[jira] [Commented] (FLINK-10672) Task stuck while writing output to flink
[
https://issues.apache.org/jira/browse/FLINK-10672?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=17305035#comment-17305035
]
Kyle Weaver commented on FLINK-10672:
-
I tried upgrading to Flink 1.12.1 and this problem no longer seems to happen
with the default memory and pipelining options.
> Task stuck while writing output to flink
>
>
> Key: FLINK-10672
> URL: https://issues.apache.org/jira/browse/FLINK-10672
> Project: Flink
> Issue Type: Bug
> Components: Runtime / Coordination
>Affects Versions: 1.5.4
> Environment: OS: Debuan rodente 4.17
> Flink version: 1.5.4
> ||Key||Value||
> |jobmanager.heap.mb|1024|
> |jobmanager.rpc.address|localhost|
> |jobmanager.rpc.port|6123|
> |metrics.reporter.jmx.class|org.apache.flink.metrics.jmx.JMXReporter|
> |metrics.reporter.jmx.port|9250-9260|
> |metrics.reporters|jmx|
> |parallelism.default|1|
> |rest.port|8081|
> |taskmanager.heap.mb|1024|
> |taskmanager.numberOfTaskSlots|1|
> |web.tmpdir|/tmp/flink-web-bdb73d6c-5b9e-47b5-9ebf-eed0a7c82c26|
>
> h1. Overview
> ||Data Port||All Slots||Free Slots||CPU Cores||Physical Memory||JVM Heap
> Size||Flink Managed Memory||
> |43501|1|0|12|62.9 GB|922 MB|642 MB|
> h1. Memory
> h2. JVM (Heap/Non-Heap)
> ||Type||Committed||Used||Maximum||
> |Heap|922 MB|575 MB|922 MB|
> |Non-Heap|68.8 MB|64.3 MB|-1 B|
> |Total|991 MB|639 MB|922 MB|
> h2. Outside JVM
> ||Type||Count||Used||Capacity||
> |Direct|3,292|105 MB|105 MB|
> |Mapped|0|0 B|0 B|
> h1. Network
> h2. Memory Segments
> ||Type||Count||
> |Available|3,194|
> |Total|3,278|
> h1. Garbage Collection
> ||Collector||Count||Time||
> |G1_Young_Generation|13|336|
> |G1_Old_Generation|1|21|
>Reporter: Ankur Goenka
>Assignee: Yun Gao
>Priority: Major
> Labels: beam
> Attachments: 0.14_all_jobs.jpg, 1uruvakHxBu.png, 3aDKQ24WvKk.png,
> Po89UGDn58V.png, WithBroadcastJob.png, jmx_dump.json, jmx_dump_detailed.json,
> jstack_129827.log, jstack_163822.log, jstack_66985.log
>
>
> I am running a fairly complex pipleline with 200+ task.
> The pipeline works fine with small data (order of 10kb input) but gets stuck
> with a slightly larger data (300kb input).
>
> The task gets stuck while writing the output toFlink, more specifically it
> gets stuck while requesting memory segment in local buffer pool. The Task
> manager UI shows that it has enough memory and memory segments to work with.
> The relevant stack trace is
> {quote}"grpc-default-executor-0" #138 daemon prio=5 os_prio=0
> tid=0x7fedb0163800 nid=0x30b7f in Object.wait() [0x7fedb4f9]
> java.lang.Thread.State: TIMED_WAITING (on object monitor)
> at (C/C++) 0x7fef201c7dae (Unknown Source)
> at (C/C++) 0x7fef1f2aea07 (Unknown Source)
> at (C/C++) 0x7fef1f241cd3 (Unknown Source)
> at java.lang.Object.wait(Native Method)
> - waiting on <0xf6d56450> (a java.util.ArrayDeque)
> at
> org.apache.flink.runtime.io.network.buffer.LocalBufferPool.requestMemorySegment(LocalBufferPool.java:247)
> - locked <0xf6d56450> (a java.util.ArrayDeque)
> at
> org.apache.flink.runtime.io.network.buffer.LocalBufferPool.requestBufferBuilderBlocking(LocalBufferPool.java:204)
> at
> org.apache.flink.runtime.io.network.api.writer.RecordWriter.requestNewBufferBuilder(RecordWriter.java:213)
> at
> org.apache.flink.runtime.io.network.api.writer.RecordWriter.sendToTarget(RecordWriter.java:144)
> at
> org.apache.flink.runtime.io.network.api.writer.RecordWriter.emit(RecordWriter.java:107)
> at
> org.apache.flink.runtime.operators.shipping.OutputCollector.collect(OutputCollector.java:65)
> at
> org.apache.flink.runtime.operators.util.metrics.CountingCollector.collect(CountingCollector.java:35)
> at
> org.apache.beam.runners.flink.translation.functions.FlinkExecutableStagePruningFunction.flatMap(FlinkExecutableStagePruningFunction.java:42)
> at
> org.apache.beam.runners.flink.translation.functions.FlinkExecutableStagePruningFunction.flatMap(FlinkExecutableStagePruningFunction.java:26)
> at
> org.apache.flink.runtime.operators.chaining.ChainedFlatMapDriver.collect(ChainedFlatMapDriver.java:80)
> at
> org.apache.flink.runtime.operators.util.metrics.CountingCollector.collect(CountingCollector.java:35)
> at
> org.apache.beam.runners.flink.translation.functions.FlinkExecutableStageFunction$MyDataReceiver.accept(FlinkExecutableStageFunction.java:230)
> - locked <0xf6a60bd0> (a java.lang.Object)
> at
> org.apache.beam.sdk.fn.data.BeamFnDataInboundObserver.accept(BeamFnDataInboundObserver.java:81)
> at
> org.apache.beam.sdk.fn.data.BeamFnDataInboundObserver.accept(BeamFnDataInboundObserver.java:32)
> at
> org.apache.beam.sdk.fn.data.BeamFnDataGrpcMultiplexer$InboundObserver.onNext(BeamFnDataGrpcMultiplexer.java:139)
> at
>
[jira] [Commented] (FLINK-10672) Task stuck while writing output to flink
[
https://issues.apache.org/jira/browse/FLINK-10672?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=17158504#comment-17158504
]
Yun Gao commented on FLINK-10672:
-
Hi [~ibzib], very sorry to disturb, do you have any updates from you side for
this issue ? Since it seems that Tfx / Beam / Flink all have evolved for some
versions, does this issue still exists?
> Task stuck while writing output to flink
>
>
> Key: FLINK-10672
> URL: https://issues.apache.org/jira/browse/FLINK-10672
> Project: Flink
> Issue Type: Bug
> Components: Runtime / Coordination
>Affects Versions: 1.5.4
> Environment: OS: Debuan rodente 4.17
> Flink version: 1.5.4
> ||Key||Value||
> |jobmanager.heap.mb|1024|
> |jobmanager.rpc.address|localhost|
> |jobmanager.rpc.port|6123|
> |metrics.reporter.jmx.class|org.apache.flink.metrics.jmx.JMXReporter|
> |metrics.reporter.jmx.port|9250-9260|
> |metrics.reporters|jmx|
> |parallelism.default|1|
> |rest.port|8081|
> |taskmanager.heap.mb|1024|
> |taskmanager.numberOfTaskSlots|1|
> |web.tmpdir|/tmp/flink-web-bdb73d6c-5b9e-47b5-9ebf-eed0a7c82c26|
>
> h1. Overview
> ||Data Port||All Slots||Free Slots||CPU Cores||Physical Memory||JVM Heap
> Size||Flink Managed Memory||
> |43501|1|0|12|62.9 GB|922 MB|642 MB|
> h1. Memory
> h2. JVM (Heap/Non-Heap)
> ||Type||Committed||Used||Maximum||
> |Heap|922 MB|575 MB|922 MB|
> |Non-Heap|68.8 MB|64.3 MB|-1 B|
> |Total|991 MB|639 MB|922 MB|
> h2. Outside JVM
> ||Type||Count||Used||Capacity||
> |Direct|3,292|105 MB|105 MB|
> |Mapped|0|0 B|0 B|
> h1. Network
> h2. Memory Segments
> ||Type||Count||
> |Available|3,194|
> |Total|3,278|
> h1. Garbage Collection
> ||Collector||Count||Time||
> |G1_Young_Generation|13|336|
> |G1_Old_Generation|1|21|
>Reporter: Ankur Goenka
>Assignee: Yun Gao
>Priority: Major
> Labels: beam
> Attachments: 0.14_all_jobs.jpg, 1uruvakHxBu.png, 3aDKQ24WvKk.png,
> Po89UGDn58V.png, WithBroadcastJob.png, jmx_dump.json, jmx_dump_detailed.json,
> jstack_129827.log, jstack_163822.log, jstack_66985.log
>
>
> I am running a fairly complex pipleline with 200+ task.
> The pipeline works fine with small data (order of 10kb input) but gets stuck
> with a slightly larger data (300kb input).
>
> The task gets stuck while writing the output toFlink, more specifically it
> gets stuck while requesting memory segment in local buffer pool. The Task
> manager UI shows that it has enough memory and memory segments to work with.
> The relevant stack trace is
> {quote}"grpc-default-executor-0" #138 daemon prio=5 os_prio=0
> tid=0x7fedb0163800 nid=0x30b7f in Object.wait() [0x7fedb4f9]
> java.lang.Thread.State: TIMED_WAITING (on object monitor)
> at (C/C++) 0x7fef201c7dae (Unknown Source)
> at (C/C++) 0x7fef1f2aea07 (Unknown Source)
> at (C/C++) 0x7fef1f241cd3 (Unknown Source)
> at java.lang.Object.wait(Native Method)
> - waiting on <0xf6d56450> (a java.util.ArrayDeque)
> at
> org.apache.flink.runtime.io.network.buffer.LocalBufferPool.requestMemorySegment(LocalBufferPool.java:247)
> - locked <0xf6d56450> (a java.util.ArrayDeque)
> at
> org.apache.flink.runtime.io.network.buffer.LocalBufferPool.requestBufferBuilderBlocking(LocalBufferPool.java:204)
> at
> org.apache.flink.runtime.io.network.api.writer.RecordWriter.requestNewBufferBuilder(RecordWriter.java:213)
> at
> org.apache.flink.runtime.io.network.api.writer.RecordWriter.sendToTarget(RecordWriter.java:144)
> at
> org.apache.flink.runtime.io.network.api.writer.RecordWriter.emit(RecordWriter.java:107)
> at
> org.apache.flink.runtime.operators.shipping.OutputCollector.collect(OutputCollector.java:65)
> at
> org.apache.flink.runtime.operators.util.metrics.CountingCollector.collect(CountingCollector.java:35)
> at
> org.apache.beam.runners.flink.translation.functions.FlinkExecutableStagePruningFunction.flatMap(FlinkExecutableStagePruningFunction.java:42)
> at
> org.apache.beam.runners.flink.translation.functions.FlinkExecutableStagePruningFunction.flatMap(FlinkExecutableStagePruningFunction.java:26)
> at
> org.apache.flink.runtime.operators.chaining.ChainedFlatMapDriver.collect(ChainedFlatMapDriver.java:80)
> at
> org.apache.flink.runtime.operators.util.metrics.CountingCollector.collect(CountingCollector.java:35)
> at
> org.apache.beam.runners.flink.translation.functions.FlinkExecutableStageFunction$MyDataReceiver.accept(FlinkExecutableStageFunction.java:230)
> - locked <0xf6a60bd0> (a java.lang.Object)
> at
> org.apache.beam.sdk.fn.data.BeamFnDataInboundObserver.accept(BeamFnDataInboundObserver.java:81)
> at
> org.apache.beam.sdk.fn.data.BeamFnDataInboundObserver.accept(BeamFnDataInboundObserver.java:32)
> at
>
[jira] [Commented] (FLINK-10672) Task stuck while writing output to flink
[
https://issues.apache.org/jira/browse/FLINK-10672?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=17148271#comment-17148271
]
Peidian Li commented on FLINK-10672:
I'm sorry for the misinformation. Although I found the same stack trace, mine
was caused by another reason. I introduced incremental cleanup in my job, which
resulted in slower job processing.
> Task stuck while writing output to flink
>
>
> Key: FLINK-10672
> URL: https://issues.apache.org/jira/browse/FLINK-10672
> Project: Flink
> Issue Type: Bug
> Components: Runtime / Coordination
>Affects Versions: 1.5.4
> Environment: OS: Debuan rodente 4.17
> Flink version: 1.5.4
> ||Key||Value||
> |jobmanager.heap.mb|1024|
> |jobmanager.rpc.address|localhost|
> |jobmanager.rpc.port|6123|
> |metrics.reporter.jmx.class|org.apache.flink.metrics.jmx.JMXReporter|
> |metrics.reporter.jmx.port|9250-9260|
> |metrics.reporters|jmx|
> |parallelism.default|1|
> |rest.port|8081|
> |taskmanager.heap.mb|1024|
> |taskmanager.numberOfTaskSlots|1|
> |web.tmpdir|/tmp/flink-web-bdb73d6c-5b9e-47b5-9ebf-eed0a7c82c26|
>
> h1. Overview
> ||Data Port||All Slots||Free Slots||CPU Cores||Physical Memory||JVM Heap
> Size||Flink Managed Memory||
> |43501|1|0|12|62.9 GB|922 MB|642 MB|
> h1. Memory
> h2. JVM (Heap/Non-Heap)
> ||Type||Committed||Used||Maximum||
> |Heap|922 MB|575 MB|922 MB|
> |Non-Heap|68.8 MB|64.3 MB|-1 B|
> |Total|991 MB|639 MB|922 MB|
> h2. Outside JVM
> ||Type||Count||Used||Capacity||
> |Direct|3,292|105 MB|105 MB|
> |Mapped|0|0 B|0 B|
> h1. Network
> h2. Memory Segments
> ||Type||Count||
> |Available|3,194|
> |Total|3,278|
> h1. Garbage Collection
> ||Collector||Count||Time||
> |G1_Young_Generation|13|336|
> |G1_Old_Generation|1|21|
>Reporter: Ankur Goenka
>Assignee: Yun Gao
>Priority: Major
> Labels: beam
> Attachments: 0.14_all_jobs.jpg, 1uruvakHxBu.png, 3aDKQ24WvKk.png,
> Po89UGDn58V.png, WithBroadcastJob.png, jmx_dump.json, jmx_dump_detailed.json,
> jstack_129827.log, jstack_163822.log, jstack_66985.log
>
>
> I am running a fairly complex pipleline with 200+ task.
> The pipeline works fine with small data (order of 10kb input) but gets stuck
> with a slightly larger data (300kb input).
>
> The task gets stuck while writing the output toFlink, more specifically it
> gets stuck while requesting memory segment in local buffer pool. The Task
> manager UI shows that it has enough memory and memory segments to work with.
> The relevant stack trace is
> {quote}"grpc-default-executor-0" #138 daemon prio=5 os_prio=0
> tid=0x7fedb0163800 nid=0x30b7f in Object.wait() [0x7fedb4f9]
> java.lang.Thread.State: TIMED_WAITING (on object monitor)
> at (C/C++) 0x7fef201c7dae (Unknown Source)
> at (C/C++) 0x7fef1f2aea07 (Unknown Source)
> at (C/C++) 0x7fef1f241cd3 (Unknown Source)
> at java.lang.Object.wait(Native Method)
> - waiting on <0xf6d56450> (a java.util.ArrayDeque)
> at
> org.apache.flink.runtime.io.network.buffer.LocalBufferPool.requestMemorySegment(LocalBufferPool.java:247)
> - locked <0xf6d56450> (a java.util.ArrayDeque)
> at
> org.apache.flink.runtime.io.network.buffer.LocalBufferPool.requestBufferBuilderBlocking(LocalBufferPool.java:204)
> at
> org.apache.flink.runtime.io.network.api.writer.RecordWriter.requestNewBufferBuilder(RecordWriter.java:213)
> at
> org.apache.flink.runtime.io.network.api.writer.RecordWriter.sendToTarget(RecordWriter.java:144)
> at
> org.apache.flink.runtime.io.network.api.writer.RecordWriter.emit(RecordWriter.java:107)
> at
> org.apache.flink.runtime.operators.shipping.OutputCollector.collect(OutputCollector.java:65)
> at
> org.apache.flink.runtime.operators.util.metrics.CountingCollector.collect(CountingCollector.java:35)
> at
> org.apache.beam.runners.flink.translation.functions.FlinkExecutableStagePruningFunction.flatMap(FlinkExecutableStagePruningFunction.java:42)
> at
> org.apache.beam.runners.flink.translation.functions.FlinkExecutableStagePruningFunction.flatMap(FlinkExecutableStagePruningFunction.java:26)
> at
> org.apache.flink.runtime.operators.chaining.ChainedFlatMapDriver.collect(ChainedFlatMapDriver.java:80)
> at
> org.apache.flink.runtime.operators.util.metrics.CountingCollector.collect(CountingCollector.java:35)
> at
> org.apache.beam.runners.flink.translation.functions.FlinkExecutableStageFunction$MyDataReceiver.accept(FlinkExecutableStageFunction.java:230)
> - locked <0xf6a60bd0> (a java.lang.Object)
> at
> org.apache.beam.sdk.fn.data.BeamFnDataInboundObserver.accept(BeamFnDataInboundObserver.java:81)
> at
> org.apache.beam.sdk.fn.data.BeamFnDataInboundObserver.accept(BeamFnDataInboundObserver.java:32)
> at
>
[jira] [Commented] (FLINK-10672) Task stuck while writing output to flink
[
https://issues.apache.org/jira/browse/FLINK-10672?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=17147486#comment-17147486
]
Yun Gao commented on FLINK-10672:
-
[~lipeidian], very thanks for the reporting, are you also using Beam on Flink
? could you also attach detailed program and exceptions ? Thanks a lot.
> Task stuck while writing output to flink
>
>
> Key: FLINK-10672
> URL: https://issues.apache.org/jira/browse/FLINK-10672
> Project: Flink
> Issue Type: Bug
> Components: Runtime / Coordination
>Affects Versions: 1.5.4
> Environment: OS: Debuan rodente 4.17
> Flink version: 1.5.4
> ||Key||Value||
> |jobmanager.heap.mb|1024|
> |jobmanager.rpc.address|localhost|
> |jobmanager.rpc.port|6123|
> |metrics.reporter.jmx.class|org.apache.flink.metrics.jmx.JMXReporter|
> |metrics.reporter.jmx.port|9250-9260|
> |metrics.reporters|jmx|
> |parallelism.default|1|
> |rest.port|8081|
> |taskmanager.heap.mb|1024|
> |taskmanager.numberOfTaskSlots|1|
> |web.tmpdir|/tmp/flink-web-bdb73d6c-5b9e-47b5-9ebf-eed0a7c82c26|
>
> h1. Overview
> ||Data Port||All Slots||Free Slots||CPU Cores||Physical Memory||JVM Heap
> Size||Flink Managed Memory||
> |43501|1|0|12|62.9 GB|922 MB|642 MB|
> h1. Memory
> h2. JVM (Heap/Non-Heap)
> ||Type||Committed||Used||Maximum||
> |Heap|922 MB|575 MB|922 MB|
> |Non-Heap|68.8 MB|64.3 MB|-1 B|
> |Total|991 MB|639 MB|922 MB|
> h2. Outside JVM
> ||Type||Count||Used||Capacity||
> |Direct|3,292|105 MB|105 MB|
> |Mapped|0|0 B|0 B|
> h1. Network
> h2. Memory Segments
> ||Type||Count||
> |Available|3,194|
> |Total|3,278|
> h1. Garbage Collection
> ||Collector||Count||Time||
> |G1_Young_Generation|13|336|
> |G1_Old_Generation|1|21|
>Reporter: Ankur Goenka
>Assignee: Yun Gao
>Priority: Major
> Labels: beam
> Attachments: 0.14_all_jobs.jpg, 1uruvakHxBu.png, 3aDKQ24WvKk.png,
> Po89UGDn58V.png, WithBroadcastJob.png, jmx_dump.json, jmx_dump_detailed.json,
> jstack_129827.log, jstack_163822.log, jstack_66985.log
>
>
> I am running a fairly complex pipleline with 200+ task.
> The pipeline works fine with small data (order of 10kb input) but gets stuck
> with a slightly larger data (300kb input).
>
> The task gets stuck while writing the output toFlink, more specifically it
> gets stuck while requesting memory segment in local buffer pool. The Task
> manager UI shows that it has enough memory and memory segments to work with.
> The relevant stack trace is
> {quote}"grpc-default-executor-0" #138 daemon prio=5 os_prio=0
> tid=0x7fedb0163800 nid=0x30b7f in Object.wait() [0x7fedb4f9]
> java.lang.Thread.State: TIMED_WAITING (on object monitor)
> at (C/C++) 0x7fef201c7dae (Unknown Source)
> at (C/C++) 0x7fef1f2aea07 (Unknown Source)
> at (C/C++) 0x7fef1f241cd3 (Unknown Source)
> at java.lang.Object.wait(Native Method)
> - waiting on <0xf6d56450> (a java.util.ArrayDeque)
> at
> org.apache.flink.runtime.io.network.buffer.LocalBufferPool.requestMemorySegment(LocalBufferPool.java:247)
> - locked <0xf6d56450> (a java.util.ArrayDeque)
> at
> org.apache.flink.runtime.io.network.buffer.LocalBufferPool.requestBufferBuilderBlocking(LocalBufferPool.java:204)
> at
> org.apache.flink.runtime.io.network.api.writer.RecordWriter.requestNewBufferBuilder(RecordWriter.java:213)
> at
> org.apache.flink.runtime.io.network.api.writer.RecordWriter.sendToTarget(RecordWriter.java:144)
> at
> org.apache.flink.runtime.io.network.api.writer.RecordWriter.emit(RecordWriter.java:107)
> at
> org.apache.flink.runtime.operators.shipping.OutputCollector.collect(OutputCollector.java:65)
> at
> org.apache.flink.runtime.operators.util.metrics.CountingCollector.collect(CountingCollector.java:35)
> at
> org.apache.beam.runners.flink.translation.functions.FlinkExecutableStagePruningFunction.flatMap(FlinkExecutableStagePruningFunction.java:42)
> at
> org.apache.beam.runners.flink.translation.functions.FlinkExecutableStagePruningFunction.flatMap(FlinkExecutableStagePruningFunction.java:26)
> at
> org.apache.flink.runtime.operators.chaining.ChainedFlatMapDriver.collect(ChainedFlatMapDriver.java:80)
> at
> org.apache.flink.runtime.operators.util.metrics.CountingCollector.collect(CountingCollector.java:35)
> at
> org.apache.beam.runners.flink.translation.functions.FlinkExecutableStageFunction$MyDataReceiver.accept(FlinkExecutableStageFunction.java:230)
> - locked <0xf6a60bd0> (a java.lang.Object)
> at
> org.apache.beam.sdk.fn.data.BeamFnDataInboundObserver.accept(BeamFnDataInboundObserver.java:81)
> at
> org.apache.beam.sdk.fn.data.BeamFnDataInboundObserver.accept(BeamFnDataInboundObserver.java:32)
> at
> org.apache.beam.sdk.fn.data.BeamFnDataGrpcMultiplexer$InboundObserver.onNext(BeamFnDataGrpcMultiplexer.java:139)
> at
>
[jira] [Commented] (FLINK-10672) Task stuck while writing output to flink
[
https://issues.apache.org/jira/browse/FLINK-10672?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=17147323#comment-17147323
]
Peidian Li commented on FLINK-10672:
I met the same problem with Flink 1.9. Is there any conclusion about this
problem?
> Task stuck while writing output to flink
>
>
> Key: FLINK-10672
> URL: https://issues.apache.org/jira/browse/FLINK-10672
> Project: Flink
> Issue Type: Bug
> Components: Runtime / Coordination
>Affects Versions: 1.5.4
> Environment: OS: Debuan rodente 4.17
> Flink version: 1.5.4
> ||Key||Value||
> |jobmanager.heap.mb|1024|
> |jobmanager.rpc.address|localhost|
> |jobmanager.rpc.port|6123|
> |metrics.reporter.jmx.class|org.apache.flink.metrics.jmx.JMXReporter|
> |metrics.reporter.jmx.port|9250-9260|
> |metrics.reporters|jmx|
> |parallelism.default|1|
> |rest.port|8081|
> |taskmanager.heap.mb|1024|
> |taskmanager.numberOfTaskSlots|1|
> |web.tmpdir|/tmp/flink-web-bdb73d6c-5b9e-47b5-9ebf-eed0a7c82c26|
>
> h1. Overview
> ||Data Port||All Slots||Free Slots||CPU Cores||Physical Memory||JVM Heap
> Size||Flink Managed Memory||
> |43501|1|0|12|62.9 GB|922 MB|642 MB|
> h1. Memory
> h2. JVM (Heap/Non-Heap)
> ||Type||Committed||Used||Maximum||
> |Heap|922 MB|575 MB|922 MB|
> |Non-Heap|68.8 MB|64.3 MB|-1 B|
> |Total|991 MB|639 MB|922 MB|
> h2. Outside JVM
> ||Type||Count||Used||Capacity||
> |Direct|3,292|105 MB|105 MB|
> |Mapped|0|0 B|0 B|
> h1. Network
> h2. Memory Segments
> ||Type||Count||
> |Available|3,194|
> |Total|3,278|
> h1. Garbage Collection
> ||Collector||Count||Time||
> |G1_Young_Generation|13|336|
> |G1_Old_Generation|1|21|
>Reporter: Ankur Goenka
>Assignee: Yun Gao
>Priority: Major
> Labels: beam
> Attachments: 0.14_all_jobs.jpg, 1uruvakHxBu.png, 3aDKQ24WvKk.png,
> Po89UGDn58V.png, WithBroadcastJob.png, jmx_dump.json, jmx_dump_detailed.json,
> jstack_129827.log, jstack_163822.log, jstack_66985.log
>
>
> I am running a fairly complex pipleline with 200+ task.
> The pipeline works fine with small data (order of 10kb input) but gets stuck
> with a slightly larger data (300kb input).
>
> The task gets stuck while writing the output toFlink, more specifically it
> gets stuck while requesting memory segment in local buffer pool. The Task
> manager UI shows that it has enough memory and memory segments to work with.
