[jira] [Comment Edited] (FLINK-9693) Possible memory leak in jobmanager retaining archived checkpoints
[ https://issues.apache.org/jira/browse/FLINK-9693?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16556325#comment-16556325 ] Steven Zhen Wu edited comment on FLINK-9693 at 7/25/18 9:56 PM: One more observation. we are seeing this issue right after the jobmanager node got killed and replaced. however, it is not reproducible when I trying to kill the jobmanager when job is healthy. so still don't know what exactly conditions triggered the issue was (Author: stevenz3wu): One more observation. we are seeing this issue right after the jobmanager node got killed and replaced. however, it is not reproducible when I trying to kill the jobmanager when job is healthy > Possible memory leak in jobmanager retaining archived checkpoints > - > > Key: FLINK-9693 > URL: https://issues.apache.org/jira/browse/FLINK-9693 > Project: Flink > Issue Type: Bug > Components: JobManager, State Backends, Checkpointing >Affects Versions: 1.5.0, 1.6.0 > Environment: !image.png!!image (1).png! >Reporter: Steven Zhen Wu >Assignee: Till Rohrmann >Priority: Major > Labels: pull-request-available > Fix For: 1.4.3, 1.5.1, 1.6.0 > > Attachments: 20180725_jm_mem_leak.png, > 41K_ExecutionVertex_objs_retained_9GB.png, ExecutionVertexZoomIn.png > > > First, some context about the job > * Flink 1.4.1 > * stand-alone deployment mode > * embarrassingly parallel: all operators are chained together > * parallelism is over 1,000 > * stateless except for Kafka source operators. checkpoint size is 8.4 MB. > * set "state.backend.fs.memory-threshold" so that only jobmanager writes to > S3 to checkpoint > * internal checkpoint with 10 checkpoints retained in history > > Summary of the observations > * 41,567 ExecutionVertex objects retained 9+ GB of memory > * Expanded in one ExecutionVertex. it seems to storing the kafka offsets for > source operator -- This message was sent by Atlassian JIRA (v7.6.3#76005)
[jira] [Comment Edited] (FLINK-9693) Possible memory leak in jobmanager retaining archived checkpoints
[ https://issues.apache.org/jira/browse/FLINK-9693?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16556325#comment-16556325 ] Steven Zhen Wu edited comment on FLINK-9693 at 7/25/18 11:23 PM: - We can actually reproduce the issue by killing jobmanager node for very large jobs, like parallelism over 1,000. This issue starts to appear when replacement jobmanager node came up. was (Author: stevenz3wu): One more observation. we are seeing this issue right after the jobmanager node got killed and replaced. however, it is not reproducible when I trying to kill the jobmanager when job is healthy. so still don't know what exactly conditions triggered the issue > Possible memory leak in jobmanager retaining archived checkpoints > - > > Key: FLINK-9693 > URL: https://issues.apache.org/jira/browse/FLINK-9693 > Project: Flink > Issue Type: Bug > Components: JobManager, State Backends, Checkpointing >Affects Versions: 1.5.0, 1.6.0 > Environment: !image.png!!image (1).png! >Reporter: Steven Zhen Wu >Assignee: Till Rohrmann >Priority: Major > Labels: pull-request-available > Fix For: 1.4.3, 1.5.1, 1.6.0 > > Attachments: 20180725_jm_mem_leak.png, > 41K_ExecutionVertex_objs_retained_9GB.png, ExecutionVertexZoomIn.png > > > First, some context about the job > * Flink 1.4.1 > * stand-alone deployment mode > * embarrassingly parallel: all operators are chained together > * parallelism is over 1,000 > * stateless except for Kafka source operators. checkpoint size is 8.4 MB. > * set "state.backend.fs.memory-threshold" so that only jobmanager writes to > S3 to checkpoint > * internal checkpoint with 10 checkpoints retained in history > > Summary of the observations > * 41,567 ExecutionVertex objects retained 9+ GB of memory > * Expanded in one ExecutionVertex. it seems to storing the kafka offsets for > source operator -- This message was sent by Atlassian JIRA (v7.6.3#76005)
[jira] [Comment Edited] (FLINK-9693) Possible memory leak in jobmanager retaining archived checkpoints