> The relevant stack trace is
> {quote}"grpc-default-executor-0" #138 daemon prio=5 os_prio=0
> tid=0x7fedb0163800 nid=0x30b7f in Object.wait() [0x7fedb4f9]
> java.lang.Thread.State: TIMED_WAITING (on object monitor)
> at (C/C++) 0x7fef201c7dae (Unknown Source)
> at (C/C++) 0x7fef1f2aea07 (Unknown Source)
> at (C/C++) 0x7fef1f241cd3 (Unknown Source)
> at java.lang.Object.wait(Native Method)
> - waiting on <0xf6d56450> (a java.util.ArrayDeque)
> at
> org.apache.flink.runtime.io.network.buffer.LocalBufferPool.requestMemorySegment(LocalBufferPool.java:247)
> - locked <0xf6d56450> (a java.util.ArrayDeque)
> at
> org.apache.flink.runtime.io.network.buffer.LocalBufferPool.requestBufferBuilderBlocking(LocalBufferPool.java:204)
> at
> org.apache.flink.runtime.io.network.api.writer.RecordWriter.requestNewBufferBuilder(RecordWriter.java:213)
> at
> org.apache.flink.runtime.io.network.api.writer.RecordWriter.sendToTarget(RecordWriter.java:144)
> at
> org.apache.flink.runtime.io.network.api.writer.RecordWriter.emit(RecordWriter.java:107)
> at
> org.apache.flink.runtime.operators.shipping.OutputCollector.collect(OutputCollector.java:65)
> at
> org.apache.flink.runtime.operators.util.metrics.CountingCollector.collect(CountingCollector.java:35)
> at
> org.apache.beam.runners.flink.translation.functions.FlinkExecutableStagePruningFunction.flatMap(FlinkExecutableStagePruningFunction.java:42)
> at
> org.apache.beam.runners.flink.translation.functions.FlinkExecutableStagePruningFunction.flatMap(FlinkExecutableStagePruningFunction.java:26)
> at
> org.apache.flink.runtime.operators.chaining.ChainedFlatMapDriver.collect(ChainedFlatMapDriver.java:80)
> at
> org.apache.flink.runtime.operators.util.metrics.CountingCollector.collect(CountingCollector.java:35)
> at
> org.apache.beam.runners.flink.translation.functions.FlinkExecutableStageFunction$MyDataReceiver.accept(FlinkExecutableStageFunction.java:230)
> - locked <0xf6a60bd0> (a java.lang.Object)
> at
> org.apache.beam.sdk.fn.data.BeamFnDataInboundObserver.accept(BeamFnDataInboundObserver.java:81)
> at
> org.apache.beam.sdk.fn.data.BeamFnDataInboundObserver.accept(BeamFnDataInboundObserver.java:32)
> at
> org.apache.beam.sdk.fn.data.BeamFnDataGrpcMultiplexer$InboundObserver.onNext(BeamFnDataGrpcMultiplexer.java:139)
> at
>
[jira] [Commented] (FLINK-10672) Task stuck while writing output to flink
[
https://issues.apache.org/jira/browse/FLINK-10672?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=17101381#comment-17101381
]
Yun Gao commented on FLINK-10672:
-
Hi [~ibzib], sorry that the environment has been freed by the administrator for
met with some security problem, I will re-deploy the environment and will send
you the access method this week.
> Task stuck while writing output to flink
>
>
> Key: FLINK-10672
> URL: https://issues.apache.org/jira/browse/FLINK-10672
> Project: Flink
> Issue Type: Bug
> Components: Runtime / Coordination
>Affects Versions: 1.5.4
> Environment: OS: Debuan rodente 4.17
> Flink version: 1.5.4
> ||Key||Value||
> |jobmanager.heap.mb|1024|
> |jobmanager.rpc.address|localhost|
> |jobmanager.rpc.port|6123|
> |metrics.reporter.jmx.class|org.apache.flink.metrics.jmx.JMXReporter|
> |metrics.reporter.jmx.port|9250-9260|
> |metrics.reporters|jmx|
> |parallelism.default|1|
> |rest.port|8081|
> |taskmanager.heap.mb|1024|
> |taskmanager.numberOfTaskSlots|1|
> |web.tmpdir|/tmp/flink-web-bdb73d6c-5b9e-47b5-9ebf-eed0a7c82c26|
>
> h1. Overview
> ||Data Port||All Slots||Free Slots||CPU Cores||Physical Memory||JVM Heap
> Size||Flink Managed Memory||
> |43501|1|0|12|62.9 GB|922 MB|642 MB|
> h1. Memory
> h2. JVM (Heap/Non-Heap)
> ||Type||Committed||Used||Maximum||
> |Heap|922 MB|575 MB|922 MB|
> |Non-Heap|68.8 MB|64.3 MB|-1 B|
> |Total|991 MB|639 MB|922 MB|
> h2. Outside JVM
> ||Type||Count||Used||Capacity||
> |Direct|3,292|105 MB|105 MB|
> |Mapped|0|0 B|0 B|
> h1. Network
> h2. Memory Segments
> ||Type||Count||
> |Available|3,194|
> |Total|3,278|
> h1. Garbage Collection
> ||Collector||Count||Time||
> |G1_Young_Generation|13|336|
> |G1_Old_Generation|1|21|
>Reporter: Ankur Goenka
>Assignee: Yun Gao
>Priority: Major
> Labels: beam
> Attachments: 0.14_all_jobs.jpg, 1uruvakHxBu.png, 3aDKQ24WvKk.png,
> Po89UGDn58V.png, WithBroadcastJob.png, jmx_dump.json, jmx_dump_detailed.json,
> jstack_129827.log, jstack_163822.log, jstack_66985.log
>
>
> I am running a fairly complex pipleline with 200+ task.
> The pipeline works fine with small data (order of 10kb input) but gets stuck
> with a slightly larger data (300kb input).
>
> The task gets stuck while writing the output toFlink, more specifically it
> gets stuck while requesting memory segment in local buffer pool. The Task
> manager UI shows that it has enough memory and memory segments to work with.
> The relevant stack trace is
> {quote}"grpc-default-executor-0" #138 daemon prio=5 os_prio=0
> tid=0x7fedb0163800 nid=0x30b7f in Object.wait() [0x7fedb4f9]
> java.lang.Thread.State: TIMED_WAITING (on object monitor)
> at (C/C++) 0x7fef201c7dae (Unknown Source)
> at (C/C++) 0x7fef1f2aea07 (Unknown Source)
> at (C/C++) 0x7fef1f241cd3 (Unknown Source)
> at java.lang.Object.wait(Native Method)
> - waiting on <0xf6d56450> (a java.util.ArrayDeque)
> at
> org.apache.flink.runtime.io.network.buffer.LocalBufferPool.requestMemorySegment(LocalBufferPool.java:247)
> - locked <0xf6d56450> (a java.util.ArrayDeque)
> at
> org.apache.flink.runtime.io.network.buffer.LocalBufferPool.requestBufferBuilderBlocking(LocalBufferPool.java:204)
> at
> org.apache.flink.runtime.io.network.api.writer.RecordWriter.requestNewBufferBuilder(RecordWriter.java:213)
> at
> org.apache.flink.runtime.io.network.api.writer.RecordWriter.sendToTarget(RecordWriter.java:144)
> at
> org.apache.flink.runtime.io.network.api.writer.RecordWriter.emit(RecordWriter.java:107)
> at
> org.apache.flink.runtime.operators.shipping.OutputCollector.collect(OutputCollector.java:65)
> at
> org.apache.flink.runtime.operators.util.metrics.CountingCollector.collect(CountingCollector.java:35)
> at
> org.apache.beam.runners.flink.translation.functions.FlinkExecutableStagePruningFunction.flatMap(FlinkExecutableStagePruningFunction.java:42)
> at
> org.apache.beam.runners.flink.translation.functions.FlinkExecutableStagePruningFunction.flatMap(FlinkExecutableStagePruningFunction.java:26)
> at
> org.apache.flink.runtime.operators.chaining.ChainedFlatMapDriver.collect(ChainedFlatMapDriver.java:80)
> at
> org.apache.flink.runtime.operators.util.metrics.CountingCollector.collect(CountingCollector.java:35)
> at
> org.apache.beam.runners.flink.translation.functions.FlinkExecutableStageFunction$MyDataReceiver.accept(FlinkExecutableStageFunction.java:230)
> - locked <0xf6a60bd0> (a java.lang.Object)
> at
> org.apache.beam.sdk.fn.data.BeamFnDataInboundObserver.accept(BeamFnDataInboundObserver.java:81)
> at
> org.apache.beam.sdk.fn.data.BeamFnDataInboundObserver.accept(BeamFnDataInboundObserver.java:32)
> at
>
[jira] [Commented] (FLINK-10672) Task stuck while writing output to flink
[
https://issues.apache.org/jira/browse/FLINK-10672?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=17097696#comment-17097696
]
Kyle Weaver commented on FLINK-10672:
-
Hi Yun, sorry I haven't had much time to look at this lately. If you still have
the test environment available, you can send info to ib...@apache.org.
> Task stuck while writing output to flink
>
>
> Key: FLINK-10672
> URL: https://issues.apache.org/jira/browse/FLINK-10672
> Project: Flink
> Issue Type: Bug
> Components: Runtime / Coordination
>Affects Versions: 1.5.4
> Environment: OS: Debuan rodente 4.17
> Flink version: 1.5.4
> ||Key||Value||
> |jobmanager.heap.mb|1024|
> |jobmanager.rpc.address|localhost|
> |jobmanager.rpc.port|6123|
> |metrics.reporter.jmx.class|org.apache.flink.metrics.jmx.JMXReporter|
> |metrics.reporter.jmx.port|9250-9260|
> |metrics.reporters|jmx|
> |parallelism.default|1|
> |rest.port|8081|
> |taskmanager.heap.mb|1024|
> |taskmanager.numberOfTaskSlots|1|
> |web.tmpdir|/tmp/flink-web-bdb73d6c-5b9e-47b5-9ebf-eed0a7c82c26|
>
> h1. Overview
> ||Data Port||All Slots||Free Slots||CPU Cores||Physical Memory||JVM Heap
> Size||Flink Managed Memory||
> |43501|1|0|12|62.9 GB|922 MB|642 MB|
> h1. Memory
> h2. JVM (Heap/Non-Heap)
> ||Type||Committed||Used||Maximum||
> |Heap|922 MB|575 MB|922 MB|
> |Non-Heap|68.8 MB|64.3 MB|-1 B|
> |Total|991 MB|639 MB|922 MB|
> h2. Outside JVM
> ||Type||Count||Used||Capacity||
> |Direct|3,292|105 MB|105 MB|
> |Mapped|0|0 B|0 B|
> h1. Network
> h2. Memory Segments
> ||Type||Count||
> |Available|3,194|
> |Total|3,278|
> h1. Garbage Collection
> ||Collector||Count||Time||
> |G1_Young_Generation|13|336|
> |G1_Old_Generation|1|21|
>Reporter: Ankur Goenka
>Assignee: Yun Gao
>Priority: Major
> Labels: beam
> Attachments: 0.14_all_jobs.jpg, 1uruvakHxBu.png, 3aDKQ24WvKk.png,
> Po89UGDn58V.png, WithBroadcastJob.png, jmx_dump.json, jmx_dump_detailed.json,
> jstack_129827.log, jstack_163822.log, jstack_66985.log
>
>
> I am running a fairly complex pipleline with 200+ task.
> The pipeline works fine with small data (order of 10kb input) but gets stuck
> with a slightly larger data (300kb input).
>
> The task gets stuck while writing the output toFlink, more specifically it
> gets stuck while requesting memory segment in local buffer pool. The Task
> manager UI shows that it has enough memory and memory segments to work with.
> The relevant stack trace is
> {quote}"grpc-default-executor-0" #138 daemon prio=5 os_prio=0
> tid=0x7fedb0163800 nid=0x30b7f in Object.wait() [0x7fedb4f9]
> java.lang.Thread.State: TIMED_WAITING (on object monitor)
> at (C/C++) 0x7fef201c7dae (Unknown Source)
> at (C/C++) 0x7fef1f2aea07 (Unknown Source)
> at (C/C++) 0x7fef1f241cd3 (Unknown Source)
> at java.lang.Object.wait(Native Method)
> - waiting on <0xf6d56450> (a java.util.ArrayDeque)
> at
> org.apache.flink.runtime.io.network.buffer.LocalBufferPool.requestMemorySegment(LocalBufferPool.java:247)
> - locked <0xf6d56450> (a java.util.ArrayDeque)
> at
> org.apache.flink.runtime.io.network.buffer.LocalBufferPool.requestBufferBuilderBlocking(LocalBufferPool.java:204)
> at
> org.apache.flink.runtime.io.network.api.writer.RecordWriter.requestNewBufferBuilder(RecordWriter.java:213)
> at
> org.apache.flink.runtime.io.network.api.writer.RecordWriter.sendToTarget(RecordWriter.java:144)
> at
> org.apache.flink.runtime.io.network.api.writer.RecordWriter.emit(RecordWriter.java:107)
> at
> org.apache.flink.runtime.operators.shipping.OutputCollector.collect(OutputCollector.java:65)
> at
> org.apache.flink.runtime.operators.util.metrics.CountingCollector.collect(CountingCollector.java:35)
> at
> org.apache.beam.runners.flink.translation.functions.FlinkExecutableStagePruningFunction.flatMap(FlinkExecutableStagePruningFunction.java:42)
> at
> org.apache.beam.runners.flink.translation.functions.FlinkExecutableStagePruningFunction.flatMap(FlinkExecutableStagePruningFunction.java:26)
> at
> org.apache.flink.runtime.operators.chaining.ChainedFlatMapDriver.collect(ChainedFlatMapDriver.java:80)
> at
> org.apache.flink.runtime.operators.util.metrics.CountingCollector.collect(CountingCollector.java:35)
> at
> org.apache.beam.runners.flink.translation.functions.FlinkExecutableStageFunction$MyDataReceiver.accept(FlinkExecutableStageFunction.java:230)
> - locked <0xf6a60bd0> (a java.lang.Object)
> at
> org.apache.beam.sdk.fn.data.BeamFnDataInboundObserver.accept(BeamFnDataInboundObserver.java:81)
> at
> org.apache.beam.sdk.fn.data.BeamFnDataInboundObserver.accept(BeamFnDataInboundObserver.java:32)
> at
> org.apache.beam.sdk.fn.data.BeamFnDataGrpcMultiplexer$InboundObserver.onNext(BeamFnDataGrpcMultiplexer.java:139)
> at
[jira] [Commented] (FLINK-10672) Task stuck while writing output to flink
[
https://issues.apache.org/jira/browse/FLINK-10672?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=17070745#comment-17070745
]
Yun Gao commented on FLINK-10672:
-
Hi [~ibzib] I have deploy an environment with Beam 2.14 and tfx 0.14 on a
public accessible machine. Could you provide me an email and I will send you
the IP and account ?
> Task stuck while writing output to flink
>
>
> Key: FLINK-10672
> URL: https://issues.apache.org/jira/browse/FLINK-10672
> Project: Flink
> Issue Type: Bug
> Components: Runtime / Coordination
>Affects Versions: 1.5.4
> Environment: OS: Debuan rodente 4.17
> Flink version: 1.5.4
> ||Key||Value||
> |jobmanager.heap.mb|1024|
> |jobmanager.rpc.address|localhost|
> |jobmanager.rpc.port|6123|
> |metrics.reporter.jmx.class|org.apache.flink.metrics.jmx.JMXReporter|
> |metrics.reporter.jmx.port|9250-9260|
> |metrics.reporters|jmx|
> |parallelism.default|1|
> |rest.port|8081|
> |taskmanager.heap.mb|1024|
> |taskmanager.numberOfTaskSlots|1|
> |web.tmpdir|/tmp/flink-web-bdb73d6c-5b9e-47b5-9ebf-eed0a7c82c26|
>
> h1. Overview
> ||Data Port||All Slots||Free Slots||CPU Cores||Physical Memory||JVM Heap
> Size||Flink Managed Memory||
> |43501|1|0|12|62.9 GB|922 MB|642 MB|
> h1. Memory
> h2. JVM (Heap/Non-Heap)
> ||Type||Committed||Used||Maximum||
> |Heap|922 MB|575 MB|922 MB|
> |Non-Heap|68.8 MB|64.3 MB|-1 B|
> |Total|991 MB|639 MB|922 MB|
> h2. Outside JVM
> ||Type||Count||Used||Capacity||
> |Direct|3,292|105 MB|105 MB|
> |Mapped|0|0 B|0 B|
> h1. Network
> h2. Memory Segments
> ||Type||Count||
> |Available|3,194|
> |Total|3,278|
> h1. Garbage Collection
> ||Collector||Count||Time||
> |G1_Young_Generation|13|336|
> |G1_Old_Generation|1|21|
>Reporter: Ankur Goenka
>Assignee: Yun Gao
>Priority: Major
> Labels: beam
> Attachments: 0.14_all_jobs.jpg, 1uruvakHxBu.png, 3aDKQ24WvKk.png,
> Po89UGDn58V.png, WithBroadcastJob.png, jmx_dump.json, jmx_dump_detailed.json,
> jstack_129827.log, jstack_163822.log, jstack_66985.log
>
>
> I am running a fairly complex pipleline with 200+ task.
> The pipeline works fine with small data (order of 10kb input) but gets stuck
> with a slightly larger data (300kb input).
>
> The task gets stuck while writing the output toFlink, more specifically it
> gets stuck while requesting memory segment in local buffer pool. The Task
> manager UI shows that it has enough memory and memory segments to work with.
> The relevant stack trace is
> {quote}"grpc-default-executor-0" #138 daemon prio=5 os_prio=0
> tid=0x7fedb0163800 nid=0x30b7f in Object.wait() [0x7fedb4f9]
> java.lang.Thread.State: TIMED_WAITING (on object monitor)
> at (C/C++) 0x7fef201c7dae (Unknown Source)
> at (C/C++) 0x7fef1f2aea07 (Unknown Source)
> at (C/C++) 0x7fef1f241cd3 (Unknown Source)
> at java.lang.Object.wait(Native Method)
> - waiting on <0xf6d56450> (a java.util.ArrayDeque)
> at
> org.apache.flink.runtime.io.network.buffer.LocalBufferPool.requestMemorySegment(LocalBufferPool.java:247)
> - locked <0xf6d56450> (a java.util.ArrayDeque)
> at
> org.apache.flink.runtime.io.network.buffer.LocalBufferPool.requestBufferBuilderBlocking(LocalBufferPool.java:204)
> at
> org.apache.flink.runtime.io.network.api.writer.RecordWriter.requestNewBufferBuilder(RecordWriter.java:213)
> at
> org.apache.flink.runtime.io.network.api.writer.RecordWriter.sendToTarget(RecordWriter.java:144)
> at
> org.apache.flink.runtime.io.network.api.writer.RecordWriter.emit(RecordWriter.java:107)
> at
> org.apache.flink.runtime.operators.shipping.OutputCollector.collect(OutputCollector.java:65)
> at
> org.apache.flink.runtime.operators.util.metrics.CountingCollector.collect(CountingCollector.java:35)
> at
> org.apache.beam.runners.flink.translation.functions.FlinkExecutableStagePruningFunction.flatMap(FlinkExecutableStagePruningFunction.java:42)
> at
> org.apache.beam.runners.flink.translation.functions.FlinkExecutableStagePruningFunction.flatMap(FlinkExecutableStagePruningFunction.java:26)
> at
> org.apache.flink.runtime.operators.chaining.ChainedFlatMapDriver.collect(ChainedFlatMapDriver.java:80)
> at
> org.apache.flink.runtime.operators.util.metrics.CountingCollector.collect(CountingCollector.java:35)
> at
> org.apache.beam.runners.flink.translation.functions.FlinkExecutableStageFunction$MyDataReceiver.accept(FlinkExecutableStageFunction.java:230)
> - locked <0xf6a60bd0> (a java.lang.Object)
> at
> org.apache.beam.sdk.fn.data.BeamFnDataInboundObserver.accept(BeamFnDataInboundObserver.java:81)
> at
> org.apache.beam.sdk.fn.data.BeamFnDataInboundObserver.accept(BeamFnDataInboundObserver.java:32)
> at
>
[jira] [Commented] (FLINK-10672) Task stuck while writing output to flink
[
https://issues.apache.org/jira/browse/FLINK-10672?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=17065410#comment-17065410
]
Yun Gao commented on FLINK-10672:
-
Hi Kyle, very sorry that I still have some problem reproducing the issue.
> Task stuck while writing output to flink
>
>
> Key: FLINK-10672
> URL: https://issues.apache.org/jira/browse/FLINK-10672
> Project: Flink
> Issue Type: Bug
> Components: Runtime / Coordination
>Affects Versions: 1.5.4
> Environment: OS: Debuan rodente 4.17
> Flink version: 1.5.4
> ||Key||Value||
> |jobmanager.heap.mb|1024|
> |jobmanager.rpc.address|localhost|
> |jobmanager.rpc.port|6123|
> |metrics.reporter.jmx.class|org.apache.flink.metrics.jmx.JMXReporter|
> |metrics.reporter.jmx.port|9250-9260|
> |metrics.reporters|jmx|
> |parallelism.default|1|
> |rest.port|8081|
> |taskmanager.heap.mb|1024|
> |taskmanager.numberOfTaskSlots|1|
> |web.tmpdir|/tmp/flink-web-bdb73d6c-5b9e-47b5-9ebf-eed0a7c82c26|
>
> h1. Overview
> ||Data Port||All Slots||Free Slots||CPU Cores||Physical Memory||JVM Heap
> Size||Flink Managed Memory||
> |43501|1|0|12|62.9 GB|922 MB|642 MB|
> h1. Memory
> h2. JVM (Heap/Non-Heap)
> ||Type||Committed||Used||Maximum||
> |Heap|922 MB|575 MB|922 MB|
> |Non-Heap|68.8 MB|64.3 MB|-1 B|
> |Total|991 MB|639 MB|922 MB|
> h2. Outside JVM
> ||Type||Count||Used||Capacity||
> |Direct|3,292|105 MB|105 MB|
> |Mapped|0|0 B|0 B|
> h1. Network
> h2. Memory Segments
> ||Type||Count||
> |Available|3,194|
> |Total|3,278|
> h1. Garbage Collection
> ||Collector||Count||Time||
> |G1_Young_Generation|13|336|
> |G1_Old_Generation|1|21|
>Reporter: Ankur Goenka
>Assignee: Yun Gao
>Priority: Major
> Labels: beam
> Attachments: 0.14_all_jobs.jpg, 1uruvakHxBu.png, 3aDKQ24WvKk.png,
> Po89UGDn58V.png, WithBroadcastJob.png, jmx_dump.json, jmx_dump_detailed.json,
> jstack_129827.log, jstack_163822.log, jstack_66985.log
>
>
> I am running a fairly complex pipleline with 200+ task.
> The pipeline works fine with small data (order of 10kb input) but gets stuck
> with a slightly larger data (300kb input).
>
> The task gets stuck while writing the output toFlink, more specifically it
> gets stuck while requesting memory segment in local buffer pool. The Task
> manager UI shows that it has enough memory and memory segments to work with.
> The relevant stack trace is
> {quote}"grpc-default-executor-0" #138 daemon prio=5 os_prio=0
> tid=0x7fedb0163800 nid=0x30b7f in Object.wait() [0x7fedb4f9]
> java.lang.Thread.State: TIMED_WAITING (on object monitor)
> at (C/C++) 0x7fef201c7dae (Unknown Source)
> at (C/C++) 0x7fef1f2aea07 (Unknown Source)
> at (C/C++) 0x7fef1f241cd3 (Unknown Source)
> at java.lang.Object.wait(Native Method)
> - waiting on <0xf6d56450> (a java.util.ArrayDeque)
> at
> org.apache.flink.runtime.io.network.buffer.LocalBufferPool.requestMemorySegment(LocalBufferPool.java:247)
> - locked <0xf6d56450> (a java.util.ArrayDeque)
> at
> org.apache.flink.runtime.io.network.buffer.LocalBufferPool.requestBufferBuilderBlocking(LocalBufferPool.java:204)
> at
> org.apache.flink.runtime.io.network.api.writer.RecordWriter.requestNewBufferBuilder(RecordWriter.java:213)
> at
> org.apache.flink.runtime.io.network.api.writer.RecordWriter.sendToTarget(RecordWriter.java:144)
> at
> org.apache.flink.runtime.io.network.api.writer.RecordWriter.emit(RecordWriter.java:107)
> at
> org.apache.flink.runtime.operators.shipping.OutputCollector.collect(OutputCollector.java:65)
> at
> org.apache.flink.runtime.operators.util.metrics.CountingCollector.collect(CountingCollector.java:35)
> at
> org.apache.beam.runners.flink.translation.functions.FlinkExecutableStagePruningFunction.flatMap(FlinkExecutableStagePruningFunction.java:42)
> at
> org.apache.beam.runners.flink.translation.functions.FlinkExecutableStagePruningFunction.flatMap(FlinkExecutableStagePruningFunction.java:26)
> at
> org.apache.flink.runtime.operators.chaining.ChainedFlatMapDriver.collect(ChainedFlatMapDriver.java:80)
> at
> org.apache.flink.runtime.operators.util.metrics.CountingCollector.collect(CountingCollector.java:35)
> at
> org.apache.beam.runners.flink.translation.functions.FlinkExecutableStageFunction$MyDataReceiver.accept(FlinkExecutableStageFunction.java:230)
> - locked <0xf6a60bd0> (a java.lang.Object)
> at
> org.apache.beam.sdk.fn.data.BeamFnDataInboundObserver.accept(BeamFnDataInboundObserver.java:81)
> at
> org.apache.beam.sdk.fn.data.BeamFnDataInboundObserver.accept(BeamFnDataInboundObserver.java:32)
> at
> org.apache.beam.sdk.fn.data.BeamFnDataGrpcMultiplexer$InboundObserver.onNext(BeamFnDataGrpcMultiplexer.java:139)
> at
>
[jira] [Commented] (FLINK-10672) Task stuck while writing output to flink
[
https://issues.apache.org/jira/browse/FLINK-10672?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=17065414#comment-17065414
]
Yun Gao commented on FLINK-10672:
-
Hi [~ibzib], very sorry that I still have some problems when reproducing the
issue. I will try to find an public accessible machine and try to reproduce the
issue on that machine so that we could all access the environment. I will
concentrate on this issue and get the environment prepared in the next few
days.
> Task stuck while writing output to flink
>
>
> Key: FLINK-10672
> URL: https://issues.apache.org/jira/browse/FLINK-10672
> Project: Flink
> Issue Type: Bug
> Components: Runtime / Coordination
>Affects Versions: 1.5.4
> Environment: OS: Debuan rodente 4.17
> Flink version: 1.5.4
> ||Key||Value||
> |jobmanager.heap.mb|1024|
> |jobmanager.rpc.address|localhost|
> |jobmanager.rpc.port|6123|
> |metrics.reporter.jmx.class|org.apache.flink.metrics.jmx.JMXReporter|
> |metrics.reporter.jmx.port|9250-9260|
> |metrics.reporters|jmx|
> |parallelism.default|1|
> |rest.port|8081|
> |taskmanager.heap.mb|1024|
> |taskmanager.numberOfTaskSlots|1|
> |web.tmpdir|/tmp/flink-web-bdb73d6c-5b9e-47b5-9ebf-eed0a7c82c26|
>
> h1. Overview
> ||Data Port||All Slots||Free Slots||CPU Cores||Physical Memory||JVM Heap
> Size||Flink Managed Memory||
> |43501|1|0|12|62.9 GB|922 MB|642 MB|
> h1. Memory
> h2. JVM (Heap/Non-Heap)
> ||Type||Committed||Used||Maximum||
> |Heap|922 MB|575 MB|922 MB|
> |Non-Heap|68.8 MB|64.3 MB|-1 B|
> |Total|991 MB|639 MB|922 MB|
> h2. Outside JVM
> ||Type||Count||Used||Capacity||
> |Direct|3,292|105 MB|105 MB|
> |Mapped|0|0 B|0 B|
> h1. Network
> h2. Memory Segments
> ||Type||Count||
> |Available|3,194|
> |Total|3,278|
> h1. Garbage Collection
> ||Collector||Count||Time||
> |G1_Young_Generation|13|336|
> |G1_Old_Generation|1|21|
>Reporter: Ankur Goenka
>Assignee: Yun Gao
>Priority: Major
> Labels: beam
> Attachments: 0.14_all_jobs.jpg, 1uruvakHxBu.png, 3aDKQ24WvKk.png,
> Po89UGDn58V.png, WithBroadcastJob.png, jmx_dump.json, jmx_dump_detailed.json,
> jstack_129827.log, jstack_163822.log, jstack_66985.log
>
>
> I am running a fairly complex pipleline with 200+ task.
> The pipeline works fine with small data (order of 10kb input) but gets stuck
> with a slightly larger data (300kb input).
>
> The task gets stuck while writing the output toFlink, more specifically it
> gets stuck while requesting memory segment in local buffer pool. The Task
> manager UI shows that it has enough memory and memory segments to work with.
> The relevant stack trace is
> {quote}"grpc-default-executor-0" #138 daemon prio=5 os_prio=0
> tid=0x7fedb0163800 nid=0x30b7f in Object.wait() [0x7fedb4f9]
> java.lang.Thread.State: TIMED_WAITING (on object monitor)
> at (C/C++) 0x7fef201c7dae (Unknown Source)
> at (C/C++) 0x7fef1f2aea07 (Unknown Source)
> at (C/C++) 0x7fef1f241cd3 (Unknown Source)
> at java.lang.Object.wait(Native Method)
> - waiting on <0xf6d56450> (a java.util.ArrayDeque)
> at
> org.apache.flink.runtime.io.network.buffer.LocalBufferPool.requestMemorySegment(LocalBufferPool.java:247)
> - locked <0xf6d56450> (a java.util.ArrayDeque)
> at
> org.apache.flink.runtime.io.network.buffer.LocalBufferPool.requestBufferBuilderBlocking(LocalBufferPool.java:204)
> at
> org.apache.flink.runtime.io.network.api.writer.RecordWriter.requestNewBufferBuilder(RecordWriter.java:213)
> at
> org.apache.flink.runtime.io.network.api.writer.RecordWriter.sendToTarget(RecordWriter.java:144)
> at
> org.apache.flink.runtime.io.network.api.writer.RecordWriter.emit(RecordWriter.java:107)
> at
> org.apache.flink.runtime.operators.shipping.OutputCollector.collect(OutputCollector.java:65)
> at
> org.apache.flink.runtime.operators.util.metrics.CountingCollector.collect(CountingCollector.java:35)
> at
> org.apache.beam.runners.flink.translation.functions.FlinkExecutableStagePruningFunction.flatMap(FlinkExecutableStagePruningFunction.java:42)
> at
> org.apache.beam.runners.flink.translation.functions.FlinkExecutableStagePruningFunction.flatMap(FlinkExecutableStagePruningFunction.java:26)
> at
> org.apache.flink.runtime.operators.chaining.ChainedFlatMapDriver.collect(ChainedFlatMapDriver.java:80)
> at
> org.apache.flink.runtime.operators.util.metrics.CountingCollector.collect(CountingCollector.java:35)
> at
> org.apache.beam.runners.flink.translation.functions.FlinkExecutableStageFunction$MyDataReceiver.accept(FlinkExecutableStageFunction.java:230)
> - locked <0xf6a60bd0> (a java.lang.Object)
> at
> org.apache.beam.sdk.fn.data.BeamFnDataInboundObserver.accept(BeamFnDataInboundObserver.java:81)
> at
>
[jira] [Commented] (FLINK-10672) Task stuck while writing output to flink
[
https://issues.apache.org/jira/browse/FLINK-10672?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=17065155#comment-17065155
]
Kyle Weaver commented on FLINK-10672:
-
Hi [~gaoyunhaii], any updates on this?
> Task stuck while writing output to flink
>
>
> Key: FLINK-10672
> URL: https://issues.apache.org/jira/browse/FLINK-10672
> Project: Flink
> Issue Type: Bug
> Components: Runtime / Coordination
>Affects Versions: 1.5.4
> Environment: OS: Debuan rodente 4.17
> Flink version: 1.5.4
> ||Key||Value||
> |jobmanager.heap.mb|1024|
> |jobmanager.rpc.address|localhost|
> |jobmanager.rpc.port|6123|
> |metrics.reporter.jmx.class|org.apache.flink.metrics.jmx.JMXReporter|
> |metrics.reporter.jmx.port|9250-9260|
> |metrics.reporters|jmx|
> |parallelism.default|1|
> |rest.port|8081|
> |taskmanager.heap.mb|1024|
> |taskmanager.numberOfTaskSlots|1|
> |web.tmpdir|/tmp/flink-web-bdb73d6c-5b9e-47b5-9ebf-eed0a7c82c26|
>
> h1. Overview
> ||Data Port||All Slots||Free Slots||CPU Cores||Physical Memory||JVM Heap
> Size||Flink Managed Memory||
> |43501|1|0|12|62.9 GB|922 MB|642 MB|
> h1. Memory
> h2. JVM (Heap/Non-Heap)
> ||Type||Committed||Used||Maximum||
> |Heap|922 MB|575 MB|922 MB|
> |Non-Heap|68.8 MB|64.3 MB|-1 B|
> |Total|991 MB|639 MB|922 MB|
> h2. Outside JVM
> ||Type||Count||Used||Capacity||
> |Direct|3,292|105 MB|105 MB|
> |Mapped|0|0 B|0 B|
> h1. Network
> h2. Memory Segments
> ||Type||Count||
> |Available|3,194|
> |Total|3,278|
> h1. Garbage Collection
> ||Collector||Count||Time||
> |G1_Young_Generation|13|336|
> |G1_Old_Generation|1|21|
>Reporter: Ankur Goenka
>Assignee: Yun Gao
>Priority: Major
> Labels: beam
> Attachments: 0.14_all_jobs.jpg, 1uruvakHxBu.png, 3aDKQ24WvKk.png,
> Po89UGDn58V.png, WithBroadcastJob.png, jmx_dump.json, jmx_dump_detailed.json,
> jstack_129827.log, jstack_163822.log, jstack_66985.log
>
>
> I am running a fairly complex pipleline with 200+ task.
> The pipeline works fine with small data (order of 10kb input) but gets stuck
> with a slightly larger data (300kb input).
>
> The task gets stuck while writing the output toFlink, more specifically it
> gets stuck while requesting memory segment in local buffer pool. The Task
> manager UI shows that it has enough memory and memory segments to work with.
> The relevant stack trace is
> {quote}"grpc-default-executor-0" #138 daemon prio=5 os_prio=0
> tid=0x7fedb0163800 nid=0x30b7f in Object.wait() [0x7fedb4f9]
> java.lang.Thread.State: TIMED_WAITING (on object monitor)
> at (C/C++) 0x7fef201c7dae (Unknown Source)
> at (C/C++) 0x7fef1f2aea07 (Unknown Source)
> at (C/C++) 0x7fef1f241cd3 (Unknown Source)
> at java.lang.Object.wait(Native Method)
> - waiting on <0xf6d56450> (a java.util.ArrayDeque)
> at
> org.apache.flink.runtime.io.network.buffer.LocalBufferPool.requestMemorySegment(LocalBufferPool.java:247)
> - locked <0xf6d56450> (a java.util.ArrayDeque)
> at
> org.apache.flink.runtime.io.network.buffer.LocalBufferPool.requestBufferBuilderBlocking(LocalBufferPool.java:204)
> at
> org.apache.flink.runtime.io.network.api.writer.RecordWriter.requestNewBufferBuilder(RecordWriter.java:213)
> at
> org.apache.flink.runtime.io.network.api.writer.RecordWriter.sendToTarget(RecordWriter.java:144)
> at
> org.apache.flink.runtime.io.network.api.writer.RecordWriter.emit(RecordWriter.java:107)
> at
> org.apache.flink.runtime.operators.shipping.OutputCollector.collect(OutputCollector.java:65)
> at
> org.apache.flink.runtime.operators.util.metrics.CountingCollector.collect(CountingCollector.java:35)
> at
> org.apache.beam.runners.flink.translation.functions.FlinkExecutableStagePruningFunction.flatMap(FlinkExecutableStagePruningFunction.java:42)
> at
> org.apache.beam.runners.flink.translation.functions.FlinkExecutableStagePruningFunction.flatMap(FlinkExecutableStagePruningFunction.java:26)
> at
> org.apache.flink.runtime.operators.chaining.ChainedFlatMapDriver.collect(ChainedFlatMapDriver.java:80)
> at
> org.apache.flink.runtime.operators.util.metrics.CountingCollector.collect(CountingCollector.java:35)
> at
> org.apache.beam.runners.flink.translation.functions.FlinkExecutableStageFunction$MyDataReceiver.accept(FlinkExecutableStageFunction.java:230)
> - locked <0xf6a60bd0> (a java.lang.Object)
> at
> org.apache.beam.sdk.fn.data.BeamFnDataInboundObserver.accept(BeamFnDataInboundObserver.java:81)
> at
> org.apache.beam.sdk.fn.data.BeamFnDataInboundObserver.accept(BeamFnDataInboundObserver.java:32)
> at
> org.apache.beam.sdk.fn.data.BeamFnDataGrpcMultiplexer$InboundObserver.onNext(BeamFnDataGrpcMultiplexer.java:139)
> at
>
[jira] [Commented] (FLINK-10672) Task stuck while writing output to flink
[
https://issues.apache.org/jira/browse/FLINK-10672?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=16973391#comment-16973391
]
venkata subbarao chunduri commented on FLINK-10672:
---
Thank you, Yun.
Yes, the indefinite loop is not an issue. It is my application error, one of
the subpartition is not read by partition client and that lead to exhaustion of
the buffers assigned per partition on the server .
> Task stuck while writing output to flink
>
>
> Key: FLINK-10672
> URL: https://issues.apache.org/jira/browse/FLINK-10672
> Project: Flink
> Issue Type: Bug
> Components: Runtime / Coordination
>Affects Versions: 1.5.4
> Environment: OS: Debuan rodente 4.17
> Flink version: 1.5.4
> ||Key||Value||
> |jobmanager.heap.mb|1024|
> |jobmanager.rpc.address|localhost|
> |jobmanager.rpc.port|6123|
> |metrics.reporter.jmx.class|org.apache.flink.metrics.jmx.JMXReporter|
> |metrics.reporter.jmx.port|9250-9260|
> |metrics.reporters|jmx|
> |parallelism.default|1|
> |rest.port|8081|
> |taskmanager.heap.mb|1024|
> |taskmanager.numberOfTaskSlots|1|
> |web.tmpdir|/tmp/flink-web-bdb73d6c-5b9e-47b5-9ebf-eed0a7c82c26|
>
> h1. Overview
> ||Data Port||All Slots||Free Slots||CPU Cores||Physical Memory||JVM Heap
> Size||Flink Managed Memory||
> |43501|1|0|12|62.9 GB|922 MB|642 MB|
> h1. Memory
> h2. JVM (Heap/Non-Heap)
> ||Type||Committed||Used||Maximum||
> |Heap|922 MB|575 MB|922 MB|
> |Non-Heap|68.8 MB|64.3 MB|-1 B|
> |Total|991 MB|639 MB|922 MB|
> h2. Outside JVM
> ||Type||Count||Used||Capacity||
> |Direct|3,292|105 MB|105 MB|
> |Mapped|0|0 B|0 B|
> h1. Network
> h2. Memory Segments
> ||Type||Count||
> |Available|3,194|
> |Total|3,278|
> h1. Garbage Collection
> ||Collector||Count||Time||
> |G1_Young_Generation|13|336|
> |G1_Old_Generation|1|21|
>Reporter: Ankur Goenka
>Assignee: Yun Gao
>Priority: Major
> Labels: beam
> Attachments: 0.14_all_jobs.jpg, 1uruvakHxBu.png, 3aDKQ24WvKk.png,
> Po89UGDn58V.png, WithBroadcastJob.png, jmx_dump.json, jmx_dump_detailed.json,
> jstack_129827.log, jstack_163822.log, jstack_66985.log
>
>
> I am running a fairly complex pipleline with 200+ task.
> The pipeline works fine with small data (order of 10kb input) but gets stuck
> with a slightly larger data (300kb input).
>
> The task gets stuck while writing the output toFlink, more specifically it
> gets stuck while requesting memory segment in local buffer pool. The Task
> manager UI shows that it has enough memory and memory segments to work with.
> The relevant stack trace is
> {quote}"grpc-default-executor-0" #138 daemon prio=5 os_prio=0
> tid=0x7fedb0163800 nid=0x30b7f in Object.wait() [0x7fedb4f9]
> java.lang.Thread.State: TIMED_WAITING (on object monitor)
> at (C/C++) 0x7fef201c7dae (Unknown Source)
> at (C/C++) 0x7fef1f2aea07 (Unknown Source)
> at (C/C++) 0x7fef1f241cd3 (Unknown Source)
> at java.lang.Object.wait(Native Method)
> - waiting on <0xf6d56450> (a java.util.ArrayDeque)
> at
> org.apache.flink.runtime.io.network.buffer.LocalBufferPool.requestMemorySegment(LocalBufferPool.java:247)
> - locked <0xf6d56450> (a java.util.ArrayDeque)
> at
> org.apache.flink.runtime.io.network.buffer.LocalBufferPool.requestBufferBuilderBlocking(LocalBufferPool.java:204)
> at
> org.apache.flink.runtime.io.network.api.writer.RecordWriter.requestNewBufferBuilder(RecordWriter.java:213)
> at
> org.apache.flink.runtime.io.network.api.writer.RecordWriter.sendToTarget(RecordWriter.java:144)
> at
> org.apache.flink.runtime.io.network.api.writer.RecordWriter.emit(RecordWriter.java:107)
> at
> org.apache.flink.runtime.operators.shipping.OutputCollector.collect(OutputCollector.java:65)
> at
> org.apache.flink.runtime.operators.util.metrics.CountingCollector.collect(CountingCollector.java:35)
> at
> org.apache.beam.runners.flink.translation.functions.FlinkExecutableStagePruningFunction.flatMap(FlinkExecutableStagePruningFunction.java:42)
> at
> org.apache.beam.runners.flink.translation.functions.FlinkExecutableStagePruningFunction.flatMap(FlinkExecutableStagePruningFunction.java:26)
> at
> org.apache.flink.runtime.operators.chaining.ChainedFlatMapDriver.collect(ChainedFlatMapDriver.java:80)
> at
> org.apache.flink.runtime.operators.util.metrics.CountingCollector.collect(CountingCollector.java:35)
> at
> org.apache.beam.runners.flink.translation.functions.FlinkExecutableStageFunction$MyDataReceiver.accept(FlinkExecutableStageFunction.java:230)
> - locked <0xf6a60bd0> (a java.lang.Object)
> at
> org.apache.beam.sdk.fn.data.BeamFnDataInboundObserver.accept(BeamFnDataInboundObserver.java:81)
> at
> org.apache.beam.sdk.fn.data.BeamFnDataInboundObserver.accept(BeamFnDataInboundObserver.java:32)
> at
>
[jira] [Commented] (FLINK-10672) Task stuck while writing output to flink
[
https://issues.apache.org/jira/browse/FLINK-10672?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=16970692#comment-16970692
]
Yun Gao commented on FLINK-10672:
-
Hi venkata, very thanks for the reporting! From my side I think by design local
input channel should not be stuck since when new source start producing data
again, the source thread (which holds the result partition that the local input
channel reads data from) will
[notify|https://github.com/apache/flink/blob/6322618bb0f1b7942d86cb1b2b7bc55290d9e330/flink-runtime/src/main/java/org/apache/flink/runtime/io/network/partition/consumer/LocalInputChannel.java#L202]
the gate about the new data, and then it will unblock the consumer task thread
from the wait() call. Have you also changed this part of codes when
incorporating the RDMA functionality ?
> Task stuck while writing output to flink
>
>
> Key: FLINK-10672
> URL: https://issues.apache.org/jira/browse/FLINK-10672
> Project: Flink
> Issue Type: Bug
> Components: Runtime / Coordination
>Affects Versions: 1.5.4
> Environment: OS: Debuan rodente 4.17
> Flink version: 1.5.4
> ||Key||Value||
> |jobmanager.heap.mb|1024|
> |jobmanager.rpc.address|localhost|
> |jobmanager.rpc.port|6123|
> |metrics.reporter.jmx.class|org.apache.flink.metrics.jmx.JMXReporter|
> |metrics.reporter.jmx.port|9250-9260|
> |metrics.reporters|jmx|
> |parallelism.default|1|
> |rest.port|8081|
> |taskmanager.heap.mb|1024|
> |taskmanager.numberOfTaskSlots|1|
> |web.tmpdir|/tmp/flink-web-bdb73d6c-5b9e-47b5-9ebf-eed0a7c82c26|
>
> h1. Overview
> ||Data Port||All Slots||Free Slots||CPU Cores||Physical Memory||JVM Heap
> Size||Flink Managed Memory||
> |43501|1|0|12|62.9 GB|922 MB|642 MB|
> h1. Memory
> h2. JVM (Heap/Non-Heap)
> ||Type||Committed||Used||Maximum||
> |Heap|922 MB|575 MB|922 MB|
> |Non-Heap|68.8 MB|64.3 MB|-1 B|
> |Total|991 MB|639 MB|922 MB|
> h2. Outside JVM
> ||Type||Count||Used||Capacity||
> |Direct|3,292|105 MB|105 MB|
> |Mapped|0|0 B|0 B|
> h1. Network
> h2. Memory Segments
> ||Type||Count||
> |Available|3,194|
> |Total|3,278|
> h1. Garbage Collection
> ||Collector||Count||Time||
> |G1_Young_Generation|13|336|
> |G1_Old_Generation|1|21|
>Reporter: Ankur Goenka
>Assignee: Yun Gao
>Priority: Major
> Labels: beam
> Attachments: 0.14_all_jobs.jpg, 1uruvakHxBu.png, 3aDKQ24WvKk.png,
> Po89UGDn58V.png, WithBroadcastJob.png, jmx_dump.json, jmx_dump_detailed.json,
> jstack_129827.log, jstack_163822.log, jstack_66985.log
>
>
> I am running a fairly complex pipleline with 200+ task.
> The pipeline works fine with small data (order of 10kb input) but gets stuck
> with a slightly larger data (300kb input).
>
> The task gets stuck while writing the output toFlink, more specifically it
> gets stuck while requesting memory segment in local buffer pool. The Task
> manager UI shows that it has enough memory and memory segments to work with.
> The relevant stack trace is
> {quote}"grpc-default-executor-0" #138 daemon prio=5 os_prio=0
> tid=0x7fedb0163800 nid=0x30b7f in Object.wait() [0x7fedb4f9]
> java.lang.Thread.State: TIMED_WAITING (on object monitor)
> at (C/C++) 0x7fef201c7dae (Unknown Source)
> at (C/C++) 0x7fef1f2aea07 (Unknown Source)
> at (C/C++) 0x7fef1f241cd3 (Unknown Source)
> at java.lang.Object.wait(Native Method)
> - waiting on <0xf6d56450> (a java.util.ArrayDeque)
> at
> org.apache.flink.runtime.io.network.buffer.LocalBufferPool.requestMemorySegment(LocalBufferPool.java:247)
> - locked <0xf6d56450> (a java.util.ArrayDeque)
> at
> org.apache.flink.runtime.io.network.buffer.LocalBufferPool.requestBufferBuilderBlocking(LocalBufferPool.java:204)
> at
> org.apache.flink.runtime.io.network.api.writer.RecordWriter.requestNewBufferBuilder(RecordWriter.java:213)
> at
> org.apache.flink.runtime.io.network.api.writer.RecordWriter.sendToTarget(RecordWriter.java:144)
> at
> org.apache.flink.runtime.io.network.api.writer.RecordWriter.emit(RecordWriter.java:107)
> at
> org.apache.flink.runtime.operators.shipping.OutputCollector.collect(OutputCollector.java:65)
> at
> org.apache.flink.runtime.operators.util.metrics.CountingCollector.collect(CountingCollector.java:35)
> at
> org.apache.beam.runners.flink.translation.functions.FlinkExecutableStagePruningFunction.flatMap(FlinkExecutableStagePruningFunction.java:42)
> at
> org.apache.beam.runners.flink.translation.functions.FlinkExecutableStagePruningFunction.flatMap(FlinkExecutableStagePruningFunction.java:26)
> at
> org.apache.flink.runtime.operators.chaining.ChainedFlatMapDriver.collect(ChainedFlatMapDriver.java:80)
> at
> org.apache.flink.runtime.operators.util.metrics.CountingCollector.collect(CountingCollector.java:35)
> at
>
[jira] [Commented] (FLINK-10672) Task stuck while writing output to flink
[
https://issues.apache.org/jira/browse/FLINK-10672?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=16970360#comment-16970360
]
venkata subbarao chunduri commented on FLINK-10672:
---
Could it be due to local channel? It looks, so.
I am also faced with the same issue. Seems, the local channel [can go to
wait|https://github.com/apache/flink/blob/6322618bb0f1b7942d86cb1b2b7bc55290d9e330/flink-runtime/src/main/java/org/apache/flink/runtime/io/network/partition/consumer/SingleInputGate.java#L539]
and may never wake up as data from the [local channel pulled in the same
thread|https://github.com/apache/flink/blob/6322618bb0f1b7942d86cb1b2b7bc55290d9e330/flink-runtime/src/main/java/org/apache/flink/runtime/io/network/partition/consumer/LocalInputChannel.java#L104].
Could you confirm?
> Task stuck while writing output to flink
>
>
> Key: FLINK-10672
> URL: https://issues.apache.org/jira/browse/FLINK-10672
> Project: Flink
> Issue Type: Bug
> Components: Runtime / Coordination
>Affects Versions: 1.5.4
> Environment: OS: Debuan rodente 4.17
> Flink version: 1.5.4
> ||Key||Value||
> |jobmanager.heap.mb|1024|
> |jobmanager.rpc.address|localhost|
> |jobmanager.rpc.port|6123|
> |metrics.reporter.jmx.class|org.apache.flink.metrics.jmx.JMXReporter|
> |metrics.reporter.jmx.port|9250-9260|
> |metrics.reporters|jmx|
> |parallelism.default|1|
> |rest.port|8081|
> |taskmanager.heap.mb|1024|
> |taskmanager.numberOfTaskSlots|1|
> |web.tmpdir|/tmp/flink-web-bdb73d6c-5b9e-47b5-9ebf-eed0a7c82c26|
>
> h1. Overview
> ||Data Port||All Slots||Free Slots||CPU Cores||Physical Memory||JVM Heap
> Size||Flink Managed Memory||
> |43501|1|0|12|62.9 GB|922 MB|642 MB|
> h1. Memory
> h2. JVM (Heap/Non-Heap)
> ||Type||Committed||Used||Maximum||
> |Heap|922 MB|575 MB|922 MB|
> |Non-Heap|68.8 MB|64.3 MB|-1 B|
> |Total|991 MB|639 MB|922 MB|
> h2. Outside JVM
> ||Type||Count||Used||Capacity||
> |Direct|3,292|105 MB|105 MB|
> |Mapped|0|0 B|0 B|
> h1. Network
> h2. Memory Segments
> ||Type||Count||
> |Available|3,194|
> |Total|3,278|
> h1. Garbage Collection
> ||Collector||Count||Time||
> |G1_Young_Generation|13|336|
> |G1_Old_Generation|1|21|
>Reporter: Ankur Goenka
>Assignee: Yun Gao
>Priority: Major
> Labels: beam
> Attachments: 0.14_all_jobs.jpg, 1uruvakHxBu.png, 3aDKQ24WvKk.png,
> Po89UGDn58V.png, WithBroadcastJob.png, jmx_dump.json, jmx_dump_detailed.json,
> jstack_129827.log, jstack_163822.log, jstack_66985.log
>
>
> I am running a fairly complex pipleline with 200+ task.
> The pipeline works fine with small data (order of 10kb input) but gets stuck
> with a slightly larger data (300kb input).
>
> The task gets stuck while writing the output toFlink, more specifically it
> gets stuck while requesting memory segment in local buffer pool. The Task
> manager UI shows that it has enough memory and memory segments to work with.
> The relevant stack trace is
> {quote}"grpc-default-executor-0" #138 daemon prio=5 os_prio=0
> tid=0x7fedb0163800 nid=0x30b7f in Object.wait() [0x7fedb4f9]
> java.lang.Thread.State: TIMED_WAITING (on object monitor)
> at (C/C++) 0x7fef201c7dae (Unknown Source)
> at (C/C++) 0x7fef1f2aea07 (Unknown Source)
> at (C/C++) 0x7fef1f241cd3 (Unknown Source)
> at java.lang.Object.wait(Native Method)
> - waiting on <0xf6d56450> (a java.util.ArrayDeque)
> at
> org.apache.flink.runtime.io.network.buffer.LocalBufferPool.requestMemorySegment(LocalBufferPool.java:247)
> - locked <0xf6d56450> (a java.util.ArrayDeque)
> at
> org.apache.flink.runtime.io.network.buffer.LocalBufferPool.requestBufferBuilderBlocking(LocalBufferPool.java:204)
> at
> org.apache.flink.runtime.io.network.api.writer.RecordWriter.requestNewBufferBuilder(RecordWriter.java:213)
> at
> org.apache.flink.runtime.io.network.api.writer.RecordWriter.sendToTarget(RecordWriter.java:144)
> at
> org.apache.flink.runtime.io.network.api.writer.RecordWriter.emit(RecordWriter.java:107)
> at
> org.apache.flink.runtime.operators.shipping.OutputCollector.collect(OutputCollector.java:65)
> at
> org.apache.flink.runtime.operators.util.metrics.CountingCollector.collect(CountingCollector.java:35)
> at
> org.apache.beam.runners.flink.translation.functions.FlinkExecutableStagePruningFunction.flatMap(FlinkExecutableStagePruningFunction.java:42)
> at
> org.apache.beam.runners.flink.translation.functions.FlinkExecutableStagePruningFunction.flatMap(FlinkExecutableStagePruningFunction.java:26)
> at
> org.apache.flink.runtime.operators.chaining.ChainedFlatMapDriver.collect(ChainedFlatMapDriver.java:80)
> at
> org.apache.flink.runtime.operators.util.metrics.CountingCollector.collect(CountingCollector.java:35)
> at
>
[jira] [Commented] (FLINK-10672) Task stuck while writing output to flink
[
https://issues.apache.org/jira/browse/FLINK-10672?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=16953107#comment-16953107
]
Kyle Weaver commented on FLINK-10672:
-
This bug was initially filed when stuckness became an issue with parallelism=1.
That was fixed by setting BATCH_FORCED mode. Later, the problem happened again
with BATCH_FORCED and parallelism=12.
This happened for me on TFX 0.13 and 0.14, but I imagine it would happen for
other TFX versions as well. Let me know if you are still unable to reproduce
the issue.