[ https://issues.apache.org/jira/browse/FLINK-9693?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16556325#comment-16556325 ] Steven Zhen Wu edited comment on FLINK-9693 at 7/26/18 5:15 AM: We can actually reproduce the issue by killing jobmanager node for very large jobs, like parallelism over 1,000. This issue starts to appear when replacement jobmanager node came up. Another observation is that ~10 GB memory leak seems to happen very quickly (like < a few mins). was (Author: stevenz3wu): We can actually reproduce the issue by killing jobmanager node for very large jobs, like parallelism over 1,000. This issue starts to appear when replacement jobmanager node came up. > Possible memory leak in jobmanager retaining archived checkpoints > - > > Key: FLINK-9693 > URL: https://issues.apache.org/jira/browse/FLINK-9693 > Project: Flink > Issue Type: Bug > Components: JobManager, State Backends, Checkpointing >Affects Versions: 1.5.0, 1.6.0 > Environment: !image.png!!image (1).png! >Reporter: Steven Zhen Wu >Assignee: Till Rohrmann >Priority: Major > Labels: pull-request-available > Fix For: 1.4.3, 1.5.1, 1.6.0 > > Attachments: 20180725_jm_mem_leak.png, > 41K_ExecutionVertex_objs_retained_9GB.png, ExecutionVertexZoomIn.png > > > First, some context about the job > * Flink 1.4.1 > * stand-alone deployment mode > * embarrassingly parallel: all operators are chained together > * parallelism is over 1,000 > * stateless except for Kafka source operators. checkpoint size is 8.4 MB. > * set "state.backend.fs.memory-threshold" so that only jobmanager writes to > S3 to checkpoint > * internal checkpoint with 10 checkpoints retained in history > > Summary of the observations > * 41,567 ExecutionVertex objects retained 9+ GB of memory > * Expanded in one ExecutionVertex. it seems to storing the kafka offsets for > source operator -- This message was sent by Atlassian JIRA (v7.6.3#76005)
[jira] [Comment Edited] (FLINK-9693) Possible memory leak in jobmanager retaining archived checkpoints
[ https://issues.apache.org/jira/browse/FLINK-9693?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16556325#comment-16556325 ] Steven Zhen Wu edited comment on FLINK-9693 at 7/27/18 12:57 AM: - We can actually reproduce the issue by killing jobmanager node for very large jobs, like parallelism over 1,000. This issue starts to appear when replacement jobmanager node came up. Another observation is that ~10 GB memory leak seems to happen very quickly (like < a few mins). A few words about our setup * standalone cluster * HA mode using zookeeper * single jobmanager. we also tried two jobmanagers setup, same issue was (Author: stevenz3wu): We can actually reproduce the issue by killing jobmanager node for very large jobs, like parallelism over 1,000. This issue starts to appear when replacement jobmanager node came up. Another observation is that ~10 GB memory leak seems to happen very quickly (like < a few mins). > Possible memory leak in jobmanager retaining archived checkpoints > - > > Key: FLINK-9693 > URL: https://issues.apache.org/jira/browse/FLINK-9693 > Project: Flink > Issue Type: Bug > Components: JobManager, State Backends, Checkpointing >Affects Versions: 1.5.0, 1.6.0 > Environment: !image.png!!image (1).png! >Reporter: Steven Zhen Wu >Assignee: Till Rohrmann >Priority: Major > Labels: pull-request-available > Fix For: 1.4.3, 1.5.1, 1.6.0 > > Attachments: 20180725_jm_mem_leak.png, > 41K_ExecutionVertex_objs_retained_9GB.png, ExecutionVertexZoomIn.png > > > First, some context about the job > * Flink 1.4.1 > * stand-alone deployment mode > * embarrassingly parallel: all operators are chained together > * parallelism is over 1,000 > * stateless except for Kafka source operators. checkpoint size is 8.4 MB. > * set "state.backend.fs.memory-threshold" so that only jobmanager writes to > S3 to checkpoint > * internal checkpoint with 10 checkpoints retained in history > > Summary of the observations > * 41,567 ExecutionVertex objects retained 9+ GB of memory > * Expanded in one ExecutionVertex. it seems to storing the kafka offsets for > source operator -- This message was sent by Atlassian JIRA (v7.6.3#76005)