> Task stuck while writing output to flink
>
>
> Key: FLINK-10672
> URL: https://issues.apache.org/jira/browse/FLINK-10672
> Project: Flink
> Issue Type: Bug
> Components: Runtime / Coordination
>Affects Versions: 1.5.4
> Environment: OS: Debuan rodente 4.17
> Flink version: 1.5.4
> ||Key||Value||
> |jobmanager.heap.mb|1024|
> |jobmanager.rpc.address|localhost|
> |jobmanager.rpc.port|6123|
> |metrics.reporter.jmx.class|org.apache.flink.metrics.jmx.JMXReporter|
> |metrics.reporter.jmx.port|9250-9260|
> |metrics.reporters|jmx|
> |parallelism.default|1|
> |rest.port|8081|
> |taskmanager.heap.mb|1024|
> |taskmanager.numberOfTaskSlots|1|
> |web.tmpdir|/tmp/flink-web-bdb73d6c-5b9e-47b5-9ebf-eed0a7c82c26|
>
> h1. Overview
> ||Data Port||All Slots||Free Slots||CPU Cores||Physical Memory||JVM Heap
> Size||Flink Managed Memory||
> |43501|1|0|12|62.9 GB|922 MB|642 MB|
> h1. Memory
> h2. JVM (Heap/Non-Heap)
> ||Type||Committed||Used||Maximum||
> |Heap|922 MB|575 MB|922 MB|
> |Non-Heap|68.8 MB|64.3 MB|-1 B|
> |Total|991 MB|639 MB|922 MB|
> h2. Outside JVM
> ||Type||Count||Used||Capacity||
> |Direct|3,292|105 MB|105 MB|
> |Mapped|0|0 B|0 B|
> h1. Network
> h2. Memory Segments
> ||Type||Count||
> |Available|3,194|
> |Total|3,278|
> h1. Garbage Collection
> ||Collector||Count||Time||
> |G1_Young_Generation|13|336|
> |G1_Old_Generation|1|21|
>Reporter: Ankur Goenka
>Assignee: Yun Gao
>Priority: Major
> Labels: beam
> Attachments: 0.14_all_jobs.jpg, 1uruvakHxBu.png, 3aDKQ24WvKk.png,
> Po89UGDn58V.png, WithBroadcastJob.png, jmx_dump.json, jmx_dump_detailed.json,
> jstack_129827.log, jstack_163822.log, jstack_66985.log
>
>
> I am running a fairly complex pipleline with 200+ task.
> The pipeline works fine with small data (order of 10kb input) but gets stuck
> with a slightly larger data (300kb input).
>
> The task gets stuck while writing the output toFlink, more specifically it
> gets stuck while requesting memory segment in local buffer pool. The Task
> manager UI shows that it has enough memory and memory segments to work with.
> The relevant stack trace is
> {quote}"grpc-default-executor-0" #138 daemon prio=5 os_prio=0
> tid=0x7fedb0163800 nid=0x30b7f in Object.wait() [0x7fedb4f9]
> java.lang.Thread.State: TIMED_WAITING (on object monitor)
> at (C/C++) 0x7fef201c7dae (Unknown Source)
> at (C/C++) 0x7fef1f2aea07 (Unknown Source)
> at (C/C++) 0x7fef1f241cd3 (Unknown Source)
> at java.lang.Object.wait(Native Method)
> - waiting on <0xf6d56450> (a java.util.ArrayDeque)
> at
> org.apache.flink.runtime.io.network.buffer.LocalBufferPool.requestMemorySegment(LocalBufferPool.java:247)
> - locked <0xf6d56450> (a java.util.ArrayDeque)
> at
> org.apache.flink.runtime.io.network.buffer.LocalBufferPool.requestBufferBuilderBlocking(LocalBufferPool.java:204)
> at
> org.apache.flink.runtime.io.network.api.writer.RecordWriter.requestNewBufferBuilder(RecordWriter.java:213)
> at
> org.apache.flink.runtime.io.network.api.writer.RecordWriter.sendToTarget(RecordWriter.java:144)
> at
> org.apache.flink.runtime.io.network.api.writer.RecordWriter.emit(RecordWriter.java:107)
> at
> org.apache.flink.runtime.operators.shipping.OutputCollector.collect(OutputCollector.java:65)
> at
> org.apache.flink.runtime.operators.util.metrics.CountingCollector.collect(CountingCollector.java:35)
> at
> org.apache.beam.runners.flink.translation.functions.FlinkExecutableStagePruningFunction.flatMap(FlinkExecutableStagePruningFunction.java:42)
> at
> org.apache.beam.runners.flink.translation.functions.FlinkExecutableStagePruningFunction.flatMap(FlinkExecutableStagePruningFunction.java:26)
> at
> org.apache.flink.runtime.operators.chaining.ChainedFlatMapDriver.collect(ChainedFlatMapDriver.java:80)
> at
> org.apache.flink.runtime.operators.util.metrics.CountingCollector.collect(CountingCollector.java:35)
> at
> org.apache.beam.runners.flink.translation.functions.FlinkExecutableStageFunction$MyDataReceiver.accept(FlinkExecutableStageFunction.java:230)
> - locked <0xf6a60bd0> (a java.lang.Object)
> at
> org.apache.beam.sdk.fn.data.BeamFnDataInboundObserver.accept(BeamFnDataInboundObserver.java:81)
> at
>
[jira] [Commented] (FLINK-10672) Task stuck while writing output to flink
[
https://issues.apache.org/jira/browse/FLINK-10672?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=16953016#comment-16953016
]
Yun Gao commented on FLINK-10672:
-
Hi Kyle, very thanks for the input! I still have two confusions, the first is
that the the attached jstack logs seems to be running with parallelism equals
to 1, therefore is the stuck also exists for parallelism = 1? Besides, could
you also provide me the version of TFX/Beam/Tensorflow used ? The job I ran
seems to be different from the job corresponding to the attached jstack.
> Task stuck while writing output to flink
>
>
> Key: FLINK-10672
> URL: https://issues.apache.org/jira/browse/FLINK-10672
> Project: Flink
> Issue Type: Bug
> Components: Runtime / Coordination
>Affects Versions: 1.5.4
> Environment: OS: Debuan rodente 4.17
> Flink version: 1.5.4
> ||Key||Value||
> |jobmanager.heap.mb|1024|
> |jobmanager.rpc.address|localhost|
> |jobmanager.rpc.port|6123|
> |metrics.reporter.jmx.class|org.apache.flink.metrics.jmx.JMXReporter|
> |metrics.reporter.jmx.port|9250-9260|
> |metrics.reporters|jmx|
> |parallelism.default|1|
> |rest.port|8081|
> |taskmanager.heap.mb|1024|
> |taskmanager.numberOfTaskSlots|1|
> |web.tmpdir|/tmp/flink-web-bdb73d6c-5b9e-47b5-9ebf-eed0a7c82c26|
>
> h1. Overview
> ||Data Port||All Slots||Free Slots||CPU Cores||Physical Memory||JVM Heap
> Size||Flink Managed Memory||
> |43501|1|0|12|62.9 GB|922 MB|642 MB|
> h1. Memory
> h2. JVM (Heap/Non-Heap)
> ||Type||Committed||Used||Maximum||
> |Heap|922 MB|575 MB|922 MB|
> |Non-Heap|68.8 MB|64.3 MB|-1 B|
> |Total|991 MB|639 MB|922 MB|
> h2. Outside JVM
> ||Type||Count||Used||Capacity||
> |Direct|3,292|105 MB|105 MB|
> |Mapped|0|0 B|0 B|
> h1. Network
> h2. Memory Segments
> ||Type||Count||
> |Available|3,194|
> |Total|3,278|
> h1. Garbage Collection
> ||Collector||Count||Time||
> |G1_Young_Generation|13|336|
> |G1_Old_Generation|1|21|
>Reporter: Ankur Goenka
>Assignee: Yun Gao
>Priority: Major
> Labels: beam
> Attachments: 0.14_all_jobs.jpg, 1uruvakHxBu.png, 3aDKQ24WvKk.png,
> Po89UGDn58V.png, WithBroadcastJob.png, jmx_dump.json, jmx_dump_detailed.json,
> jstack_129827.log, jstack_163822.log, jstack_66985.log
>
>
> I am running a fairly complex pipleline with 200+ task.
> The pipeline works fine with small data (order of 10kb input) but gets stuck
> with a slightly larger data (300kb input).
>
> The task gets stuck while writing the output toFlink, more specifically it
> gets stuck while requesting memory segment in local buffer pool. The Task
> manager UI shows that it has enough memory and memory segments to work with.
> The relevant stack trace is
> {quote}"grpc-default-executor-0" #138 daemon prio=5 os_prio=0
> tid=0x7fedb0163800 nid=0x30b7f in Object.wait() [0x7fedb4f9]
> java.lang.Thread.State: TIMED_WAITING (on object monitor)
> at (C/C++) 0x7fef201c7dae (Unknown Source)
> at (C/C++) 0x7fef1f2aea07 (Unknown Source)
> at (C/C++) 0x7fef1f241cd3 (Unknown Source)
> at java.lang.Object.wait(Native Method)
> - waiting on <0xf6d56450> (a java.util.ArrayDeque)
> at
> org.apache.flink.runtime.io.network.buffer.LocalBufferPool.requestMemorySegment(LocalBufferPool.java:247)
> - locked <0xf6d56450> (a java.util.ArrayDeque)
> at
> org.apache.flink.runtime.io.network.buffer.LocalBufferPool.requestBufferBuilderBlocking(LocalBufferPool.java:204)
> at
> org.apache.flink.runtime.io.network.api.writer.RecordWriter.requestNewBufferBuilder(RecordWriter.java:213)
> at
> org.apache.flink.runtime.io.network.api.writer.RecordWriter.sendToTarget(RecordWriter.java:144)
> at
> org.apache.flink.runtime.io.network.api.writer.RecordWriter.emit(RecordWriter.java:107)
> at
> org.apache.flink.runtime.operators.shipping.OutputCollector.collect(OutputCollector.java:65)
> at
> org.apache.flink.runtime.operators.util.metrics.CountingCollector.collect(CountingCollector.java:35)
> at
> org.apache.beam.runners.flink.translation.functions.FlinkExecutableStagePruningFunction.flatMap(FlinkExecutableStagePruningFunction.java:42)
> at
> org.apache.beam.runners.flink.translation.functions.FlinkExecutableStagePruningFunction.flatMap(FlinkExecutableStagePruningFunction.java:26)
> at
> org.apache.flink.runtime.operators.chaining.ChainedFlatMapDriver.collect(ChainedFlatMapDriver.java:80)
> at
> org.apache.flink.runtime.operators.util.metrics.CountingCollector.collect(CountingCollector.java:35)
> at
> org.apache.beam.runners.flink.translation.functions.FlinkExecutableStageFunction$MyDataReceiver.accept(FlinkExecutableStageFunction.java:230)
> - locked <0xf6a60bd0> (a java.lang.Object)
> at
> org.apache.beam.sdk.fn.data.BeamFnDataInboundObserver.accept(BeamFnDataInboundObserver.java:81)
>
[jira] [Commented] (FLINK-10672) Task stuck while writing output to flink
[
https://issues.apache.org/jira/browse/FLINK-10672?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=16952809#comment-16952809
]
Kyle Weaver commented on FLINK-10672:
-
You will need to set the task parallelism to reproduce the stuck job. For me it
happens when parallelism=12, which is also the number of cores on my machine
(might or might not be coincidence). Anyway, you can experiment with different
parallelism settings and see what happens.
> Task stuck while writing output to flink
>
>
> Key: FLINK-10672
> URL: https://issues.apache.org/jira/browse/FLINK-10672
> Project: Flink
> Issue Type: Bug
> Components: Runtime / Coordination
>Affects Versions: 1.5.4
> Environment: OS: Debuan rodente 4.17
> Flink version: 1.5.4
> ||Key||Value||
> |jobmanager.heap.mb|1024|
> |jobmanager.rpc.address|localhost|
> |jobmanager.rpc.port|6123|
> |metrics.reporter.jmx.class|org.apache.flink.metrics.jmx.JMXReporter|
> |metrics.reporter.jmx.port|9250-9260|
> |metrics.reporters|jmx|
> |parallelism.default|1|
> |rest.port|8081|
> |taskmanager.heap.mb|1024|
> |taskmanager.numberOfTaskSlots|1|
> |web.tmpdir|/tmp/flink-web-bdb73d6c-5b9e-47b5-9ebf-eed0a7c82c26|
>
> h1. Overview
> ||Data Port||All Slots||Free Slots||CPU Cores||Physical Memory||JVM Heap
> Size||Flink Managed Memory||
> |43501|1|0|12|62.9 GB|922 MB|642 MB|
> h1. Memory
> h2. JVM (Heap/Non-Heap)
> ||Type||Committed||Used||Maximum||
> |Heap|922 MB|575 MB|922 MB|
> |Non-Heap|68.8 MB|64.3 MB|-1 B|
> |Total|991 MB|639 MB|922 MB|
> h2. Outside JVM
> ||Type||Count||Used||Capacity||
> |Direct|3,292|105 MB|105 MB|
> |Mapped|0|0 B|0 B|
> h1. Network
> h2. Memory Segments
> ||Type||Count||
> |Available|3,194|
> |Total|3,278|
> h1. Garbage Collection
> ||Collector||Count||Time||
> |G1_Young_Generation|13|336|
> |G1_Old_Generation|1|21|
>Reporter: Ankur Goenka
>Assignee: Yun Gao
>Priority: Major
> Labels: beam
> Attachments: 0.14_all_jobs.jpg, 1uruvakHxBu.png, 3aDKQ24WvKk.png,
> Po89UGDn58V.png, WithBroadcastJob.png, jmx_dump.json, jmx_dump_detailed.json,
> jstack_129827.log, jstack_163822.log, jstack_66985.log
>
>
> I am running a fairly complex pipleline with 200+ task.
> The pipeline works fine with small data (order of 10kb input) but gets stuck
> with a slightly larger data (300kb input).
>
> The task gets stuck while writing the output toFlink, more specifically it
> gets stuck while requesting memory segment in local buffer pool. The Task
> manager UI shows that it has enough memory and memory segments to work with.
> The relevant stack trace is
> {quote}"grpc-default-executor-0" #138 daemon prio=5 os_prio=0
> tid=0x7fedb0163800 nid=0x30b7f in Object.wait() [0x7fedb4f9]
> java.lang.Thread.State: TIMED_WAITING (on object monitor)
> at (C/C++) 0x7fef201c7dae (Unknown Source)
> at (C/C++) 0x7fef1f2aea07 (Unknown Source)
> at (C/C++) 0x7fef1f241cd3 (Unknown Source)
> at java.lang.Object.wait(Native Method)
> - waiting on <0xf6d56450> (a java.util.ArrayDeque)
> at
> org.apache.flink.runtime.io.network.buffer.LocalBufferPool.requestMemorySegment(LocalBufferPool.java:247)
> - locked <0xf6d56450> (a java.util.ArrayDeque)
> at
> org.apache.flink.runtime.io.network.buffer.LocalBufferPool.requestBufferBuilderBlocking(LocalBufferPool.java:204)
> at
> org.apache.flink.runtime.io.network.api.writer.RecordWriter.requestNewBufferBuilder(RecordWriter.java:213)
> at
> org.apache.flink.runtime.io.network.api.writer.RecordWriter.sendToTarget(RecordWriter.java:144)
> at
> org.apache.flink.runtime.io.network.api.writer.RecordWriter.emit(RecordWriter.java:107)
> at
> org.apache.flink.runtime.operators.shipping.OutputCollector.collect(OutputCollector.java:65)
> at
> org.apache.flink.runtime.operators.util.metrics.CountingCollector.collect(CountingCollector.java:35)
> at
> org.apache.beam.runners.flink.translation.functions.FlinkExecutableStagePruningFunction.flatMap(FlinkExecutableStagePruningFunction.java:42)
> at
> org.apache.beam.runners.flink.translation.functions.FlinkExecutableStagePruningFunction.flatMap(FlinkExecutableStagePruningFunction.java:26)
> at
> org.apache.flink.runtime.operators.chaining.ChainedFlatMapDriver.collect(ChainedFlatMapDriver.java:80)
> at
> org.apache.flink.runtime.operators.util.metrics.CountingCollector.collect(CountingCollector.java:35)
> at
> org.apache.beam.runners.flink.translation.functions.FlinkExecutableStageFunction$MyDataReceiver.accept(FlinkExecutableStageFunction.java:230)
> - locked <0xf6a60bd0> (a java.lang.Object)
> at
> org.apache.beam.sdk.fn.data.BeamFnDataInboundObserver.accept(BeamFnDataInboundObserver.java:81)
> at
> org.apache.beam.sdk.fn.data.BeamFnDataInboundObserver.accept(BeamFnDataInboundObserver.java:32)
[jira] [Commented] (FLINK-10672) Task stuck while writing output to flink
[
https://issues.apache.org/jira/browse/FLINK-10672?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=16951507#comment-16951507
]
Yun Gao commented on FLINK-10672:
-
Hi [~angoenka], very sorry for the long delay because of meeting with some
package conflict problems. I have run the chicago taxi test with TFX 0.14, Beam
2.14 and Tensorflow 1.14, however the job seems to be different from the
attached one, as [^0.14_all_jobs.jpg] shows, there are 4 jobs in total and the
largest one only has 145 tasks in total. Therefore could you also provide me
the version or branch id of TFX/Beam/Tensorflow ? Besides, do I need to change
some configuration to reproduce the stuck job (For example, do I need to change
the default dataset) ? Very thanks for the explanation. :)
> Task stuck while writing output to flink
>
>
> Key: FLINK-10672
> URL: https://issues.apache.org/jira/browse/FLINK-10672
> Project: Flink
> Issue Type: Bug
> Components: Runtime / Coordination
>Affects Versions: 1.5.4
> Environment: OS: Debuan rodente 4.17
> Flink version: 1.5.4
> ||Key||Value||
> |jobmanager.heap.mb|1024|
> |jobmanager.rpc.address|localhost|
> |jobmanager.rpc.port|6123|
> |metrics.reporter.jmx.class|org.apache.flink.metrics.jmx.JMXReporter|
> |metrics.reporter.jmx.port|9250-9260|
> |metrics.reporters|jmx|
> |parallelism.default|1|
> |rest.port|8081|
> |taskmanager.heap.mb|1024|
> |taskmanager.numberOfTaskSlots|1|
> |web.tmpdir|/tmp/flink-web-bdb73d6c-5b9e-47b5-9ebf-eed0a7c82c26|
>
> h1. Overview
> ||Data Port||All Slots||Free Slots||CPU Cores||Physical Memory||JVM Heap
> Size||Flink Managed Memory||
> |43501|1|0|12|62.9 GB|922 MB|642 MB|
> h1. Memory
> h2. JVM (Heap/Non-Heap)
> ||Type||Committed||Used||Maximum||
> |Heap|922 MB|575 MB|922 MB|
> |Non-Heap|68.8 MB|64.3 MB|-1 B|
> |Total|991 MB|639 MB|922 MB|
> h2. Outside JVM
> ||Type||Count||Used||Capacity||
> |Direct|3,292|105 MB|105 MB|
> |Mapped|0|0 B|0 B|
> h1. Network
> h2. Memory Segments
> ||Type||Count||
> |Available|3,194|
> |Total|3,278|
> h1. Garbage Collection
> ||Collector||Count||Time||
> |G1_Young_Generation|13|336|
> |G1_Old_Generation|1|21|
>Reporter: Ankur Goenka
>Assignee: Yun Gao
>Priority: Major
> Labels: beam
> Attachments: 0.14_all_jobs.jpg, 1uruvakHxBu.png, 3aDKQ24WvKk.png,
> Po89UGDn58V.png, jmx_dump.json, jmx_dump_detailed.json, jstack_129827.log,
> jstack_163822.log, jstack_66985.log
>
>
> I am running a fairly complex pipleline with 200+ task.
> The pipeline works fine with small data (order of 10kb input) but gets stuck
> with a slightly larger data (300kb input).
>
> The task gets stuck while writing the output toFlink, more specifically it
> gets stuck while requesting memory segment in local buffer pool. The Task
> manager UI shows that it has enough memory and memory segments to work with.
> The relevant stack trace is
> {quote}"grpc-default-executor-0" #138 daemon prio=5 os_prio=0
> tid=0x7fedb0163800 nid=0x30b7f in Object.wait() [0x7fedb4f9]
> java.lang.Thread.State: TIMED_WAITING (on object monitor)
> at (C/C++) 0x7fef201c7dae (Unknown Source)
> at (C/C++) 0x7fef1f2aea07 (Unknown Source)
> at (C/C++) 0x7fef1f241cd3 (Unknown Source)
> at java.lang.Object.wait(Native Method)
> - waiting on <0xf6d56450> (a java.util.ArrayDeque)
> at
> org.apache.flink.runtime.io.network.buffer.LocalBufferPool.requestMemorySegment(LocalBufferPool.java:247)
> - locked <0xf6d56450> (a java.util.ArrayDeque)
> at
> org.apache.flink.runtime.io.network.buffer.LocalBufferPool.requestBufferBuilderBlocking(LocalBufferPool.java:204)
> at
> org.apache.flink.runtime.io.network.api.writer.RecordWriter.requestNewBufferBuilder(RecordWriter.java:213)
> at
> org.apache.flink.runtime.io.network.api.writer.RecordWriter.sendToTarget(RecordWriter.java:144)
> at
> org.apache.flink.runtime.io.network.api.writer.RecordWriter.emit(RecordWriter.java:107)
> at
> org.apache.flink.runtime.operators.shipping.OutputCollector.collect(OutputCollector.java:65)
> at
> org.apache.flink.runtime.operators.util.metrics.CountingCollector.collect(CountingCollector.java:35)
> at
> org.apache.beam.runners.flink.translation.functions.FlinkExecutableStagePruningFunction.flatMap(FlinkExecutableStagePruningFunction.java:42)
> at
> org.apache.beam.runners.flink.translation.functions.FlinkExecutableStagePruningFunction.flatMap(FlinkExecutableStagePruningFunction.java:26)
> at
> org.apache.flink.runtime.operators.chaining.ChainedFlatMapDriver.collect(ChainedFlatMapDriver.java:80)
> at
> org.apache.flink.runtime.operators.util.metrics.CountingCollector.collect(CountingCollector.java:35)
> at
>
[jira] [Commented] (FLINK-10672) Task stuck while writing output to flink
[
https://issues.apache.org/jira/browse/FLINK-10672?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=16946410#comment-16946410
]
Yun Gao commented on FLINK-10672:
-
Hi [~angoenka] Very thanks for the input and very sorry for responding so late
because of being in the holiday of National's Day in the past week, I will
reproduce the problem as soon as possible.
> Task stuck while writing output to flink
>
>
> Key: FLINK-10672
> URL: https://issues.apache.org/jira/browse/FLINK-10672
> Project: Flink
> Issue Type: Bug
> Components: Runtime / Coordination
>Affects Versions: 1.5.4
> Environment: OS: Debuan rodente 4.17
> Flink version: 1.5.4
> ||Key||Value||
> |jobmanager.heap.mb|1024|
> |jobmanager.rpc.address|localhost|
> |jobmanager.rpc.port|6123|
> |metrics.reporter.jmx.class|org.apache.flink.metrics.jmx.JMXReporter|
> |metrics.reporter.jmx.port|9250-9260|
> |metrics.reporters|jmx|
> |parallelism.default|1|
> |rest.port|8081|
> |taskmanager.heap.mb|1024|
> |taskmanager.numberOfTaskSlots|1|
> |web.tmpdir|/tmp/flink-web-bdb73d6c-5b9e-47b5-9ebf-eed0a7c82c26|
>
> h1. Overview
> ||Data Port||All Slots||Free Slots||CPU Cores||Physical Memory||JVM Heap
> Size||Flink Managed Memory||
> |43501|1|0|12|62.9 GB|922 MB|642 MB|
> h1. Memory
> h2. JVM (Heap/Non-Heap)
> ||Type||Committed||Used||Maximum||
> |Heap|922 MB|575 MB|922 MB|
> |Non-Heap|68.8 MB|64.3 MB|-1 B|
> |Total|991 MB|639 MB|922 MB|
> h2. Outside JVM
> ||Type||Count||Used||Capacity||
> |Direct|3,292|105 MB|105 MB|
> |Mapped|0|0 B|0 B|
> h1. Network
> h2. Memory Segments
> ||Type||Count||
> |Available|3,194|
> |Total|3,278|
> h1. Garbage Collection
> ||Collector||Count||Time||
> |G1_Young_Generation|13|336|
> |G1_Old_Generation|1|21|
>Reporter: Ankur Goenka
>Assignee: Yun Gao
>Priority: Major
> Labels: beam
> Attachments: 1uruvakHxBu.png, 3aDKQ24WvKk.png, Po89UGDn58V.png,
> jmx_dump.json, jmx_dump_detailed.json, jstack_129827.log, jstack_163822.log,
> jstack_66985.log
>
>
> I am running a fairly complex pipleline with 200+ task.
> The pipeline works fine with small data (order of 10kb input) but gets stuck
> with a slightly larger data (300kb input).
>
> The task gets stuck while writing the output toFlink, more specifically it
> gets stuck while requesting memory segment in local buffer pool. The Task
> manager UI shows that it has enough memory and memory segments to work with.
> The relevant stack trace is
> {quote}"grpc-default-executor-0" #138 daemon prio=5 os_prio=0
> tid=0x7fedb0163800 nid=0x30b7f in Object.wait() [0x7fedb4f9]
> java.lang.Thread.State: TIMED_WAITING (on object monitor)
> at (C/C++) 0x7fef201c7dae (Unknown Source)
> at (C/C++) 0x7fef1f2aea07 (Unknown Source)
> at (C/C++) 0x7fef1f241cd3 (Unknown Source)
> at java.lang.Object.wait(Native Method)
> - waiting on <0xf6d56450> (a java.util.ArrayDeque)
> at
> org.apache.flink.runtime.io.network.buffer.LocalBufferPool.requestMemorySegment(LocalBufferPool.java:247)
> - locked <0xf6d56450> (a java.util.ArrayDeque)
> at
> org.apache.flink.runtime.io.network.buffer.LocalBufferPool.requestBufferBuilderBlocking(LocalBufferPool.java:204)
> at
> org.apache.flink.runtime.io.network.api.writer.RecordWriter.requestNewBufferBuilder(RecordWriter.java:213)
> at
> org.apache.flink.runtime.io.network.api.writer.RecordWriter.sendToTarget(RecordWriter.java:144)
> at
> org.apache.flink.runtime.io.network.api.writer.RecordWriter.emit(RecordWriter.java:107)
> at
> org.apache.flink.runtime.operators.shipping.OutputCollector.collect(OutputCollector.java:65)
> at
> org.apache.flink.runtime.operators.util.metrics.CountingCollector.collect(CountingCollector.java:35)
> at
> org.apache.beam.runners.flink.translation.functions.FlinkExecutableStagePruningFunction.flatMap(FlinkExecutableStagePruningFunction.java:42)
> at
> org.apache.beam.runners.flink.translation.functions.FlinkExecutableStagePruningFunction.flatMap(FlinkExecutableStagePruningFunction.java:26)
> at
> org.apache.flink.runtime.operators.chaining.ChainedFlatMapDriver.collect(ChainedFlatMapDriver.java:80)
> at
> org.apache.flink.runtime.operators.util.metrics.CountingCollector.collect(CountingCollector.java:35)
> at
> org.apache.beam.runners.flink.translation.functions.FlinkExecutableStageFunction$MyDataReceiver.accept(FlinkExecutableStageFunction.java:230)
> - locked <0xf6a60bd0> (a java.lang.Object)
> at
> org.apache.beam.sdk.fn.data.BeamFnDataInboundObserver.accept(BeamFnDataInboundObserver.java:81)
> at
> org.apache.beam.sdk.fn.data.BeamFnDataInboundObserver.accept(BeamFnDataInboundObserver.java:32)
> at
> org.apache.beam.sdk.fn.data.BeamFnDataGrpcMultiplexer$InboundObserver.onNext(BeamFnDataGrpcMultiplexer.java:139)
> at
>
[jira] [Commented] (FLINK-10672) Task stuck while writing output to flink
[
https://issues.apache.org/jira/browse/FLINK-10672?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=16942291#comment-16942291
]
Ankur Goenka commented on FLINK-10672:
--
I don't have the original graph as the code has changed.
But the problem still exists when running the following steps with parallelism
12.
Setup flink using
[https://github.com/tensorflow/tfx/blob/master/tfx/examples/chicago_taxi/setup_beam_on_flink.sh]
Run the pipeline
#
[https://github.com/tensorflow/tfx/blob/master/tfx/examples/chicago_taxi/tfdv_analyze_and_validate_portable_beam.sh]
#
[https://github.com/tensorflow/tfx/blob/master/tfx/examples/chicago_taxi/preprocess_portable_beam.sh]
You can view the graph by just running it with parallelism 1.
> Task stuck while writing output to flink
>
>
> Key: FLINK-10672
> URL: https://issues.apache.org/jira/browse/FLINK-10672
> Project: Flink
> Issue Type: Bug
> Components: Runtime / Coordination
>Affects Versions: 1.5.4
> Environment: OS: Debuan rodente 4.17
> Flink version: 1.5.4
> ||Key||Value||
> |jobmanager.heap.mb|1024|
> |jobmanager.rpc.address|localhost|
> |jobmanager.rpc.port|6123|
> |metrics.reporter.jmx.class|org.apache.flink.metrics.jmx.JMXReporter|
> |metrics.reporter.jmx.port|9250-9260|
> |metrics.reporters|jmx|
> |parallelism.default|1|
> |rest.port|8081|
> |taskmanager.heap.mb|1024|
> |taskmanager.numberOfTaskSlots|1|
> |web.tmpdir|/tmp/flink-web-bdb73d6c-5b9e-47b5-9ebf-eed0a7c82c26|
>
> h1. Overview
> ||Data Port||All Slots||Free Slots||CPU Cores||Physical Memory||JVM Heap
> Size||Flink Managed Memory||
> |43501|1|0|12|62.9 GB|922 MB|642 MB|
> h1. Memory
> h2. JVM (Heap/Non-Heap)
> ||Type||Committed||Used||Maximum||
> |Heap|922 MB|575 MB|922 MB|
> |Non-Heap|68.8 MB|64.3 MB|-1 B|
> |Total|991 MB|639 MB|922 MB|
> h2. Outside JVM
> ||Type||Count||Used||Capacity||
> |Direct|3,292|105 MB|105 MB|
> |Mapped|0|0 B|0 B|
> h1. Network
> h2. Memory Segments
> ||Type||Count||
> |Available|3,194|
> |Total|3,278|
> h1. Garbage Collection
> ||Collector||Count||Time||
> |G1_Young_Generation|13|336|
> |G1_Old_Generation|1|21|
>Reporter: Ankur Goenka
>Assignee: Yun Gao
>Priority: Major
> Labels: beam
> Attachments: 1uruvakHxBu.png, 3aDKQ24WvKk.png, Po89UGDn58V.png,
> jmx_dump.json, jmx_dump_detailed.json, jstack_129827.log, jstack_163822.log,
> jstack_66985.log
>
>
> I am running a fairly complex pipleline with 200+ task.
> The pipeline works fine with small data (order of 10kb input) but gets stuck
> with a slightly larger data (300kb input).
>
> The task gets stuck while writing the output toFlink, more specifically it
> gets stuck while requesting memory segment in local buffer pool. The Task
> manager UI shows that it has enough memory and memory segments to work with.
> The relevant stack trace is
> {quote}"grpc-default-executor-0" #138 daemon prio=5 os_prio=0
> tid=0x7fedb0163800 nid=0x30b7f in Object.wait() [0x7fedb4f9]
> java.lang.Thread.State: TIMED_WAITING (on object monitor)
> at (C/C++) 0x7fef201c7dae (Unknown Source)
> at (C/C++) 0x7fef1f2aea07 (Unknown Source)
> at (C/C++) 0x7fef1f241cd3 (Unknown Source)
> at java.lang.Object.wait(Native Method)
> - waiting on <0xf6d56450> (a java.util.ArrayDeque)
> at
> org.apache.flink.runtime.io.network.buffer.LocalBufferPool.requestMemorySegment(LocalBufferPool.java:247)
> - locked <0xf6d56450> (a java.util.ArrayDeque)
> at
> org.apache.flink.runtime.io.network.buffer.LocalBufferPool.requestBufferBuilderBlocking(LocalBufferPool.java:204)
> at
> org.apache.flink.runtime.io.network.api.writer.RecordWriter.requestNewBufferBuilder(RecordWriter.java:213)
> at
> org.apache.flink.runtime.io.network.api.writer.RecordWriter.sendToTarget(RecordWriter.java:144)
> at
> org.apache.flink.runtime.io.network.api.writer.RecordWriter.emit(RecordWriter.java:107)
> at
> org.apache.flink.runtime.operators.shipping.OutputCollector.collect(OutputCollector.java:65)
> at
> org.apache.flink.runtime.operators.util.metrics.CountingCollector.collect(CountingCollector.java:35)
> at
> org.apache.beam.runners.flink.translation.functions.FlinkExecutableStagePruningFunction.flatMap(FlinkExecutableStagePruningFunction.java:42)
> at
> org.apache.beam.runners.flink.translation.functions.FlinkExecutableStagePruningFunction.flatMap(FlinkExecutableStagePruningFunction.java:26)
> at
> org.apache.flink.runtime.operators.chaining.ChainedFlatMapDriver.collect(ChainedFlatMapDriver.java:80)
> at
> org.apache.flink.runtime.operators.util.metrics.CountingCollector.collect(CountingCollector.java:35)
> at
> org.apache.beam.runners.flink.translation.functions.FlinkExecutableStageFunction$MyDataReceiver.accept(FlinkExecutableStageFunction.java:230)
>
[jira] [Commented] (FLINK-10672) Task stuck while writing output to flink
[
https://issues.apache.org/jira/browse/FLINK-10672?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=16941796#comment-16941796
]
Maximilian Michels commented on FLINK-10672:
[~angoenka] Do you think you could provide that, or at least the steps to
reproduce the behavior?
> Task stuck while writing output to flink
>
>
> Key: FLINK-10672
> URL: https://issues.apache.org/jira/browse/FLINK-10672
> Project: Flink
> Issue Type: Bug
> Components: Runtime / Coordination
>Affects Versions: 1.5.4
> Environment: OS: Debuan rodente 4.17
> Flink version: 1.5.4
> ||Key||Value||
> |jobmanager.heap.mb|1024|
> |jobmanager.rpc.address|localhost|
> |jobmanager.rpc.port|6123|
> |metrics.reporter.jmx.class|org.apache.flink.metrics.jmx.JMXReporter|
> |metrics.reporter.jmx.port|9250-9260|
> |metrics.reporters|jmx|
> |parallelism.default|1|
> |rest.port|8081|
> |taskmanager.heap.mb|1024|
> |taskmanager.numberOfTaskSlots|1|
> |web.tmpdir|/tmp/flink-web-bdb73d6c-5b9e-47b5-9ebf-eed0a7c82c26|
>
> h1. Overview
> ||Data Port||All Slots||Free Slots||CPU Cores||Physical Memory||JVM Heap
> Size||Flink Managed Memory||
> |43501|1|0|12|62.9 GB|922 MB|642 MB|
> h1. Memory
> h2. JVM (Heap/Non-Heap)
> ||Type||Committed||Used||Maximum||
> |Heap|922 MB|575 MB|922 MB|
> |Non-Heap|68.8 MB|64.3 MB|-1 B|
> |Total|991 MB|639 MB|922 MB|
> h2. Outside JVM
> ||Type||Count||Used||Capacity||
> |Direct|3,292|105 MB|105 MB|
> |Mapped|0|0 B|0 B|
> h1. Network
> h2. Memory Segments
> ||Type||Count||
> |Available|3,194|
> |Total|3,278|
> h1. Garbage Collection
> ||Collector||Count||Time||
> |G1_Young_Generation|13|336|
> |G1_Old_Generation|1|21|
>Reporter: Ankur Goenka
>Assignee: Yun Gao
>Priority: Major
> Labels: beam
> Attachments: 1uruvakHxBu.png, 3aDKQ24WvKk.png, Po89UGDn58V.png,
> jmx_dump.json, jmx_dump_detailed.json, jstack_129827.log, jstack_163822.log,
> jstack_66985.log
>
>
> I am running a fairly complex pipleline with 200+ task.
> The pipeline works fine with small data (order of 10kb input) but gets stuck
> with a slightly larger data (300kb input).
>
> The task gets stuck while writing the output toFlink, more specifically it
> gets stuck while requesting memory segment in local buffer pool. The Task
> manager UI shows that it has enough memory and memory segments to work with.
> The relevant stack trace is
> {quote}"grpc-default-executor-0" #138 daemon prio=5 os_prio=0
> tid=0x7fedb0163800 nid=0x30b7f in Object.wait() [0x7fedb4f9]
> java.lang.Thread.State: TIMED_WAITING (on object monitor)
> at (C/C++) 0x7fef201c7dae (Unknown Source)
> at (C/C++) 0x7fef1f2aea07 (Unknown Source)
> at (C/C++) 0x7fef1f241cd3 (Unknown Source)
> at java.lang.Object.wait(Native Method)
> - waiting on <0xf6d56450> (a java.util.ArrayDeque)
> at
> org.apache.flink.runtime.io.network.buffer.LocalBufferPool.requestMemorySegment(LocalBufferPool.java:247)
> - locked <0xf6d56450> (a java.util.ArrayDeque)
> at
> org.apache.flink.runtime.io.network.buffer.LocalBufferPool.requestBufferBuilderBlocking(LocalBufferPool.java:204)
> at
> org.apache.flink.runtime.io.network.api.writer.RecordWriter.requestNewBufferBuilder(RecordWriter.java:213)
> at
> org.apache.flink.runtime.io.network.api.writer.RecordWriter.sendToTarget(RecordWriter.java:144)
> at
> org.apache.flink.runtime.io.network.api.writer.RecordWriter.emit(RecordWriter.java:107)
> at
> org.apache.flink.runtime.operators.shipping.OutputCollector.collect(OutputCollector.java:65)
> at
> org.apache.flink.runtime.operators.util.metrics.CountingCollector.collect(CountingCollector.java:35)
> at
> org.apache.beam.runners.flink.translation.functions.FlinkExecutableStagePruningFunction.flatMap(FlinkExecutableStagePruningFunction.java:42)
> at
> org.apache.beam.runners.flink.translation.functions.FlinkExecutableStagePruningFunction.flatMap(FlinkExecutableStagePruningFunction.java:26)
> at
> org.apache.flink.runtime.operators.chaining.ChainedFlatMapDriver.collect(ChainedFlatMapDriver.java:80)
> at
> org.apache.flink.runtime.operators.util.metrics.CountingCollector.collect(CountingCollector.java:35)
> at
> org.apache.beam.runners.flink.translation.functions.FlinkExecutableStageFunction$MyDataReceiver.accept(FlinkExecutableStageFunction.java:230)
> - locked <0xf6a60bd0> (a java.lang.Object)
> at
> org.apache.beam.sdk.fn.data.BeamFnDataInboundObserver.accept(BeamFnDataInboundObserver.java:81)
> at
> org.apache.beam.sdk.fn.data.BeamFnDataInboundObserver.accept(BeamFnDataInboundObserver.java:32)
> at
> org.apache.beam.sdk.fn.data.BeamFnDataGrpcMultiplexer$InboundObserver.onNext(BeamFnDataGrpcMultiplexer.java:139)
> at
>
[jira] [Commented] (FLINK-10672) Task stuck while writing output to flink
[
https://issues.apache.org/jira/browse/FLINK-10672?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=16940226#comment-16940226
]
Yun Gao commented on FLINK-10672:
-
Hi [~mxm], very thanks for the explanation. Besides, is it possible to also
provide the topology of the underlying Flink Job and the status of each job
vertex when the job get stuck? I think with these information, we may be able
to analyze if there are cyclic dependency among tasks.
> Task stuck while writing output to flink
>
>
> Key: FLINK-10672
> URL: https://issues.apache.org/jira/browse/FLINK-10672
> Project: Flink
> Issue Type: Bug
> Components: Runtime / Coordination
>Affects Versions: 1.5.4
> Environment: OS: Debuan rodente 4.17
> Flink version: 1.5.4
> ||Key||Value||
> |jobmanager.heap.mb|1024|
> |jobmanager.rpc.address|localhost|
> |jobmanager.rpc.port|6123|
> |metrics.reporter.jmx.class|org.apache.flink.metrics.jmx.JMXReporter|
> |metrics.reporter.jmx.port|9250-9260|
> |metrics.reporters|jmx|
> |parallelism.default|1|
> |rest.port|8081|
> |taskmanager.heap.mb|1024|
> |taskmanager.numberOfTaskSlots|1|
> |web.tmpdir|/tmp/flink-web-bdb73d6c-5b9e-47b5-9ebf-eed0a7c82c26|
>
> h1. Overview
> ||Data Port||All Slots||Free Slots||CPU Cores||Physical Memory||JVM Heap
> Size||Flink Managed Memory||
> |43501|1|0|12|62.9 GB|922 MB|642 MB|
> h1. Memory
> h2. JVM (Heap/Non-Heap)
> ||Type||Committed||Used||Maximum||
> |Heap|922 MB|575 MB|922 MB|
> |Non-Heap|68.8 MB|64.3 MB|-1 B|
> |Total|991 MB|639 MB|922 MB|
> h2. Outside JVM
> ||Type||Count||Used||Capacity||
> |Direct|3,292|105 MB|105 MB|
> |Mapped|0|0 B|0 B|
> h1. Network
> h2. Memory Segments
> ||Type||Count||
> |Available|3,194|
> |Total|3,278|
> h1. Garbage Collection
> ||Collector||Count||Time||
> |G1_Young_Generation|13|336|
> |G1_Old_Generation|1|21|
>Reporter: Ankur Goenka
>Assignee: Yun Gao
>Priority: Major
> Labels: beam
> Attachments: 1uruvakHxBu.png, 3aDKQ24WvKk.png, Po89UGDn58V.png,
> jmx_dump.json, jmx_dump_detailed.json, jstack_129827.log, jstack_163822.log,
> jstack_66985.log
>
>
> I am running a fairly complex pipleline with 200+ task.
> The pipeline works fine with small data (order of 10kb input) but gets stuck
> with a slightly larger data (300kb input).
>
> The task gets stuck while writing the output toFlink, more specifically it
> gets stuck while requesting memory segment in local buffer pool. The Task
> manager UI shows that it has enough memory and memory segments to work with.
> The relevant stack trace is
> {quote}"grpc-default-executor-0" #138 daemon prio=5 os_prio=0
> tid=0x7fedb0163800 nid=0x30b7f in Object.wait() [0x7fedb4f9]
> java.lang.Thread.State: TIMED_WAITING (on object monitor)
> at (C/C++) 0x7fef201c7dae (Unknown Source)
> at (C/C++) 0x7fef1f2aea07 (Unknown Source)
> at (C/C++) 0x7fef1f241cd3 (Unknown Source)
> at java.lang.Object.wait(Native Method)
> - waiting on <0xf6d56450> (a java.util.ArrayDeque)
> at
> org.apache.flink.runtime.io.network.buffer.LocalBufferPool.requestMemorySegment(LocalBufferPool.java:247)
> - locked <0xf6d56450> (a java.util.ArrayDeque)
> at
> org.apache.flink.runtime.io.network.buffer.LocalBufferPool.requestBufferBuilderBlocking(LocalBufferPool.java:204)
> at
> org.apache.flink.runtime.io.network.api.writer.RecordWriter.requestNewBufferBuilder(RecordWriter.java:213)
> at
> org.apache.flink.runtime.io.network.api.writer.RecordWriter.sendToTarget(RecordWriter.java:144)
> at
> org.apache.flink.runtime.io.network.api.writer.RecordWriter.emit(RecordWriter.java:107)
> at
> org.apache.flink.runtime.operators.shipping.OutputCollector.collect(OutputCollector.java:65)
> at
> org.apache.flink.runtime.operators.util.metrics.CountingCollector.collect(CountingCollector.java:35)
> at
> org.apache.beam.runners.flink.translation.functions.FlinkExecutableStagePruningFunction.flatMap(FlinkExecutableStagePruningFunction.java:42)
> at
> org.apache.beam.runners.flink.translation.functions.FlinkExecutableStagePruningFunction.flatMap(FlinkExecutableStagePruningFunction.java:26)
> at
> org.apache.flink.runtime.operators.chaining.ChainedFlatMapDriver.collect(ChainedFlatMapDriver.java:80)
> at
> org.apache.flink.runtime.operators.util.metrics.CountingCollector.collect(CountingCollector.java:35)
> at
> org.apache.beam.runners.flink.translation.functions.FlinkExecutableStageFunction$MyDataReceiver.accept(FlinkExecutableStageFunction.java:230)
> - locked <0xf6a60bd0> (a java.lang.Object)
> at
> org.apache.beam.sdk.fn.data.BeamFnDataInboundObserver.accept(BeamFnDataInboundObserver.java:81)
> at
> org.apache.beam.sdk.fn.data.BeamFnDataInboundObserver.accept(BeamFnDataInboundObserver.java:32)
> at
>
[jira] [Commented] (FLINK-10672) Task stuck while writing output to flink
[
https://issues.apache.org/jira/browse/FLINK-10672?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=16939220#comment-16939220
]
Maximilian Michels commented on FLINK-10672:
Yes, it is one possible explanation that the tasks get stuck reading elements
from GRPC. The GRPC connections are used to transfer data and control execution
between Flink and the SDK Harness (where language-specific UDF code runs).
> Task stuck while writing output to flink
>
>
> Key: FLINK-10672
> URL: https://issues.apache.org/jira/browse/FLINK-10672
> Project: Flink
> Issue Type: Bug
> Components: Runtime / Coordination
>Affects Versions: 1.5.4
> Environment: OS: Debuan rodente 4.17
> Flink version: 1.5.4
> ||Key||Value||
> |jobmanager.heap.mb|1024|
> |jobmanager.rpc.address|localhost|
> |jobmanager.rpc.port|6123|
> |metrics.reporter.jmx.class|org.apache.flink.metrics.jmx.JMXReporter|
> |metrics.reporter.jmx.port|9250-9260|
> |metrics.reporters|jmx|
> |parallelism.default|1|
> |rest.port|8081|
> |taskmanager.heap.mb|1024|
> |taskmanager.numberOfTaskSlots|1|
> |web.tmpdir|/tmp/flink-web-bdb73d6c-5b9e-47b5-9ebf-eed0a7c82c26|
>
> h1. Overview
> ||Data Port||All Slots||Free Slots||CPU Cores||Physical Memory||JVM Heap
> Size||Flink Managed Memory||
> |43501|1|0|12|62.9 GB|922 MB|642 MB|
> h1. Memory
> h2. JVM (Heap/Non-Heap)
> ||Type||Committed||Used||Maximum||
> |Heap|922 MB|575 MB|922 MB|
> |Non-Heap|68.8 MB|64.3 MB|-1 B|
> |Total|991 MB|639 MB|922 MB|
> h2. Outside JVM
> ||Type||Count||Used||Capacity||
> |Direct|3,292|105 MB|105 MB|
> |Mapped|0|0 B|0 B|
> h1. Network
> h2. Memory Segments
> ||Type||Count||
> |Available|3,194|
> |Total|3,278|
> h1. Garbage Collection
> ||Collector||Count||Time||
> |G1_Young_Generation|13|336|
> |G1_Old_Generation|1|21|
>Reporter: Ankur Goenka
>Assignee: Yun Gao
>Priority: Major
> Labels: beam
> Attachments: 1uruvakHxBu.png, 3aDKQ24WvKk.png, Po89UGDn58V.png,
> jmx_dump.json, jmx_dump_detailed.json, jstack_129827.log, jstack_163822.log,
> jstack_66985.log
>
>
> I am running a fairly complex pipleline with 200+ task.
> The pipeline works fine with small data (order of 10kb input) but gets stuck
> with a slightly larger data (300kb input).
>
> The task gets stuck while writing the output toFlink, more specifically it
> gets stuck while requesting memory segment in local buffer pool. The Task
> manager UI shows that it has enough memory and memory segments to work with.
> The relevant stack trace is
> {quote}"grpc-default-executor-0" #138 daemon prio=5 os_prio=0
> tid=0x7fedb0163800 nid=0x30b7f in Object.wait() [0x7fedb4f9]
> java.lang.Thread.State: TIMED_WAITING (on object monitor)
> at (C/C++) 0x7fef201c7dae (Unknown Source)
> at (C/C++) 0x7fef1f2aea07 (Unknown Source)
> at (C/C++) 0x7fef1f241cd3 (Unknown Source)
> at java.lang.Object.wait(Native Method)
> - waiting on <0xf6d56450> (a java.util.ArrayDeque)
> at
> org.apache.flink.runtime.io.network.buffer.LocalBufferPool.requestMemorySegment(LocalBufferPool.java:247)
> - locked <0xf6d56450> (a java.util.ArrayDeque)
> at
> org.apache.flink.runtime.io.network.buffer.LocalBufferPool.requestBufferBuilderBlocking(LocalBufferPool.java:204)
> at
> org.apache.flink.runtime.io.network.api.writer.RecordWriter.requestNewBufferBuilder(RecordWriter.java:213)
> at
> org.apache.flink.runtime.io.network.api.writer.RecordWriter.sendToTarget(RecordWriter.java:144)
> at
> org.apache.flink.runtime.io.network.api.writer.RecordWriter.emit(RecordWriter.java:107)
> at
> org.apache.flink.runtime.operators.shipping.OutputCollector.collect(OutputCollector.java:65)
> at
> org.apache.flink.runtime.operators.util.metrics.CountingCollector.collect(CountingCollector.java:35)
> at
> org.apache.beam.runners.flink.translation.functions.FlinkExecutableStagePruningFunction.flatMap(FlinkExecutableStagePruningFunction.java:42)
> at
> org.apache.beam.runners.flink.translation.functions.FlinkExecutableStagePruningFunction.flatMap(FlinkExecutableStagePruningFunction.java:26)
> at
> org.apache.flink.runtime.operators.chaining.ChainedFlatMapDriver.collect(ChainedFlatMapDriver.java:80)
> at
> org.apache.flink.runtime.operators.util.metrics.CountingCollector.collect(CountingCollector.java:35)
> at
> org.apache.beam.runners.flink.translation.functions.FlinkExecutableStageFunction$MyDataReceiver.accept(FlinkExecutableStageFunction.java:230)
> - locked <0xf6a60bd0> (a java.lang.Object)
> at
> org.apache.beam.sdk.fn.data.BeamFnDataInboundObserver.accept(BeamFnDataInboundObserver.java:81)
> at
> org.apache.beam.sdk.fn.data.BeamFnDataInboundObserver.accept(BeamFnDataInboundObserver.java:32)
> at
>
[jira] [Commented] (FLINK-10672) Task stuck while writing output to flink
[
https://issues.apache.org/jira/browse/FLINK-10672?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=16939094#comment-16939094
]
Yun Gao commented on FLINK-10672:
-
Hi [~ibzib], very sorry for not very familiar with Beam, but I see from the
[stack log|[^jstack_66985.log]] that there is a GRPC server thread which is
also writing data. So what is the relationship between the tasks and the GRPC
？Are tasks also communicate with GRPC and do GRPC might cause cycle between
tasks ?
> Task stuck while writing output to flink
>
>
> Key: FLINK-10672
> URL: https://issues.apache.org/jira/browse/FLINK-10672
> Project: Flink
> Issue Type: Bug
> Components: Runtime / Coordination
>Affects Versions: 1.5.4
> Environment: OS: Debuan rodente 4.17
> Flink version: 1.5.4
> ||Key||Value||
> |jobmanager.heap.mb|1024|
> |jobmanager.rpc.address|localhost|
> |jobmanager.rpc.port|6123|
> |metrics.reporter.jmx.class|org.apache.flink.metrics.jmx.JMXReporter|
> |metrics.reporter.jmx.port|9250-9260|
> |metrics.reporters|jmx|
> |parallelism.default|1|
> |rest.port|8081|
> |taskmanager.heap.mb|1024|
> |taskmanager.numberOfTaskSlots|1|
> |web.tmpdir|/tmp/flink-web-bdb73d6c-5b9e-47b5-9ebf-eed0a7c82c26|
>
> h1. Overview
> ||Data Port||All Slots||Free Slots||CPU Cores||Physical Memory||JVM Heap
> Size||Flink Managed Memory||
> |43501|1|0|12|62.9 GB|922 MB|642 MB|
> h1. Memory
> h2. JVM (Heap/Non-Heap)
> ||Type||Committed||Used||Maximum||
> |Heap|922 MB|575 MB|922 MB|
> |Non-Heap|68.8 MB|64.3 MB|-1 B|
> |Total|991 MB|639 MB|922 MB|
> h2. Outside JVM
> ||Type||Count||Used||Capacity||
> |Direct|3,292|105 MB|105 MB|
> |Mapped|0|0 B|0 B|
> h1. Network
> h2. Memory Segments
> ||Type||Count||
> |Available|3,194|
> |Total|3,278|
> h1. Garbage Collection
> ||Collector||Count||Time||
> |G1_Young_Generation|13|336|
> |G1_Old_Generation|1|21|
>Reporter: Ankur Goenka
>Assignee: Yun Gao
>Priority: Major
> Labels: beam
> Attachments: 1uruvakHxBu.png, 3aDKQ24WvKk.png, Po89UGDn58V.png,
> jmx_dump.json, jmx_dump_detailed.json, jstack_129827.log, jstack_163822.log,
> jstack_66985.log
>
>
> I am running a fairly complex pipleline with 200+ task.
> The pipeline works fine with small data (order of 10kb input) but gets stuck
> with a slightly larger data (300kb input).
>
> The task gets stuck while writing the output toFlink, more specifically it
> gets stuck while requesting memory segment in local buffer pool. The Task
> manager UI shows that it has enough memory and memory segments to work with.
> The relevant stack trace is
> {quote}"grpc-default-executor-0" #138 daemon prio=5 os_prio=0
> tid=0x7fedb0163800 nid=0x30b7f in Object.wait() [0x7fedb4f9]
> java.lang.Thread.State: TIMED_WAITING (on object monitor)
> at (C/C++) 0x7fef201c7dae (Unknown Source)
> at (C/C++) 0x7fef1f2aea07 (Unknown Source)
> at (C/C++) 0x7fef1f241cd3 (Unknown Source)
> at java.lang.Object.wait(Native Method)
> - waiting on <0xf6d56450> (a java.util.ArrayDeque)
> at
> org.apache.flink.runtime.io.network.buffer.LocalBufferPool.requestMemorySegment(LocalBufferPool.java:247)
> - locked <0xf6d56450> (a java.util.ArrayDeque)
> at
> org.apache.flink.runtime.io.network.buffer.LocalBufferPool.requestBufferBuilderBlocking(LocalBufferPool.java:204)
> at
> org.apache.flink.runtime.io.network.api.writer.RecordWriter.requestNewBufferBuilder(RecordWriter.java:213)
> at
> org.apache.flink.runtime.io.network.api.writer.RecordWriter.sendToTarget(RecordWriter.java:144)
> at
> org.apache.flink.runtime.io.network.api.writer.RecordWriter.emit(RecordWriter.java:107)
> at
> org.apache.flink.runtime.operators.shipping.OutputCollector.collect(OutputCollector.java:65)
> at
> org.apache.flink.runtime.operators.util.metrics.CountingCollector.collect(CountingCollector.java:35)
> at
> org.apache.beam.runners.flink.translation.functions.FlinkExecutableStagePruningFunction.flatMap(FlinkExecutableStagePruningFunction.java:42)
> at
> org.apache.beam.runners.flink.translation.functions.FlinkExecutableStagePruningFunction.flatMap(FlinkExecutableStagePruningFunction.java:26)
> at
> org.apache.flink.runtime.operators.chaining.ChainedFlatMapDriver.collect(ChainedFlatMapDriver.java:80)
> at
> org.apache.flink.runtime.operators.util.metrics.CountingCollector.collect(CountingCollector.java:35)
> at
> org.apache.beam.runners.flink.translation.functions.FlinkExecutableStageFunction$MyDataReceiver.accept(FlinkExecutableStageFunction.java:230)
> - locked <0xf6a60bd0> (a java.lang.Object)
> at
> org.apache.beam.sdk.fn.data.BeamFnDataInboundObserver.accept(BeamFnDataInboundObserver.java:81)
> at
> org.apache.beam.sdk.fn.data.BeamFnDataInboundObserver.accept(BeamFnDataInboundObserver.java:32)
> at
>
[jira] [Commented] (FLINK-10672) Task stuck while writing output to flink
[
https://issues.apache.org/jira/browse/FLINK-10672?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=16938443#comment-16938443
]
zhijiang commented on FLINK-10672:
--
I think [~gaoyunhaii] would continue helping trace this issue .
> Task stuck while writing output to flink
>
>
> Key: FLINK-10672
> URL: https://issues.apache.org/jira/browse/FLINK-10672
> Project: Flink
> Issue Type: Bug
> Components: Runtime / Coordination
>Affects Versions: 1.5.4
> Environment: OS: Debuan rodente 4.17
> Flink version: 1.5.4
> ||Key||Value||
> |jobmanager.heap.mb|1024|
> |jobmanager.rpc.address|localhost|
> |jobmanager.rpc.port|6123|
> |metrics.reporter.jmx.class|org.apache.flink.metrics.jmx.JMXReporter|
> |metrics.reporter.jmx.port|9250-9260|
> |metrics.reporters|jmx|
> |parallelism.default|1|
> |rest.port|8081|
> |taskmanager.heap.mb|1024|
> |taskmanager.numberOfTaskSlots|1|
> |web.tmpdir|/tmp/flink-web-bdb73d6c-5b9e-47b5-9ebf-eed0a7c82c26|
>
> h1. Overview
> ||Data Port||All Slots||Free Slots||CPU Cores||Physical Memory||JVM Heap
> Size||Flink Managed Memory||
> |43501|1|0|12|62.9 GB|922 MB|642 MB|
> h1. Memory
> h2. JVM (Heap/Non-Heap)
> ||Type||Committed||Used||Maximum||
> |Heap|922 MB|575 MB|922 MB|
> |Non-Heap|68.8 MB|64.3 MB|-1 B|
> |Total|991 MB|639 MB|922 MB|
> h2. Outside JVM
> ||Type||Count||Used||Capacity||
> |Direct|3,292|105 MB|105 MB|
> |Mapped|0|0 B|0 B|
> h1. Network
> h2. Memory Segments
> ||Type||Count||
> |Available|3,194|
> |Total|3,278|
> h1. Garbage Collection
> ||Collector||Count||Time||
> |G1_Young_Generation|13|336|
> |G1_Old_Generation|1|21|
>Reporter: Ankur Goenka
>Assignee: Yun Gao
>Priority: Major
> Labels: beam
> Attachments: 1uruvakHxBu.png, 3aDKQ24WvKk.png, Po89UGDn58V.png,
> jmx_dump.json, jmx_dump_detailed.json, jstack_129827.log, jstack_163822.log,
> jstack_66985.log
>
>
> I am running a fairly complex pipleline with 200+ task.
> The pipeline works fine with small data (order of 10kb input) but gets stuck
> with a slightly larger data (300kb input).
>
> The task gets stuck while writing the output toFlink, more specifically it
> gets stuck while requesting memory segment in local buffer pool. The Task
> manager UI shows that it has enough memory and memory segments to work with.
> The relevant stack trace is
> {quote}"grpc-default-executor-0" #138 daemon prio=5 os_prio=0
> tid=0x7fedb0163800 nid=0x30b7f in Object.wait() [0x7fedb4f9]
> java.lang.Thread.State: TIMED_WAITING (on object monitor)
> at (C/C++) 0x7fef201c7dae (Unknown Source)
> at (C/C++) 0x7fef1f2aea07 (Unknown Source)
> at (C/C++) 0x7fef1f241cd3 (Unknown Source)
> at java.lang.Object.wait(Native Method)
> - waiting on <0xf6d56450> (a java.util.ArrayDeque)
> at
> org.apache.flink.runtime.io.network.buffer.LocalBufferPool.requestMemorySegment(LocalBufferPool.java:247)
> - locked <0xf6d56450> (a java.util.ArrayDeque)
> at
> org.apache.flink.runtime.io.network.buffer.LocalBufferPool.requestBufferBuilderBlocking(LocalBufferPool.java:204)
> at
> org.apache.flink.runtime.io.network.api.writer.RecordWriter.requestNewBufferBuilder(RecordWriter.java:213)
> at
> org.apache.flink.runtime.io.network.api.writer.RecordWriter.sendToTarget(RecordWriter.java:144)
> at
> org.apache.flink.runtime.io.network.api.writer.RecordWriter.emit(RecordWriter.java:107)
> at
> org.apache.flink.runtime.operators.shipping.OutputCollector.collect(OutputCollector.java:65)
> at
> org.apache.flink.runtime.operators.util.metrics.CountingCollector.collect(CountingCollector.java:35)
> at
> org.apache.beam.runners.flink.translation.functions.FlinkExecutableStagePruningFunction.flatMap(FlinkExecutableStagePruningFunction.java:42)
> at
> org.apache.beam.runners.flink.translation.functions.FlinkExecutableStagePruningFunction.flatMap(FlinkExecutableStagePruningFunction.java:26)
> at
> org.apache.flink.runtime.operators.chaining.ChainedFlatMapDriver.collect(ChainedFlatMapDriver.java:80)
> at
> org.apache.flink.runtime.operators.util.metrics.CountingCollector.collect(CountingCollector.java:35)
> at
> org.apache.beam.runners.flink.translation.functions.FlinkExecutableStageFunction$MyDataReceiver.accept(FlinkExecutableStageFunction.java:230)
> - locked <0xf6a60bd0> (a java.lang.Object)
> at
> org.apache.beam.sdk.fn.data.BeamFnDataInboundObserver.accept(BeamFnDataInboundObserver.java:81)
> at
> org.apache.beam.sdk.fn.data.BeamFnDataInboundObserver.accept(BeamFnDataInboundObserver.java:32)
> at
> org.apache.beam.sdk.fn.data.BeamFnDataGrpcMultiplexer$InboundObserver.onNext(BeamFnDataGrpcMultiplexer.java:139)
> at
> org.apache.beam.sdk.fn.data.BeamFnDataGrpcMultiplexer$InboundObserver.onNext(BeamFnDataGrpcMultiplexer.java:125)
> at
>
[jira] [Commented] (FLINK-10672) Task stuck while writing output to flink
[
https://issues.apache.org/jira/browse/FLINK-10672?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=16938147#comment-16938147
]
Kyle Weaver commented on FLINK-10672:
-
[~zjwang] I have tested the pipeline on Flink 1.8.1 and Flink 1.9.0 and the
issue is still present (even with BATCH_FORCED). Strangely, it didn't happen
when I had parallelism set to 11, but did happen when parallelism was 12.
However, it is solved if I increase the taskManager memory, as Ankur indicated
earlier.
> Task stuck while writing output to flink
>
>
> Key: FLINK-10672
> URL: https://issues.apache.org/jira/browse/FLINK-10672
> Project: Flink
> Issue Type: Bug
> Components: Runtime / Coordination
>Affects Versions: 1.5.4
> Environment: OS: Debuan rodente 4.17
> Flink version: 1.5.4
> ||Key||Value||
> |jobmanager.heap.mb|1024|
> |jobmanager.rpc.address|localhost|
> |jobmanager.rpc.port|6123|
> |metrics.reporter.jmx.class|org.apache.flink.metrics.jmx.JMXReporter|
> |metrics.reporter.jmx.port|9250-9260|
> |metrics.reporters|jmx|
> |parallelism.default|1|
> |rest.port|8081|
> |taskmanager.heap.mb|1024|
> |taskmanager.numberOfTaskSlots|1|
> |web.tmpdir|/tmp/flink-web-bdb73d6c-5b9e-47b5-9ebf-eed0a7c82c26|
>
> h1. Overview
> ||Data Port||All Slots||Free Slots||CPU Cores||Physical Memory||JVM Heap
> Size||Flink Managed Memory||
> |43501|1|0|12|62.9 GB|922 MB|642 MB|
> h1. Memory
> h2. JVM (Heap/Non-Heap)
> ||Type||Committed||Used||Maximum||
> |Heap|922 MB|575 MB|922 MB|
> |Non-Heap|68.8 MB|64.3 MB|-1 B|
> |Total|991 MB|639 MB|922 MB|
> h2. Outside JVM
> ||Type||Count||Used||Capacity||
> |Direct|3,292|105 MB|105 MB|
> |Mapped|0|0 B|0 B|
> h1. Network
> h2. Memory Segments
> ||Type||Count||
> |Available|3,194|
> |Total|3,278|
> h1. Garbage Collection
> ||Collector||Count||Time||
> |G1_Young_Generation|13|336|
> |G1_Old_Generation|1|21|
>Reporter: Ankur Goenka
>Priority: Major
> Labels: beam
> Attachments: 1uruvakHxBu.png, 3aDKQ24WvKk.png, Po89UGDn58V.png,
> jmx_dump.json, jmx_dump_detailed.json, jstack_129827.log, jstack_163822.log,
> jstack_66985.log
>
>
> I am running a fairly complex pipleline with 200+ task.
> The pipeline works fine with small data (order of 10kb input) but gets stuck
> with a slightly larger data (300kb input).
>
> The task gets stuck while writing the output toFlink, more specifically it
> gets stuck while requesting memory segment in local buffer pool. The Task
> manager UI shows that it has enough memory and memory segments to work with.
> The relevant stack trace is
> {quote}"grpc-default-executor-0" #138 daemon prio=5 os_prio=0
> tid=0x7fedb0163800 nid=0x30b7f in Object.wait() [0x7fedb4f9]
> java.lang.Thread.State: TIMED_WAITING (on object monitor)
> at (C/C++) 0x7fef201c7dae (Unknown Source)
> at (C/C++) 0x7fef1f2aea07 (Unknown Source)
> at (C/C++) 0x7fef1f241cd3 (Unknown Source)
> at java.lang.Object.wait(Native Method)
> - waiting on <0xf6d56450> (a java.util.ArrayDeque)
> at
> org.apache.flink.runtime.io.network.buffer.LocalBufferPool.requestMemorySegment(LocalBufferPool.java:247)
> - locked <0xf6d56450> (a java.util.ArrayDeque)
> at
> org.apache.flink.runtime.io.network.buffer.LocalBufferPool.requestBufferBuilderBlocking(LocalBufferPool.java:204)
> at
> org.apache.flink.runtime.io.network.api.writer.RecordWriter.requestNewBufferBuilder(RecordWriter.java:213)
> at
> org.apache.flink.runtime.io.network.api.writer.RecordWriter.sendToTarget(RecordWriter.java:144)
> at
> org.apache.flink.runtime.io.network.api.writer.RecordWriter.emit(RecordWriter.java:107)
> at
> org.apache.flink.runtime.operators.shipping.OutputCollector.collect(OutputCollector.java:65)
> at
> org.apache.flink.runtime.operators.util.metrics.CountingCollector.collect(CountingCollector.java:35)
> at
> org.apache.beam.runners.flink.translation.functions.FlinkExecutableStagePruningFunction.flatMap(FlinkExecutableStagePruningFunction.java:42)
> at
> org.apache.beam.runners.flink.translation.functions.FlinkExecutableStagePruningFunction.flatMap(FlinkExecutableStagePruningFunction.java:26)
> at
> org.apache.flink.runtime.operators.chaining.ChainedFlatMapDriver.collect(ChainedFlatMapDriver.java:80)
> at
> org.apache.flink.runtime.operators.util.metrics.CountingCollector.collect(CountingCollector.java:35)
> at
> org.apache.beam.runners.flink.translation.functions.FlinkExecutableStageFunction$MyDataReceiver.accept(FlinkExecutableStageFunction.java:230)
> - locked <0xf6a60bd0> (a java.lang.Object)
> at
> org.apache.beam.sdk.fn.data.BeamFnDataInboundObserver.accept(BeamFnDataInboundObserver.java:81)
> at
> org.apache.beam.sdk.fn.data.BeamFnDataInboundObserver.accept(BeamFnDataInboundObserver.java:32)
> at
>
[jira] [Commented] (FLINK-10672) Task stuck while writing output to flink
[
https://issues.apache.org/jira/browse/FLINK-10672?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=16897491#comment-16897491
]
zhijiang commented on FLINK-10672:
--
Could you also test the release-1.8.1 or current master branch whether it could
happen or not. We ever fixed some deadlock issues for both pipelined and
blocking partitions, which could also cause the similar problems as you
mentioned.
> Task stuck while writing output to flink
>
>
> Key: FLINK-10672
> URL: https://issues.apache.org/jira/browse/FLINK-10672
> Project: Flink
> Issue Type: Bug
> Components: Runtime / Coordination
>Affects Versions: 1.5.4
> Environment: OS: Debuan rodente 4.17
> Flink version: 1.5.4
> ||Key||Value||
> |jobmanager.heap.mb|1024|
> |jobmanager.rpc.address|localhost|
> |jobmanager.rpc.port|6123|
> |metrics.reporter.jmx.class|org.apache.flink.metrics.jmx.JMXReporter|
> |metrics.reporter.jmx.port|9250-9260|
> |metrics.reporters|jmx|
> |parallelism.default|1|
> |rest.port|8081|
> |taskmanager.heap.mb|1024|
> |taskmanager.numberOfTaskSlots|1|
> |web.tmpdir|/tmp/flink-web-bdb73d6c-5b9e-47b5-9ebf-eed0a7c82c26|
>
> h1. Overview
> ||Data Port||All Slots||Free Slots||CPU Cores||Physical Memory||JVM Heap
> Size||Flink Managed Memory||
> |43501|1|0|12|62.9 GB|922 MB|642 MB|
> h1. Memory
> h2. JVM (Heap/Non-Heap)
> ||Type||Committed||Used||Maximum||
> |Heap|922 MB|575 MB|922 MB|
> |Non-Heap|68.8 MB|64.3 MB|-1 B|
> |Total|991 MB|639 MB|922 MB|
> h2. Outside JVM
> ||Type||Count||Used||Capacity||
> |Direct|3,292|105 MB|105 MB|
> |Mapped|0|0 B|0 B|
> h1. Network
> h2. Memory Segments
> ||Type||Count||
> |Available|3,194|
> |Total|3,278|
> h1. Garbage Collection
> ||Collector||Count||Time||
> |G1_Young_Generation|13|336|
> |G1_Old_Generation|1|21|
>Reporter: Ankur Goenka
>Priority: Major
> Labels: beam
> Attachments: 1uruvakHxBu.png, 3aDKQ24WvKk.png, Po89UGDn58V.png,
> jmx_dump.json, jmx_dump_detailed.json, jstack_129827.log, jstack_163822.log,
> jstack_66985.log
>
>
> I am running a fairly complex pipleline with 200+ task.
> The pipeline works fine with small data (order of 10kb input) but gets stuck
> with a slightly larger data (300kb input).
>
> The task gets stuck while writing the output toFlink, more specifically it
> gets stuck while requesting memory segment in local buffer pool. The Task
> manager UI shows that it has enough memory and memory segments to work with.
> The relevant stack trace is
> {quote}"grpc-default-executor-0" #138 daemon prio=5 os_prio=0
> tid=0x7fedb0163800 nid=0x30b7f in Object.wait() [0x7fedb4f9]
> java.lang.Thread.State: TIMED_WAITING (on object monitor)
> at (C/C++) 0x7fef201c7dae (Unknown Source)
> at (C/C++) 0x7fef1f2aea07 (Unknown Source)
> at (C/C++) 0x7fef1f241cd3 (Unknown Source)
> at java.lang.Object.wait(Native Method)
> - waiting on <0xf6d56450> (a java.util.ArrayDeque)
> at
> org.apache.flink.runtime.io.network.buffer.LocalBufferPool.requestMemorySegment(LocalBufferPool.java:247)
> - locked <0xf6d56450> (a java.util.ArrayDeque)
> at
> org.apache.flink.runtime.io.network.buffer.LocalBufferPool.requestBufferBuilderBlocking(LocalBufferPool.java:204)
> at
> org.apache.flink.runtime.io.network.api.writer.RecordWriter.requestNewBufferBuilder(RecordWriter.java:213)
> at
> org.apache.flink.runtime.io.network.api.writer.RecordWriter.sendToTarget(RecordWriter.java:144)
> at
> org.apache.flink.runtime.io.network.api.writer.RecordWriter.emit(RecordWriter.java:107)
> at
> org.apache.flink.runtime.operators.shipping.OutputCollector.collect(OutputCollector.java:65)
> at
> org.apache.flink.runtime.operators.util.metrics.CountingCollector.collect(CountingCollector.java:35)
> at
> org.apache.beam.runners.flink.translation.functions.FlinkExecutableStagePruningFunction.flatMap(FlinkExecutableStagePruningFunction.java:42)
> at
> org.apache.beam.runners.flink.translation.functions.FlinkExecutableStagePruningFunction.flatMap(FlinkExecutableStagePruningFunction.java:26)
> at
> org.apache.flink.runtime.operators.chaining.ChainedFlatMapDriver.collect(ChainedFlatMapDriver.java:80)
> at
> org.apache.flink.runtime.operators.util.metrics.CountingCollector.collect(CountingCollector.java:35)
> at
> org.apache.beam.runners.flink.translation.functions.FlinkExecutableStageFunction$MyDataReceiver.accept(FlinkExecutableStageFunction.java:230)
> - locked <0xf6a60bd0> (a java.lang.Object)
> at
> org.apache.beam.sdk.fn.data.BeamFnDataInboundObserver.accept(BeamFnDataInboundObserver.java:81)
> at
> org.apache.beam.sdk.fn.data.BeamFnDataInboundObserver.accept(BeamFnDataInboundObserver.java:32)
> at
>
[jira] [Commented] (FLINK-10672) Task stuck while writing output to flink
[
https://issues.apache.org/jira/browse/FLINK-10672?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=16897101#comment-16897101
]
Maximilian Michels commented on FLINK-10672:
[~zjwang] From the testing that [~angoenka] did, it seemed to be more involved
then just backpressure from an upstream task; although this is how it
manifests. The attached stack traces do not provide enough evidence to track
down the problem. As [~ibzib] reported, the {{BATCH_FORCED}} execution mode did
not get rid of the problem. We want to make sure to identity the stuck task to
make sure this is not a Beam related problem.
> Task stuck while writing output to flink
>
>
> Key: FLINK-10672
> URL: https://issues.apache.org/jira/browse/FLINK-10672
> Project: Flink
> Issue Type: Bug
> Components: Runtime / Coordination
>Affects Versions: 1.5.4
> Environment: OS: Debuan rodente 4.17
> Flink version: 1.5.4
> ||Key||Value||
> |jobmanager.heap.mb|1024|
> |jobmanager.rpc.address|localhost|
> |jobmanager.rpc.port|6123|
> |metrics.reporter.jmx.class|org.apache.flink.metrics.jmx.JMXReporter|
> |metrics.reporter.jmx.port|9250-9260|
> |metrics.reporters|jmx|
> |parallelism.default|1|
> |rest.port|8081|
> |taskmanager.heap.mb|1024|
> |taskmanager.numberOfTaskSlots|1|
> |web.tmpdir|/tmp/flink-web-bdb73d6c-5b9e-47b5-9ebf-eed0a7c82c26|
>
> h1. Overview
> ||Data Port||All Slots||Free Slots||CPU Cores||Physical Memory||JVM Heap
> Size||Flink Managed Memory||
> |43501|1|0|12|62.9 GB|922 MB|642 MB|
> h1. Memory
> h2. JVM (Heap/Non-Heap)
> ||Type||Committed||Used||Maximum||
> |Heap|922 MB|575 MB|922 MB|
> |Non-Heap|68.8 MB|64.3 MB|-1 B|
> |Total|991 MB|639 MB|922 MB|
> h2. Outside JVM
> ||Type||Count||Used||Capacity||
> |Direct|3,292|105 MB|105 MB|
> |Mapped|0|0 B|0 B|
> h1. Network
> h2. Memory Segments
> ||Type||Count||
> |Available|3,194|
> |Total|3,278|
> h1. Garbage Collection
> ||Collector||Count||Time||
> |G1_Young_Generation|13|336|
> |G1_Old_Generation|1|21|
>Reporter: Ankur Goenka
>Priority: Major
> Labels: beam
> Attachments: 1uruvakHxBu.png, 3aDKQ24WvKk.png, Po89UGDn58V.png,
> jmx_dump.json, jmx_dump_detailed.json, jstack_129827.log, jstack_163822.log,
> jstack_66985.log
>
>
> I am running a fairly complex pipleline with 200+ task.
> The pipeline works fine with small data (order of 10kb input) but gets stuck
> with a slightly larger data (300kb input).
>
> The task gets stuck while writing the output toFlink, more specifically it
> gets stuck while requesting memory segment in local buffer pool. The Task
> manager UI shows that it has enough memory and memory segments to work with.
> The relevant stack trace is
> {quote}"grpc-default-executor-0" #138 daemon prio=5 os_prio=0
> tid=0x7fedb0163800 nid=0x30b7f in Object.wait() [0x7fedb4f9]
> java.lang.Thread.State: TIMED_WAITING (on object monitor)
> at (C/C++) 0x7fef201c7dae (Unknown Source)
> at (C/C++) 0x7fef1f2aea07 (Unknown Source)
> at (C/C++) 0x7fef1f241cd3 (Unknown Source)
> at java.lang.Object.wait(Native Method)
> - waiting on <0xf6d56450> (a java.util.ArrayDeque)
> at
> org.apache.flink.runtime.io.network.buffer.LocalBufferPool.requestMemorySegment(LocalBufferPool.java:247)
> - locked <0xf6d56450> (a java.util.ArrayDeque)
> at
> org.apache.flink.runtime.io.network.buffer.LocalBufferPool.requestBufferBuilderBlocking(LocalBufferPool.java:204)
> at
> org.apache.flink.runtime.io.network.api.writer.RecordWriter.requestNewBufferBuilder(RecordWriter.java:213)
> at
> org.apache.flink.runtime.io.network.api.writer.RecordWriter.sendToTarget(RecordWriter.java:144)
> at
> org.apache.flink.runtime.io.network.api.writer.RecordWriter.emit(RecordWriter.java:107)
> at
> org.apache.flink.runtime.operators.shipping.OutputCollector.collect(OutputCollector.java:65)
> at
> org.apache.flink.runtime.operators.util.metrics.CountingCollector.collect(CountingCollector.java:35)
> at
> org.apache.beam.runners.flink.translation.functions.FlinkExecutableStagePruningFunction.flatMap(FlinkExecutableStagePruningFunction.java:42)
> at
> org.apache.beam.runners.flink.translation.functions.FlinkExecutableStagePruningFunction.flatMap(FlinkExecutableStagePruningFunction.java:26)
> at
> org.apache.flink.runtime.operators.chaining.ChainedFlatMapDriver.collect(ChainedFlatMapDriver.java:80)
> at
> org.apache.flink.runtime.operators.util.metrics.CountingCollector.collect(CountingCollector.java:35)
> at
> org.apache.beam.runners.flink.translation.functions.FlinkExecutableStageFunction$MyDataReceiver.accept(FlinkExecutableStageFunction.java:230)
> - locked <0xf6a60bd0> (a java.lang.Object)
> at
> org.apache.beam.sdk.fn.data.BeamFnDataInboundObserver.accept(BeamFnDataInboundObserver.java:81)
> at
>
[jira] [Commented] (FLINK-10672) Task stuck while writing output to flink
[
https://issues.apache.org/jira/browse/FLINK-10672?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=16887642#comment-16887642
]
zhijiang commented on FLINK-10672:
--
[~ibzib] I think it is the case of back pressure. The producer is blocked in
the process of `requestMemorySegment` because its produced partition data
could not be consumed by downstream tasks. In order to trace the root issue,
the key point is to find the proper downstream task which triggers the back
pressure.
E.G. for the topology of A-->B–>C—>D, if we found vertex A is blocked by
`requestMemorySegment` long time, we can trace the state of upstream vertex in
topology. If vertex B is also blocked, we could continue tracing the upstream
until we find the vertex which is not blocked any more, assuming vertex C in
this case. Then we further check which specific parallelism task in vertex C
causes the above serious block. Such task has a feature that its inqueue buffer
size is very high and its out queue size is low even empty, and the relevant
metrics could help. If such task is found, we could further check its stack to
confirm where it is stuck.
Maybe you could get the root cause then, or you can provide further findings
for us.
> Task stuck while writing output to flink
>
>
> Key: FLINK-10672
> URL: https://issues.apache.org/jira/browse/FLINK-10672
> Project: Flink
> Issue Type: Bug
> Components: Runtime / Coordination
>Affects Versions: 1.5.4
> Environment: OS: Debuan rodente 4.17
> Flink version: 1.5.4
> ||Key||Value||
> |jobmanager.heap.mb|1024|
> |jobmanager.rpc.address|localhost|
> |jobmanager.rpc.port|6123|
> |metrics.reporter.jmx.class|org.apache.flink.metrics.jmx.JMXReporter|
> |metrics.reporter.jmx.port|9250-9260|
> |metrics.reporters|jmx|
> |parallelism.default|1|
> |rest.port|8081|
> |taskmanager.heap.mb|1024|
> |taskmanager.numberOfTaskSlots|1|
> |web.tmpdir|/tmp/flink-web-bdb73d6c-5b9e-47b5-9ebf-eed0a7c82c26|
>
> h1. Overview
> ||Data Port||All Slots||Free Slots||CPU Cores||Physical Memory||JVM Heap
> Size||Flink Managed Memory||
> |43501|1|0|12|62.9 GB|922 MB|642 MB|
> h1. Memory
> h2. JVM (Heap/Non-Heap)
> ||Type||Committed||Used||Maximum||
> |Heap|922 MB|575 MB|922 MB|
> |Non-Heap|68.8 MB|64.3 MB|-1 B|
> |Total|991 MB|639 MB|922 MB|
> h2. Outside JVM
> ||Type||Count||Used||Capacity||
> |Direct|3,292|105 MB|105 MB|
> |Mapped|0|0 B|0 B|
> h1. Network
> h2. Memory Segments
> ||Type||Count||
> |Available|3,194|
> |Total|3,278|
> h1. Garbage Collection
> ||Collector||Count||Time||
> |G1_Young_Generation|13|336|
> |G1_Old_Generation|1|21|
>Reporter: Ankur Goenka
>Priority: Major
> Labels: beam
> Attachments: 1uruvakHxBu.png, 3aDKQ24WvKk.png, Po89UGDn58V.png,
> jmx_dump.json, jmx_dump_detailed.json, jstack_129827.log, jstack_163822.log,
> jstack_66985.log
>
>
> I am running a fairly complex pipleline with 200+ task.
> The pipeline works fine with small data (order of 10kb input) but gets stuck
> with a slightly larger data (300kb input).
>
> The task gets stuck while writing the output toFlink, more specifically it
> gets stuck while requesting memory segment in local buffer pool. The Task
> manager UI shows that it has enough memory and memory segments to work with.
> The relevant stack trace is
> {quote}"grpc-default-executor-0" #138 daemon prio=5 os_prio=0
> tid=0x7fedb0163800 nid=0x30b7f in Object.wait() [0x7fedb4f9]
> java.lang.Thread.State: TIMED_WAITING (on object monitor)
> at (C/C++) 0x7fef201c7dae (Unknown Source)
> at (C/C++) 0x7fef1f2aea07 (Unknown Source)
> at (C/C++) 0x7fef1f241cd3 (Unknown Source)
> at java.lang.Object.wait(Native Method)
> - waiting on <0xf6d56450> (a java.util.ArrayDeque)
> at
> org.apache.flink.runtime.io.network.buffer.LocalBufferPool.requestMemorySegment(LocalBufferPool.java:247)
> - locked <0xf6d56450> (a java.util.ArrayDeque)
> at
> org.apache.flink.runtime.io.network.buffer.LocalBufferPool.requestBufferBuilderBlocking(LocalBufferPool.java:204)
> at
> org.apache.flink.runtime.io.network.api.writer.RecordWriter.requestNewBufferBuilder(RecordWriter.java:213)
> at
> org.apache.flink.runtime.io.network.api.writer.RecordWriter.sendToTarget(RecordWriter.java:144)
> at
> org.apache.flink.runtime.io.network.api.writer.RecordWriter.emit(RecordWriter.java:107)
> at
> org.apache.flink.runtime.operators.shipping.OutputCollector.collect(OutputCollector.java:65)
> at
> org.apache.flink.runtime.operators.util.metrics.CountingCollector.collect(CountingCollector.java:35)
> at
> org.apache.beam.runners.flink.translation.functions.FlinkExecutableStagePruningFunction.flatMap(FlinkExecutableStagePruningFunction.java:42)
> at
>
[jira] [Commented] (FLINK-10672) Task stuck while writing output to flink
[
https://issues.apache.org/jira/browse/FLINK-10672?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=16887283#comment-16887283
]
Kyle Weaver commented on FLINK-10672:
-
[~zjwang] I was on Flink 1.5.6 before. I upgraded to 1.8.1 and it looks like
this is still a problem.
> Task stuck while writing output to flink
>
>
> Key: FLINK-10672
> URL: https://issues.apache.org/jira/browse/FLINK-10672
> Project: Flink
> Issue Type: Bug
> Components: Runtime / Coordination
>Affects Versions: 1.5.4
> Environment: OS: Debuan rodente 4.17
> Flink version: 1.5.4
> ||Key||Value||
> |jobmanager.heap.mb|1024|
> |jobmanager.rpc.address|localhost|
> |jobmanager.rpc.port|6123|
> |metrics.reporter.jmx.class|org.apache.flink.metrics.jmx.JMXReporter|
> |metrics.reporter.jmx.port|9250-9260|
> |metrics.reporters|jmx|
> |parallelism.default|1|
> |rest.port|8081|
> |taskmanager.heap.mb|1024|
> |taskmanager.numberOfTaskSlots|1|
> |web.tmpdir|/tmp/flink-web-bdb73d6c-5b9e-47b5-9ebf-eed0a7c82c26|
>
> h1. Overview
> ||Data Port||All Slots||Free Slots||CPU Cores||Physical Memory||JVM Heap
> Size||Flink Managed Memory||
> |43501|1|0|12|62.9 GB|922 MB|642 MB|
> h1. Memory
> h2. JVM (Heap/Non-Heap)
> ||Type||Committed||Used||Maximum||
> |Heap|922 MB|575 MB|922 MB|
> |Non-Heap|68.8 MB|64.3 MB|-1 B|
> |Total|991 MB|639 MB|922 MB|
> h2. Outside JVM
> ||Type||Count||Used||Capacity||
> |Direct|3,292|105 MB|105 MB|
> |Mapped|0|0 B|0 B|
> h1. Network
> h2. Memory Segments
> ||Type||Count||
> |Available|3,194|
> |Total|3,278|
> h1. Garbage Collection
> ||Collector||Count||Time||
> |G1_Young_Generation|13|336|
> |G1_Old_Generation|1|21|
>Reporter: Ankur Goenka
>Priority: Major
> Labels: beam
> Attachments: 1uruvakHxBu.png, 3aDKQ24WvKk.png, Po89UGDn58V.png,
> jmx_dump.json, jmx_dump_detailed.json, jstack_129827.log, jstack_163822.log,
> jstack_66985.log
>
>
> I am running a fairly complex pipleline with 200+ task.
> The pipeline works fine with small data (order of 10kb input) but gets stuck
> with a slightly larger data (300kb input).
>
> The task gets stuck while writing the output toFlink, more specifically it
> gets stuck while requesting memory segment in local buffer pool. The Task
> manager UI shows that it has enough memory and memory segments to work with.
> The relevant stack trace is
> {quote}"grpc-default-executor-0" #138 daemon prio=5 os_prio=0
> tid=0x7fedb0163800 nid=0x30b7f in Object.wait() [0x7fedb4f9]
> java.lang.Thread.State: TIMED_WAITING (on object monitor)
> at (C/C++) 0x7fef201c7dae (Unknown Source)
> at (C/C++) 0x7fef1f2aea07 (Unknown Source)
> at (C/C++) 0x7fef1f241cd3 (Unknown Source)
> at java.lang.Object.wait(Native Method)
> - waiting on <0xf6d56450> (a java.util.ArrayDeque)
> at
> org.apache.flink.runtime.io.network.buffer.LocalBufferPool.requestMemorySegment(LocalBufferPool.java:247)
> - locked <0xf6d56450> (a java.util.ArrayDeque)
> at
> org.apache.flink.runtime.io.network.buffer.LocalBufferPool.requestBufferBuilderBlocking(LocalBufferPool.java:204)
> at
> org.apache.flink.runtime.io.network.api.writer.RecordWriter.requestNewBufferBuilder(RecordWriter.java:213)
> at
> org.apache.flink.runtime.io.network.api.writer.RecordWriter.sendToTarget(RecordWriter.java:144)
> at
> org.apache.flink.runtime.io.network.api.writer.RecordWriter.emit(RecordWriter.java:107)
> at
> org.apache.flink.runtime.operators.shipping.OutputCollector.collect(OutputCollector.java:65)
> at
> org.apache.flink.runtime.operators.util.metrics.CountingCollector.collect(CountingCollector.java:35)
> at
> org.apache.beam.runners.flink.translation.functions.FlinkExecutableStagePruningFunction.flatMap(FlinkExecutableStagePruningFunction.java:42)
> at
> org.apache.beam.runners.flink.translation.functions.FlinkExecutableStagePruningFunction.flatMap(FlinkExecutableStagePruningFunction.java:26)
> at
> org.apache.flink.runtime.operators.chaining.ChainedFlatMapDriver.collect(ChainedFlatMapDriver.java:80)
> at
> org.apache.flink.runtime.operators.util.metrics.CountingCollector.collect(CountingCollector.java:35)
> at
> org.apache.beam.runners.flink.translation.functions.FlinkExecutableStageFunction$MyDataReceiver.accept(FlinkExecutableStageFunction.java:230)
> - locked <0xf6a60bd0> (a java.lang.Object)
> at
> org.apache.beam.sdk.fn.data.BeamFnDataInboundObserver.accept(BeamFnDataInboundObserver.java:81)
> at
> org.apache.beam.sdk.fn.data.BeamFnDataInboundObserver.accept(BeamFnDataInboundObserver.java:32)
> at
> org.apache.beam.sdk.fn.data.BeamFnDataGrpcMultiplexer$InboundObserver.onNext(BeamFnDataGrpcMultiplexer.java:139)
> at
> org.apache.beam.sdk.fn.data.BeamFnDataGrpcMultiplexer$InboundObserver.onNext(BeamFnDataGrpcMultiplexer.java:125)
> at
[jira] [Commented] (FLINK-10672) Task stuck while writing output to flink
[
https://issues.apache.org/jira/browse/FLINK-10672?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=16887201#comment-16887201
]
zhijiang commented on FLINK-10672:
--
[~ibzib] which fix do you mean?
In addition which flink version is used to cause above issue? I have not looked
through this issue yet, and I would further check the jstack to confirm whether
it has been solved already.
> Task stuck while writing output to flink
>
>
> Key: FLINK-10672
> URL: https://issues.apache.org/jira/browse/FLINK-10672
> Project: Flink
> Issue Type: Bug
> Components: Runtime / Coordination
>Affects Versions: 1.5.4
> Environment: OS: Debuan rodente 4.17
> Flink version: 1.5.4
> ||Key||Value||
> |jobmanager.heap.mb|1024|
> |jobmanager.rpc.address|localhost|
> |jobmanager.rpc.port|6123|
> |metrics.reporter.jmx.class|org.apache.flink.metrics.jmx.JMXReporter|
> |metrics.reporter.jmx.port|9250-9260|
> |metrics.reporters|jmx|
> |parallelism.default|1|
> |rest.port|8081|
> |taskmanager.heap.mb|1024|
> |taskmanager.numberOfTaskSlots|1|
> |web.tmpdir|/tmp/flink-web-bdb73d6c-5b9e-47b5-9ebf-eed0a7c82c26|
>
> h1. Overview
> ||Data Port||All Slots||Free Slots||CPU Cores||Physical Memory||JVM Heap
> Size||Flink Managed Memory||
> |43501|1|0|12|62.9 GB|922 MB|642 MB|
> h1. Memory
> h2. JVM (Heap/Non-Heap)
> ||Type||Committed||Used||Maximum||
> |Heap|922 MB|575 MB|922 MB|
> |Non-Heap|68.8 MB|64.3 MB|-1 B|
> |Total|991 MB|639 MB|922 MB|
> h2. Outside JVM
> ||Type||Count||Used||Capacity||
> |Direct|3,292|105 MB|105 MB|
> |Mapped|0|0 B|0 B|
> h1. Network
> h2. Memory Segments
> ||Type||Count||
> |Available|3,194|
> |Total|3,278|
> h1. Garbage Collection
> ||Collector||Count||Time||
> |G1_Young_Generation|13|336|
> |G1_Old_Generation|1|21|
>Reporter: Ankur Goenka
>Priority: Major
> Labels: beam
> Attachments: 1uruvakHxBu.png, 3aDKQ24WvKk.png, Po89UGDn58V.png,
> jmx_dump.json, jmx_dump_detailed.json, jstack_129827.log, jstack_163822.log,
> jstack_66985.log
>
>
> I am running a fairly complex pipleline with 200+ task.
> The pipeline works fine with small data (order of 10kb input) but gets stuck
> with a slightly larger data (300kb input).
>
> The task gets stuck while writing the output toFlink, more specifically it
> gets stuck while requesting memory segment in local buffer pool. The Task
> manager UI shows that it has enough memory and memory segments to work with.
> The relevant stack trace is
> {quote}"grpc-default-executor-0" #138 daemon prio=5 os_prio=0
> tid=0x7fedb0163800 nid=0x30b7f in Object.wait() [0x7fedb4f9]
> java.lang.Thread.State: TIMED_WAITING (on object monitor)
> at (C/C++) 0x7fef201c7dae (Unknown Source)
> at (C/C++) 0x7fef1f2aea07 (Unknown Source)
> at (C/C++) 0x7fef1f241cd3 (Unknown Source)
> at java.lang.Object.wait(Native Method)
> - waiting on <0xf6d56450> (a java.util.ArrayDeque)
> at
> org.apache.flink.runtime.io.network.buffer.LocalBufferPool.requestMemorySegment(LocalBufferPool.java:247)
> - locked <0xf6d56450> (a java.util.ArrayDeque)
> at
> org.apache.flink.runtime.io.network.buffer.LocalBufferPool.requestBufferBuilderBlocking(LocalBufferPool.java:204)
> at
> org.apache.flink.runtime.io.network.api.writer.RecordWriter.requestNewBufferBuilder(RecordWriter.java:213)
> at
> org.apache.flink.runtime.io.network.api.writer.RecordWriter.sendToTarget(RecordWriter.java:144)
> at
> org.apache.flink.runtime.io.network.api.writer.RecordWriter.emit(RecordWriter.java:107)
> at
> org.apache.flink.runtime.operators.shipping.OutputCollector.collect(OutputCollector.java:65)
> at
> org.apache.flink.runtime.operators.util.metrics.CountingCollector.collect(CountingCollector.java:35)
> at
> org.apache.beam.runners.flink.translation.functions.FlinkExecutableStagePruningFunction.flatMap(FlinkExecutableStagePruningFunction.java:42)
> at
> org.apache.beam.runners.flink.translation.functions.FlinkExecutableStagePruningFunction.flatMap(FlinkExecutableStagePruningFunction.java:26)
> at
> org.apache.flink.runtime.operators.chaining.ChainedFlatMapDriver.collect(ChainedFlatMapDriver.java:80)
> at
> org.apache.flink.runtime.operators.util.metrics.CountingCollector.collect(CountingCollector.java:35)
> at
> org.apache.beam.runners.flink.translation.functions.FlinkExecutableStageFunction$MyDataReceiver.accept(FlinkExecutableStageFunction.java:230)
> - locked <0xf6a60bd0> (a java.lang.Object)
> at
> org.apache.beam.sdk.fn.data.BeamFnDataInboundObserver.accept(BeamFnDataInboundObserver.java:81)
> at
> org.apache.beam.sdk.fn.data.BeamFnDataInboundObserver.accept(BeamFnDataInboundObserver.java:32)
> at
> org.apache.beam.sdk.fn.data.BeamFnDataGrpcMultiplexer$InboundObserver.onNext(BeamFnDataGrpcMultiplexer.java:139)
> at
>
[jira] [Commented] (FLINK-10672) Task stuck while writing output to flink
[
https://issues.apache.org/jira/browse/FLINK-10672?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=16886547#comment-16886547
]
Kyle Weaver commented on FLINK-10672:
-
[~zjwang] [~pnowojski] I saw you had made previous fixes to this code. Do you
know why this might be happening?
> Task stuck while writing output to flink
>
>
> Key: FLINK-10672
> URL: https://issues.apache.org/jira/browse/FLINK-10672
> Project: Flink
> Issue Type: Bug
> Components: Runtime / Coordination
>Affects Versions: 1.5.4
> Environment: OS: Debuan rodente 4.17
> Flink version: 1.5.4
> ||Key||Value||
> |jobmanager.heap.mb|1024|
> |jobmanager.rpc.address|localhost|
> |jobmanager.rpc.port|6123|
> |metrics.reporter.jmx.class|org.apache.flink.metrics.jmx.JMXReporter|
> |metrics.reporter.jmx.port|9250-9260|
> |metrics.reporters|jmx|
> |parallelism.default|1|
> |rest.port|8081|
> |taskmanager.heap.mb|1024|
> |taskmanager.numberOfTaskSlots|1|
> |web.tmpdir|/tmp/flink-web-bdb73d6c-5b9e-47b5-9ebf-eed0a7c82c26|
>
> h1. Overview
> ||Data Port||All Slots||Free Slots||CPU Cores||Physical Memory||JVM Heap
> Size||Flink Managed Memory||
> |43501|1|0|12|62.9 GB|922 MB|642 MB|
> h1. Memory
> h2. JVM (Heap/Non-Heap)
> ||Type||Committed||Used||Maximum||
> |Heap|922 MB|575 MB|922 MB|
> |Non-Heap|68.8 MB|64.3 MB|-1 B|
> |Total|991 MB|639 MB|922 MB|
> h2. Outside JVM
> ||Type||Count||Used||Capacity||
> |Direct|3,292|105 MB|105 MB|
> |Mapped|0|0 B|0 B|
> h1. Network
> h2. Memory Segments
> ||Type||Count||
> |Available|3,194|
> |Total|3,278|
> h1. Garbage Collection
> ||Collector||Count||Time||
> |G1_Young_Generation|13|336|
> |G1_Old_Generation|1|21|
>Reporter: Ankur Goenka
>Priority: Major
> Labels: beam
> Attachments: 1uruvakHxBu.png, 3aDKQ24WvKk.png, Po89UGDn58V.png,
> jmx_dump.json, jmx_dump_detailed.json, jstack_129827.log, jstack_163822.log,
> jstack_66985.log
>
>
> I am running a fairly complex pipleline with 200+ task.
> The pipeline works fine with small data (order of 10kb input) but gets stuck
> with a slightly larger data (300kb input).
>
> The task gets stuck while writing the output toFlink, more specifically it
> gets stuck while requesting memory segment in local buffer pool. The Task
> manager UI shows that it has enough memory and memory segments to work with.
> The relevant stack trace is
> {quote}"grpc-default-executor-0" #138 daemon prio=5 os_prio=0
> tid=0x7fedb0163800 nid=0x30b7f in Object.wait() [0x7fedb4f9]
> java.lang.Thread.State: TIMED_WAITING (on object monitor)
> at (C/C++) 0x7fef201c7dae (Unknown Source)
> at (C/C++) 0x7fef1f2aea07 (Unknown Source)
> at (C/C++) 0x7fef1f241cd3 (Unknown Source)
> at java.lang.Object.wait(Native Method)
> - waiting on <0xf6d56450> (a java.util.ArrayDeque)
> at
> org.apache.flink.runtime.io.network.buffer.LocalBufferPool.requestMemorySegment(LocalBufferPool.java:247)
> - locked <0xf6d56450> (a java.util.ArrayDeque)
> at
> org.apache.flink.runtime.io.network.buffer.LocalBufferPool.requestBufferBuilderBlocking(LocalBufferPool.java:204)
> at
> org.apache.flink.runtime.io.network.api.writer.RecordWriter.requestNewBufferBuilder(RecordWriter.java:213)
> at
> org.apache.flink.runtime.io.network.api.writer.RecordWriter.sendToTarget(RecordWriter.java:144)
> at
> org.apache.flink.runtime.io.network.api.writer.RecordWriter.emit(RecordWriter.java:107)
> at
> org.apache.flink.runtime.operators.shipping.OutputCollector.collect(OutputCollector.java:65)
> at
> org.apache.flink.runtime.operators.util.metrics.CountingCollector.collect(CountingCollector.java:35)
> at
> org.apache.beam.runners.flink.translation.functions.FlinkExecutableStagePruningFunction.flatMap(FlinkExecutableStagePruningFunction.java:42)
> at
> org.apache.beam.runners.flink.translation.functions.FlinkExecutableStagePruningFunction.flatMap(FlinkExecutableStagePruningFunction.java:26)
> at
> org.apache.flink.runtime.operators.chaining.ChainedFlatMapDriver.collect(ChainedFlatMapDriver.java:80)
> at
> org.apache.flink.runtime.operators.util.metrics.CountingCollector.collect(CountingCollector.java:35)
> at
> org.apache.beam.runners.flink.translation.functions.FlinkExecutableStageFunction$MyDataReceiver.accept(FlinkExecutableStageFunction.java:230)
> - locked <0xf6a60bd0> (a java.lang.Object)
> at
> org.apache.beam.sdk.fn.data.BeamFnDataInboundObserver.accept(BeamFnDataInboundObserver.java:81)
> at
> org.apache.beam.sdk.fn.data.BeamFnDataInboundObserver.accept(BeamFnDataInboundObserver.java:32)
> at
> org.apache.beam.sdk.fn.data.BeamFnDataGrpcMultiplexer$InboundObserver.onNext(BeamFnDataGrpcMultiplexer.java:139)
> at
>
[jira] [Commented] (FLINK-10672) Task stuck while writing output to flink
[
https://issues.apache.org/jira/browse/FLINK-10672?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=16882538#comment-16882538
]
Kyle Weaver commented on FLINK-10672:
-
This is still an issue with the TFX preprocess Beam pipeline when the
parallelism and number of task slots are increased, even with the execution
mode set to BATCH_FORCED.
||Key||Value||
|jobmanager.heap.mb|1024|
|jobmanager.rpc.address|localhost|
|jobmanager.rpc.port|6123|
|parallelism.default|12|
|rest.port|8081|
|taskmanager.heap.mb|1024|
|taskmanager.numberOfTaskSlots|12|
|web.tmpdir|/tmp/flink-web-794ab85f-d04a-4b32-a535-eb00eeeadb98|
h1. Overview
||Data Port||All Slots||Free Slots||CPU Cores||Physical Memory||JVM Heap
Size||Flink Managed Memory||
|41199|12|0|12|62.6 GB|922 MB|639 MB|
h1. Memory
h2. JVM (Heap/Non-Heap)
||Type||Committed||Used||Maximum||
|Heap|922 MB|497 MB|922 MB|
|Non-Heap|70.5 MB|65.9 MB|-1 B|
|Total|992 MB|563 MB|922 MB|
h2. Outside JVM
||Type||Count||Used||Capacity||
|Direct|3,286|104 MB|104 MB|
|Mapped|0|0 B|0 B|
h1. Network
h2. Memory Segments
||Type||Count||
|Available|3,203|
|Total|3,278|
h1. Garbage Collection
||Collector||Count||Time||
|G1_Young_Generation|27|871|
|G1_Old_Generation|1|27|
> Task stuck while writing output to flink
>
>
> Key: FLINK-10672
> URL: https://issues.apache.org/jira/browse/FLINK-10672
> Project: Flink
> Issue Type: Bug
> Components: Runtime / Coordination
>Affects Versions: 1.5.4
> Environment: OS: Debuan rodente 4.17
> Flink version: 1.5.4
> ||Key||Value||
> |jobmanager.heap.mb|1024|
> |jobmanager.rpc.address|localhost|
> |jobmanager.rpc.port|6123|
> |metrics.reporter.jmx.class|org.apache.flink.metrics.jmx.JMXReporter|
> |metrics.reporter.jmx.port|9250-9260|
> |metrics.reporters|jmx|
> |parallelism.default|1|
> |rest.port|8081|
> |taskmanager.heap.mb|1024|
> |taskmanager.numberOfTaskSlots|1|
> |web.tmpdir|/tmp/flink-web-bdb73d6c-5b9e-47b5-9ebf-eed0a7c82c26|
>
> h1. Overview
> ||Data Port||All Slots||Free Slots||CPU Cores||Physical Memory||JVM Heap
> Size||Flink Managed Memory||
> |43501|1|0|12|62.9 GB|922 MB|642 MB|
> h1. Memory
> h2. JVM (Heap/Non-Heap)
> ||Type||Committed||Used||Maximum||
> |Heap|922 MB|575 MB|922 MB|
> |Non-Heap|68.8 MB|64.3 MB|-1 B|
> |Total|991 MB|639 MB|922 MB|
> h2. Outside JVM
> ||Type||Count||Used||Capacity||
> |Direct|3,292|105 MB|105 MB|
> |Mapped|0|0 B|0 B|
> h1. Network
> h2. Memory Segments
> ||Type||Count||
> |Available|3,194|
> |Total|3,278|
> h1. Garbage Collection
> ||Collector||Count||Time||
> |G1_Young_Generation|13|336|
> |G1_Old_Generation|1|21|
>Reporter: Ankur Goenka
>Priority: Major
> Labels: beam
> Attachments: 1uruvakHxBu.png, 3aDKQ24WvKk.png, Po89UGDn58V.png,
> jmx_dump.json, jmx_dump_detailed.json, jstack_129827.log, jstack_163822.log,
> jstack_66985.log
>
>
> I am running a fairly complex pipleline with 200+ task.
> The pipeline works fine with small data (order of 10kb input) but gets stuck
> with a slightly larger data (300kb input).
>
> The task gets stuck while writing the output toFlink, more specifically it
> gets stuck while requesting memory segment in local buffer pool. The Task
> manager UI shows that it has enough memory and memory segments to work with.
> The relevant stack trace is
> {quote}"grpc-default-executor-0" #138 daemon prio=5 os_prio=0
> tid=0x7fedb0163800 nid=0x30b7f in Object.wait() [0x7fedb4f9]
> java.lang.Thread.State: TIMED_WAITING (on object monitor)
> at (C/C++) 0x7fef201c7dae (Unknown Source)
> at (C/C++) 0x7fef1f2aea07 (Unknown Source)
> at (C/C++) 0x7fef1f241cd3 (Unknown Source)
> at java.lang.Object.wait(Native Method)
> - waiting on <0xf6d56450> (a java.util.ArrayDeque)
> at
> org.apache.flink.runtime.io.network.buffer.LocalBufferPool.requestMemorySegment(LocalBufferPool.java:247)
> - locked <0xf6d56450> (a java.util.ArrayDeque)
> at
> org.apache.flink.runtime.io.network.buffer.LocalBufferPool.requestBufferBuilderBlocking(LocalBufferPool.java:204)
> at
> org.apache.flink.runtime.io.network.api.writer.RecordWriter.requestNewBufferBuilder(RecordWriter.java:213)
> at
> org.apache.flink.runtime.io.network.api.writer.RecordWriter.sendToTarget(RecordWriter.java:144)
> at
> org.apache.flink.runtime.io.network.api.writer.RecordWriter.emit(RecordWriter.java:107)
> at
> org.apache.flink.runtime.operators.shipping.OutputCollector.collect(OutputCollector.java:65)
> at
> org.apache.flink.runtime.operators.util.metrics.CountingCollector.collect(CountingCollector.java:35)
> at
> org.apache.beam.runners.flink.translation.functions.FlinkExecutableStagePruningFunction.flatMap(FlinkExecutableStagePruningFunction.java:42)
> at
>
[jira] [Commented] (FLINK-10672) Task stuck while writing output to flink
[
https://issues.apache.org/jira/browse/FLINK-10672?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=16724901#comment-16724901
]
Maximilian Michels commented on FLINK-10672:
No, [~thw] just set the Fix Version to empty because this is not fixed in 1.5.6.
> Task stuck while writing output to flink
>
>
> Key: FLINK-10672
> URL: https://issues.apache.org/jira/browse/FLINK-10672
> Project: Flink
> Issue Type: Bug
>Affects Versions: 1.5.4
> Environment: OS: Debuan rodente 4.17
> Flink version: 1.5.4
> ||Key||Value||
> |jobmanager.heap.mb|1024|
> |jobmanager.rpc.address|localhost|
> |jobmanager.rpc.port|6123|
> |metrics.reporter.jmx.class|org.apache.flink.metrics.jmx.JMXReporter|
> |metrics.reporter.jmx.port|9250-9260|
> |metrics.reporters|jmx|
> |parallelism.default|1|
> |rest.port|8081|
> |taskmanager.heap.mb|1024|
> |taskmanager.numberOfTaskSlots|1|
> |web.tmpdir|/tmp/flink-web-bdb73d6c-5b9e-47b5-9ebf-eed0a7c82c26|
>
> h1. Overview
> ||Data Port||All Slots||Free Slots||CPU Cores||Physical Memory||JVM Heap
> Size||Flink Managed Memory||
> |43501|1|0|12|62.9 GB|922 MB|642 MB|
> h1. Memory
> h2. JVM (Heap/Non-Heap)
> ||Type||Committed||Used||Maximum||
> |Heap|922 MB|575 MB|922 MB|
> |Non-Heap|68.8 MB|64.3 MB|-1 B|
> |Total|991 MB|639 MB|922 MB|
> h2. Outside JVM
> ||Type||Count||Used||Capacity||
> |Direct|3,292|105 MB|105 MB|
> |Mapped|0|0 B|0 B|
> h1. Network
> h2. Memory Segments
> ||Type||Count||
> |Available|3,194|
> |Total|3,278|
> h1. Garbage Collection
> ||Collector||Count||Time||
> |G1_Young_Generation|13|336|
> |G1_Old_Generation|1|21|
>Reporter: Ankur Goenka
>Priority: Major
> Labels: beam
> Attachments: 1uruvakHxBu.png, 3aDKQ24WvKk.png, Po89UGDn58V.png,
> jmx_dump.json, jmx_dump_detailed.json, jstack_129827.log, jstack_163822.log,
> jstack_66985.log
>
>
> I am running a fairly complex pipleline with 200+ task.
> The pipeline works fine with small data (order of 10kb input) but gets stuck
> with a slightly larger data (300kb input).
>
> The task gets stuck while writing the output toFlink, more specifically it
> gets stuck while requesting memory segment in local buffer pool. The Task
> manager UI shows that it has enough memory and memory segments to work with.
> The relevant stack trace is
> {quote}"grpc-default-executor-0" #138 daemon prio=5 os_prio=0
> tid=0x7fedb0163800 nid=0x30b7f in Object.wait() [0x7fedb4f9]
> java.lang.Thread.State: TIMED_WAITING (on object monitor)
> at (C/C++) 0x7fef201c7dae (Unknown Source)
> at (C/C++) 0x7fef1f2aea07 (Unknown Source)
> at (C/C++) 0x7fef1f241cd3 (Unknown Source)
> at java.lang.Object.wait(Native Method)
> - waiting on <0xf6d56450> (a java.util.ArrayDeque)
> at
> org.apache.flink.runtime.io.network.buffer.LocalBufferPool.requestMemorySegment(LocalBufferPool.java:247)
> - locked <0xf6d56450> (a java.util.ArrayDeque)
> at
> org.apache.flink.runtime.io.network.buffer.LocalBufferPool.requestBufferBuilderBlocking(LocalBufferPool.java:204)
> at
> org.apache.flink.runtime.io.network.api.writer.RecordWriter.requestNewBufferBuilder(RecordWriter.java:213)
> at
> org.apache.flink.runtime.io.network.api.writer.RecordWriter.sendToTarget(RecordWriter.java:144)
> at
> org.apache.flink.runtime.io.network.api.writer.RecordWriter.emit(RecordWriter.java:107)
> at
> org.apache.flink.runtime.operators.shipping.OutputCollector.collect(OutputCollector.java:65)
> at
> org.apache.flink.runtime.operators.util.metrics.CountingCollector.collect(CountingCollector.java:35)
> at
> org.apache.beam.runners.flink.translation.functions.FlinkExecutableStagePruningFunction.flatMap(FlinkExecutableStagePruningFunction.java:42)
> at
> org.apache.beam.runners.flink.translation.functions.FlinkExecutableStagePruningFunction.flatMap(FlinkExecutableStagePruningFunction.java:26)
> at
> org.apache.flink.runtime.operators.chaining.ChainedFlatMapDriver.collect(ChainedFlatMapDriver.java:80)
> at
> org.apache.flink.runtime.operators.util.metrics.CountingCollector.collect(CountingCollector.java:35)
> at
> org.apache.beam.runners.flink.translation.functions.FlinkExecutableStageFunction$MyDataReceiver.accept(FlinkExecutableStageFunction.java:230)
> - locked <0xf6a60bd0> (a java.lang.Object)
> at
> org.apache.beam.sdk.fn.data.BeamFnDataInboundObserver.accept(BeamFnDataInboundObserver.java:81)
> at
> org.apache.beam.sdk.fn.data.BeamFnDataInboundObserver.accept(BeamFnDataInboundObserver.java:32)
> at
> org.apache.beam.sdk.fn.data.BeamFnDataGrpcMultiplexer$InboundObserver.onNext(BeamFnDataGrpcMultiplexer.java:139)
> at
> org.apache.beam.sdk.fn.data.BeamFnDataGrpcMultiplexer$InboundObserver.onNext(BeamFnDataGrpcMultiplexer.java:125)
> at
>
[jira] [Commented] (FLINK-10672) Task stuck while writing output to flink
[
https://issues.apache.org/jira/browse/FLINK-10672?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=16724576#comment-16724576
]
Ankur Goenka commented on FLINK-10672:
--
Is this fixed in 1.5.6?
> Task stuck while writing output to flink
>
>
> Key: FLINK-10672
> URL: https://issues.apache.org/jira/browse/FLINK-10672
> Project: Flink
> Issue Type: Bug
>Affects Versions: 1.5.4
> Environment: OS: Debuan rodente 4.17
> Flink version: 1.5.4
> ||Key||Value||
> |jobmanager.heap.mb|1024|
> |jobmanager.rpc.address|localhost|
> |jobmanager.rpc.port|6123|
> |metrics.reporter.jmx.class|org.apache.flink.metrics.jmx.JMXReporter|
> |metrics.reporter.jmx.port|9250-9260|
> |metrics.reporters|jmx|
> |parallelism.default|1|
> |rest.port|8081|
> |taskmanager.heap.mb|1024|
> |taskmanager.numberOfTaskSlots|1|
> |web.tmpdir|/tmp/flink-web-bdb73d6c-5b9e-47b5-9ebf-eed0a7c82c26|
>
> h1. Overview
> ||Data Port||All Slots||Free Slots||CPU Cores||Physical Memory||JVM Heap
> Size||Flink Managed Memory||
> |43501|1|0|12|62.9 GB|922 MB|642 MB|
> h1. Memory
> h2. JVM (Heap/Non-Heap)
> ||Type||Committed||Used||Maximum||
> |Heap|922 MB|575 MB|922 MB|
> |Non-Heap|68.8 MB|64.3 MB|-1 B|
> |Total|991 MB|639 MB|922 MB|
> h2. Outside JVM
> ||Type||Count||Used||Capacity||
> |Direct|3,292|105 MB|105 MB|
> |Mapped|0|0 B|0 B|
> h1. Network
> h2. Memory Segments
> ||Type||Count||
> |Available|3,194|
> |Total|3,278|
> h1. Garbage Collection
> ||Collector||Count||Time||
> |G1_Young_Generation|13|336|
> |G1_Old_Generation|1|21|
>Reporter: Ankur Goenka
>Priority: Major
> Labels: beam
> Attachments: 1uruvakHxBu.png, 3aDKQ24WvKk.png, Po89UGDn58V.png,
> jmx_dump.json, jmx_dump_detailed.json, jstack_129827.log, jstack_163822.log,
> jstack_66985.log
>
>
> I am running a fairly complex pipleline with 200+ task.
> The pipeline works fine with small data (order of 10kb input) but gets stuck
> with a slightly larger data (300kb input).
>
> The task gets stuck while writing the output toFlink, more specifically it
> gets stuck while requesting memory segment in local buffer pool. The Task
> manager UI shows that it has enough memory and memory segments to work with.
> The relevant stack trace is
> {quote}"grpc-default-executor-0" #138 daemon prio=5 os_prio=0
> tid=0x7fedb0163800 nid=0x30b7f in Object.wait() [0x7fedb4f9]
> java.lang.Thread.State: TIMED_WAITING (on object monitor)
> at (C/C++) 0x7fef201c7dae (Unknown Source)
> at (C/C++) 0x7fef1f2aea07 (Unknown Source)
> at (C/C++) 0x7fef1f241cd3 (Unknown Source)
> at java.lang.Object.wait(Native Method)
> - waiting on <0xf6d56450> (a java.util.ArrayDeque)
> at
> org.apache.flink.runtime.io.network.buffer.LocalBufferPool.requestMemorySegment(LocalBufferPool.java:247)
> - locked <0xf6d56450> (a java.util.ArrayDeque)
> at
> org.apache.flink.runtime.io.network.buffer.LocalBufferPool.requestBufferBuilderBlocking(LocalBufferPool.java:204)
> at
> org.apache.flink.runtime.io.network.api.writer.RecordWriter.requestNewBufferBuilder(RecordWriter.java:213)
> at
> org.apache.flink.runtime.io.network.api.writer.RecordWriter.sendToTarget(RecordWriter.java:144)
> at
> org.apache.flink.runtime.io.network.api.writer.RecordWriter.emit(RecordWriter.java:107)
> at
> org.apache.flink.runtime.operators.shipping.OutputCollector.collect(OutputCollector.java:65)
> at
> org.apache.flink.runtime.operators.util.metrics.CountingCollector.collect(CountingCollector.java:35)
> at
> org.apache.beam.runners.flink.translation.functions.FlinkExecutableStagePruningFunction.flatMap(FlinkExecutableStagePruningFunction.java:42)
> at
> org.apache.beam.runners.flink.translation.functions.FlinkExecutableStagePruningFunction.flatMap(FlinkExecutableStagePruningFunction.java:26)
> at
> org.apache.flink.runtime.operators.chaining.ChainedFlatMapDriver.collect(ChainedFlatMapDriver.java:80)
> at
> org.apache.flink.runtime.operators.util.metrics.CountingCollector.collect(CountingCollector.java:35)
> at
> org.apache.beam.runners.flink.translation.functions.FlinkExecutableStageFunction$MyDataReceiver.accept(FlinkExecutableStageFunction.java:230)
> - locked <0xf6a60bd0> (a java.lang.Object)
> at
> org.apache.beam.sdk.fn.data.BeamFnDataInboundObserver.accept(BeamFnDataInboundObserver.java:81)
> at
> org.apache.beam.sdk.fn.data.BeamFnDataInboundObserver.accept(BeamFnDataInboundObserver.java:32)
> at
> org.apache.beam.sdk.fn.data.BeamFnDataGrpcMultiplexer$InboundObserver.onNext(BeamFnDataGrpcMultiplexer.java:139)
> at
> org.apache.beam.sdk.fn.data.BeamFnDataGrpcMultiplexer$InboundObserver.onNext(BeamFnDataGrpcMultiplexer.java:125)
> at
>
[jira] [Commented] (FLINK-10672) Task stuck while writing output to flink
[
https://issues.apache.org/jira/browse/FLINK-10672?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=16671319#comment-16671319
]
Maximilian Michels commented on FLINK-10672:
Thanks for investigating the matter [~angoenka].
This seems like an issue with the deadlock detection code. It's not a blocker
for us anymore because we have the {{BATCH_FORCED}} workaround. We will
continue to investigate the pipelined execution issue.
The TFX pipeline makes heavy use of broadcast variables and there is a
suspicion these might not be scheduled correctly under certain conditions.
> Task stuck while writing output to flink
>
>
> Key: FLINK-10672
> URL: https://issues.apache.org/jira/browse/FLINK-10672
> Project: Flink
> Issue Type: Bug
>Affects Versions: 1.5.4
> Environment: OS: Debuan rodente 4.17
> Flink version: 1.5.4
> ||Key||Value||
> |jobmanager.heap.mb|1024|
> |jobmanager.rpc.address|localhost|
> |jobmanager.rpc.port|6123|
> |metrics.reporter.jmx.class|org.apache.flink.metrics.jmx.JMXReporter|
> |metrics.reporter.jmx.port|9250-9260|
> |metrics.reporters|jmx|
> |parallelism.default|1|
> |rest.port|8081|
> |taskmanager.heap.mb|1024|
> |taskmanager.numberOfTaskSlots|1|
> |web.tmpdir|/tmp/flink-web-bdb73d6c-5b9e-47b5-9ebf-eed0a7c82c26|
>
> h1. Overview
> ||Data Port||All Slots||Free Slots||CPU Cores||Physical Memory||JVM Heap
> Size||Flink Managed Memory||
> |43501|1|0|12|62.9 GB|922 MB|642 MB|
> h1. Memory
> h2. JVM (Heap/Non-Heap)
> ||Type||Committed||Used||Maximum||
> |Heap|922 MB|575 MB|922 MB|
> |Non-Heap|68.8 MB|64.3 MB|-1 B|
> |Total|991 MB|639 MB|922 MB|
> h2. Outside JVM
> ||Type||Count||Used||Capacity||
> |Direct|3,292|105 MB|105 MB|
> |Mapped|0|0 B|0 B|
> h1. Network
> h2. Memory Segments
> ||Type||Count||
> |Available|3,194|
> |Total|3,278|
> h1. Garbage Collection
> ||Collector||Count||Time||
> |G1_Young_Generation|13|336|
> |G1_Old_Generation|1|21|
>Reporter: Ankur Goenka
>Priority: Major
> Attachments: 1uruvakHxBu.png, 3aDKQ24WvKk.png, Po89UGDn58V.png,
> jmx_dump.json, jmx_dump_detailed.json, jstack_129827.log, jstack_163822.log,
> jstack_66985.log
>
>
> I am running a fairly complex pipleline with 200+ task.
> The pipeline works fine with small data (order of 10kb input) but gets stuck
> with a slightly larger data (300kb input).
>
> The task gets stuck while writing the output toFlink, more specifically it
> gets stuck while requesting memory segment in local buffer pool. The Task
> manager UI shows that it has enough memory and memory segments to work with.
> The relevant stack trace is
> {quote}"grpc-default-executor-0" #138 daemon prio=5 os_prio=0
> tid=0x7fedb0163800 nid=0x30b7f in Object.wait() [0x7fedb4f9]
> java.lang.Thread.State: TIMED_WAITING (on object monitor)
> at (C/C++) 0x7fef201c7dae (Unknown Source)
> at (C/C++) 0x7fef1f2aea07 (Unknown Source)
> at (C/C++) 0x7fef1f241cd3 (Unknown Source)
> at java.lang.Object.wait(Native Method)
> - waiting on <0xf6d56450> (a java.util.ArrayDeque)
> at
> org.apache.flink.runtime.io.network.buffer.LocalBufferPool.requestMemorySegment(LocalBufferPool.java:247)
> - locked <0xf6d56450> (a java.util.ArrayDeque)
> at
> org.apache.flink.runtime.io.network.buffer.LocalBufferPool.requestBufferBuilderBlocking(LocalBufferPool.java:204)
> at
> org.apache.flink.runtime.io.network.api.writer.RecordWriter.requestNewBufferBuilder(RecordWriter.java:213)
> at
> org.apache.flink.runtime.io.network.api.writer.RecordWriter.sendToTarget(RecordWriter.java:144)
> at
> org.apache.flink.runtime.io.network.api.writer.RecordWriter.emit(RecordWriter.java:107)
> at
> org.apache.flink.runtime.operators.shipping.OutputCollector.collect(OutputCollector.java:65)
> at
> org.apache.flink.runtime.operators.util.metrics.CountingCollector.collect(CountingCollector.java:35)
> at
> org.apache.beam.runners.flink.translation.functions.FlinkExecutableStagePruningFunction.flatMap(FlinkExecutableStagePruningFunction.java:42)
> at
> org.apache.beam.runners.flink.translation.functions.FlinkExecutableStagePruningFunction.flatMap(FlinkExecutableStagePruningFunction.java:26)
> at
> org.apache.flink.runtime.operators.chaining.ChainedFlatMapDriver.collect(ChainedFlatMapDriver.java:80)
> at
> org.apache.flink.runtime.operators.util.metrics.CountingCollector.collect(CountingCollector.java:35)
> at
> org.apache.beam.runners.flink.translation.functions.FlinkExecutableStageFunction$MyDataReceiver.accept(FlinkExecutableStageFunction.java:230)
> - locked <0xf6a60bd0> (a java.lang.Object)
> at
> org.apache.beam.sdk.fn.data.BeamFnDataInboundObserver.accept(BeamFnDataInboundObserver.java:81)
> at
> org.apache.beam.sdk.fn.data.BeamFnDataInboundObserver.accept(BeamFnDataInboundObserver.java:32)
> at
>
[jira] [Commented] (FLINK-10672) Task stuck while writing output to flink
[
https://issues.apache.org/jira/browse/FLINK-10672?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=16669502#comment-16669502
]
Ankur Goenka commented on FLINK-10672:
--
Investigation notes:
* Pipeline seems to be waiting on writing data. I verified back pressure in an
stuck task. The consuming task was waiting on data while producing task was
waiting to memorySegment to write the data.
* Many of the broadcasting tasks were in created state and were not scheduled.
* Using only BATCH execution mode did not help.
* Using BATCH_FORCED execution mode makes the pipeline work. This suggests
that we might not be identifying a breakPipeline correctly.
* Pipeline seems to also work if we increase the taskManager memory. This
might be because with higher memory, pipelined buffers can push all the output
to memory before starting the consuming task.
> Task stuck while writing output to flink
>
>
> Key: FLINK-10672
> URL: https://issues.apache.org/jira/browse/FLINK-10672
> Project: Flink
> Issue Type: Bug
>Affects Versions: 1.5.4
> Environment: OS: Debuan rodente 4.17
> Flink version: 1.5.4
> ||Key||Value||
> |jobmanager.heap.mb|1024|
> |jobmanager.rpc.address|localhost|
> |jobmanager.rpc.port|6123|
> |metrics.reporter.jmx.class|org.apache.flink.metrics.jmx.JMXReporter|
> |metrics.reporter.jmx.port|9250-9260|
> |metrics.reporters|jmx|
> |parallelism.default|1|
> |rest.port|8081|
> |taskmanager.heap.mb|1024|
> |taskmanager.numberOfTaskSlots|1|
> |web.tmpdir|/tmp/flink-web-bdb73d6c-5b9e-47b5-9ebf-eed0a7c82c26|
>
> h1. Overview
> ||Data Port||All Slots||Free Slots||CPU Cores||Physical Memory||JVM Heap
> Size||Flink Managed Memory||
> |43501|1|0|12|62.9 GB|922 MB|642 MB|
> h1. Memory
> h2. JVM (Heap/Non-Heap)
> ||Type||Committed||Used||Maximum||
> |Heap|922 MB|575 MB|922 MB|
> |Non-Heap|68.8 MB|64.3 MB|-1 B|
> |Total|991 MB|639 MB|922 MB|
> h2. Outside JVM
> ||Type||Count||Used||Capacity||
> |Direct|3,292|105 MB|105 MB|
> |Mapped|0|0 B|0 B|
> h1. Network
> h2. Memory Segments
> ||Type||Count||
> |Available|3,194|
> |Total|3,278|
> h1. Garbage Collection
> ||Collector||Count||Time||
> |G1_Young_Generation|13|336|
> |G1_Old_Generation|1|21|
>Reporter: Ankur Goenka
>Priority: Major
> Attachments: 1uruvakHxBu.png, 3aDKQ24WvKk.png, Po89UGDn58V.png,
> jmx_dump.json, jmx_dump_detailed.json, jstack_129827.log, jstack_163822.log,
> jstack_66985.log
>
>
> I am running a fairly complex pipleline with 200+ task.
> The pipeline works fine with small data (order of 10kb input) but gets stuck
> with a slightly larger data (300kb input).
>
> The task gets stuck while writing the output toFlink, more specifically it
> gets stuck while requesting memory segment in local buffer pool. The Task
> manager UI shows that it has enough memory and memory segments to work with.
> The relevant stack trace is
> {quote}"grpc-default-executor-0" #138 daemon prio=5 os_prio=0
> tid=0x7fedb0163800 nid=0x30b7f in Object.wait() [0x7fedb4f9]
> java.lang.Thread.State: TIMED_WAITING (on object monitor)
> at (C/C++) 0x7fef201c7dae (Unknown Source)
> at (C/C++) 0x7fef1f2aea07 (Unknown Source)
> at (C/C++) 0x7fef1f241cd3 (Unknown Source)
> at java.lang.Object.wait(Native Method)
> - waiting on <0xf6d56450> (a java.util.ArrayDeque)
> at
> org.apache.flink.runtime.io.network.buffer.LocalBufferPool.requestMemorySegment(LocalBufferPool.java:247)
> - locked <0xf6d56450> (a java.util.ArrayDeque)
> at
> org.apache.flink.runtime.io.network.buffer.LocalBufferPool.requestBufferBuilderBlocking(LocalBufferPool.java:204)
> at
> org.apache.flink.runtime.io.network.api.writer.RecordWriter.requestNewBufferBuilder(RecordWriter.java:213)
> at
> org.apache.flink.runtime.io.network.api.writer.RecordWriter.sendToTarget(RecordWriter.java:144)
> at
> org.apache.flink.runtime.io.network.api.writer.RecordWriter.emit(RecordWriter.java:107)
> at
> org.apache.flink.runtime.operators.shipping.OutputCollector.collect(OutputCollector.java:65)
> at
> org.apache.flink.runtime.operators.util.metrics.CountingCollector.collect(CountingCollector.java:35)
> at
> org.apache.beam.runners.flink.translation.functions.FlinkExecutableStagePruningFunction.flatMap(FlinkExecutableStagePruningFunction.java:42)
> at
> org.apache.beam.runners.flink.translation.functions.FlinkExecutableStagePruningFunction.flatMap(FlinkExecutableStagePruningFunction.java:26)
> at
> org.apache.flink.runtime.operators.chaining.ChainedFlatMapDriver.collect(ChainedFlatMapDriver.java:80)
> at
> org.apache.flink.runtime.operators.util.metrics.CountingCollector.collect(CountingCollector.java:35)
> at
>
[jira] [Commented] (FLINK-10672) Task stuck while writing output to flink
[
https://issues.apache.org/jira/browse/FLINK-10672?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=16662937#comment-16662937
]
Ankur Goenka commented on FLINK-10672:
--
cc: [~mxm] [~robertwb]
> Task stuck while writing output to flink
>
>
> Key: FLINK-10672
> URL: https://issues.apache.org/jira/browse/FLINK-10672
> Project: Flink
> Issue Type: Bug
>Affects Versions: 1.5.4
> Environment: OS: Debuan rodente 4.17
> Flink version: 1.5.4
> ||Key||Value||
> |jobmanager.heap.mb|1024|
> |jobmanager.rpc.address|localhost|
> |jobmanager.rpc.port|6123|
> |metrics.reporter.jmx.class|org.apache.flink.metrics.jmx.JMXReporter|
> |metrics.reporter.jmx.port|9250-9260|
> |metrics.reporters|jmx|
> |parallelism.default|1|
> |rest.port|8081|
> |taskmanager.heap.mb|1024|
> |taskmanager.numberOfTaskSlots|1|
> |web.tmpdir|/tmp/flink-web-bdb73d6c-5b9e-47b5-9ebf-eed0a7c82c26|
>
> h1. Overview
> ||Data Port||All Slots||Free Slots||CPU Cores||Physical Memory||JVM Heap
> Size||Flink Managed Memory||
> |43501|1|0|12|62.9 GB|922 MB|642 MB|
> h1. Memory
> h2. JVM (Heap/Non-Heap)
> ||Type||Committed||Used||Maximum||
> |Heap|922 MB|575 MB|922 MB|
> |Non-Heap|68.8 MB|64.3 MB|-1 B|
> |Total|991 MB|639 MB|922 MB|
> h2. Outside JVM
> ||Type||Count||Used||Capacity||
> |Direct|3,292|105 MB|105 MB|
> |Mapped|0|0 B|0 B|
> h1. Network
> h2. Memory Segments
> ||Type||Count||
> |Available|3,194|
> |Total|3,278|
> h1. Garbage Collection
> ||Collector||Count||Time||
> |G1_Young_Generation|13|336|
> |G1_Old_Generation|1|21|
>Reporter: Ankur Goenka
>Priority: Major
> Attachments: 1uruvakHxBu.png, 3aDKQ24WvKk.png, Po89UGDn58V.png,
> jmx_dump.json, jmx_dump_detailed.json, jstack_129827.log, jstack_163822.log,
> jstack_66985.log
>
>
> I am running a fairly complex pipleline with 200+ task.
> The pipeline works fine with small data (order of 10kb input) but gets stuck
> with a slightly larger data (300kb input).
>
> The task gets stuck while writing the output toFlink, more specifically it
> gets stuck while requesting memory segment in local buffer pool. The Task
> manager UI shows that it has enough memory and memory segments to work with.
> The relevant stack trace is
> {{"grpc-default-executor-0" #138 daemon prio=5 os_prio=0
> tid=0x7fedb0163800 nid=0x30b7f in Object.wait() [0x7fedb4f9]}}
> {{ java.lang.Thread.State: TIMED_WAITING (on object monitor)}}
> {{ at (C/C++) 0x7fef201c7dae (Unknown Source)}}
> {{ at (C/C++) 0x7fef1f2aea07 (Unknown Source)}}
> {{ at (C/C++) 0x7fef1f241cd3 (Unknown Source)}}
> {{ at java.lang.Object.wait(Native Method)}}
> {{ - waiting on <0xf6d56450> (a java.util.ArrayDeque)}}
> {{ at
> org.apache.flink.runtime.io.network.buffer.LocalBufferPool.requestMemorySegment(LocalBufferPool.java:247)}}
> {{ - locked <0xf6d56450> (a java.util.ArrayDeque)}}
> {{ at
> org.apache.flink.runtime.io.network.buffer.LocalBufferPool.requestBufferBuilderBlocking(LocalBufferPool.java:204)}}
> {{ at
> org.apache.flink.runtime.io.network.api.writer.RecordWriter.requestNewBufferBuilder(RecordWriter.java:213)}}
> {{ at
> org.apache.flink.runtime.io.network.api.writer.RecordWriter.sendToTarget(RecordWriter.java:144)}}
> {{ at
> org.apache.flink.runtime.io.network.api.writer.RecordWriter.emit(RecordWriter.java:107)}}
> {{ at
> org.apache.flink.runtime.operators.shipping.OutputCollector.collect(OutputCollector.java:65)}}
> {{ at
> org.apache.flink.runtime.operators.util.metrics.CountingCollector.collect(CountingCollector.java:35)}}
> {{ at
> org.apache.beam.runners.flink.translation.functions.FlinkExecutableStagePruningFunction.flatMap(FlinkExecutableStagePruningFunction.java:42)}}
> {{ at
> org.apache.beam.runners.flink.translation.functions.FlinkExecutableStagePruningFunction.flatMap(FlinkExecutableStagePruningFunction.java:26)}}
> {{ at
> org.apache.flink.runtime.operators.chaining.ChainedFlatMapDriver.collect(ChainedFlatMapDriver.java:80)}}
> {{ at
> org.apache.flink.runtime.operators.util.metrics.CountingCollector.collect(CountingCollector.java:35)}}
> {{ at
> org.apache.beam.runners.flink.translation.functions.FlinkExecutableStageFunction$MyDataReceiver.accept(FlinkExecutableStageFunction.java:230)}}
> {{ - locked <0xf6a60bd0> (a java.lang.Object)}}
> {{ at
> org.apache.beam.sdk.fn.data.BeamFnDataInboundObserver.accept(BeamFnDataInboundObserver.java:81)}}
> {{ at
> org.apache.beam.sdk.fn.data.BeamFnDataInboundObserver.accept(BeamFnDataInboundObserver.java:32)}}
> {{ at
> org.apache.beam.sdk.fn.data.BeamFnDataGrpcMultiplexer$InboundObserver.onNext(BeamFnDataGrpcMultiplexer.java:139)}}
> {{ at
> org.apache.beam.sdk.fn.data.BeamFnDataGrpcMultiplexer$InboundObserver.onNext(BeamFnDataGrpcMultiplexer.java:125)}}
> {{ at
>
