[jira] [Commented] (FLINK-10423) Forward RocksDB memory metrics to Flink metrics reporter
[ https://issues.apache.org/jira/browse/FLINK-10423?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=16627866#comment-16627866 ] Monal Daxini commented on FLINK-10423: -- It will be great to have this ported to 1.6 release as well. Spoke with [~srichter] about this offline. > Forward RocksDB memory metrics to Flink metrics reporter > - > > Key: FLINK-10423 > URL: https://issues.apache.org/jira/browse/FLINK-10423 > Project: Flink > Issue Type: New Feature > Components: Metrics, State Backends, Checkpointing >Reporter: Seth Wiesman >Assignee: Seth Wiesman >Priority: Major > > RocksDB contains a number of metrics at the column family level about current > memory usage, open memtables, etc that would be useful to users wishing > greater insight what rocksdb is doing. This work is inspired heavily by the > comments on this rocksdb issue thread > (https://github.com/facebook/rocksdb/issues/3216#issuecomment-348779233) -- This message was sent by Atlassian JIRA (v7.6.3#76005)
[jira] [Comment Edited] (FLINK-10423) Forward RocksDB memory metrics to Flink metrics reporter
[ https://issues.apache.org/jira/browse/FLINK-10423?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=16627866#comment-16627866 ] Monal Daxini edited comment on FLINK-10423 at 9/25/18 8:15 PM: --- It will be great to have this ported to 1.6 release. Spoke with [~srichter] about this offline. was (Author: mdaxini): It will be great to have this ported to 1.6 release as well. Spoke with [~srichter] about this offline. > Forward RocksDB memory metrics to Flink metrics reporter > - > > Key: FLINK-10423 > URL: https://issues.apache.org/jira/browse/FLINK-10423 > Project: Flink > Issue Type: New Feature > Components: Metrics, State Backends, Checkpointing >Reporter: Seth Wiesman >Assignee: Seth Wiesman >Priority: Major > > RocksDB contains a number of metrics at the column family level about current > memory usage, open memtables, etc that would be useful to users wishing > greater insight what rocksdb is doing. This work is inspired heavily by the > comments on this rocksdb issue thread > (https://github.com/facebook/rocksdb/issues/3216#issuecomment-348779233) -- This message was sent by Atlassian JIRA (v7.6.3#76005)
[jira] [Commented] (FLINK-9061) add entropy to s3 path for better scalability
[ https://issues.apache.org/jira/browse/FLINK-9061?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=16505447#comment-16505447 ] Monal Daxini commented on FLINK-9061: - In addition to what [~stevenz3wu] and [~jgrier] suggest, it will be good to make the entropy generation pluggable. This way users can override the default entropy generation, if they need to. {code:java} // code placeholder state.backend.fs.checkpointdir.injectEntropy.strategy=com.foo.MyStaticEntropyGenerator {code} > add entropy to s3 path for better scalability > - > > Key: FLINK-9061 > URL: https://issues.apache.org/jira/browse/FLINK-9061 > Project: Flink > Issue Type: Bug > Components: FileSystem, State Backends, Checkpointing >Affects Versions: 1.5.0, 1.4.2 >Reporter: Jamie Grier >Assignee: Indrajit Roychoudhury >Priority: Critical > > I think we need to modify the way we write checkpoints to S3 for high-scale > jobs (those with many total tasks). The issue is that we are writing all the > checkpoint data under a common key prefix. This is the worst case scenario > for S3 performance since the key is used as a partition key. > > In the worst case checkpoints fail with a 500 status code coming back from S3 > and an internal error type of TooBusyException. > > One possible solution would be to add a hook in the Flink filesystem code > that allows me to "rewrite" paths. For example say I have the checkpoint > directory set to: > > s3://bucket/flink/checkpoints > > I would hook that and rewrite that path to: > > s3://bucket/[HASH]/flink/checkpoints, where HASH is the hash of the original > path > > This would distribute the checkpoint write load around the S3 cluster evenly. > > For reference: > https://aws.amazon.com/premiumsupport/knowledge-center/s3-bucket-performance-improve/ > > Any other people hit this issue? Any other ideas for solutions? This is a > pretty serious problem for people trying to checkpoint to S3. > > -Jamie > -- This message was sent by Atlassian JIRA (v7.6.3#76005)
[jira] [Commented] (FLINK-8571) Provide an enhanced KeyedStream implementation to use ForwardPartitioner
[ https://issues.apache.org/jira/browse/FLINK-8571?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=16355877#comment-16355877 ] Monal Daxini commented on FLINK-8571: - Thanks Stefan for the quick turn around time! Can this patch be applied directly to 1.4, and would this be available in the next 1.4.1 release as well? > Provide an enhanced KeyedStream implementation to use ForwardPartitioner > > > Key: FLINK-8571 > URL: https://issues.apache.org/jira/browse/FLINK-8571 > Project: Flink > Issue Type: Improvement >Reporter: Nagarjun Guraja >Assignee: Stefan Richter >Priority: Major > > This enhancement would help in modeling problems with pre partitioned input > sources(for e.g. Kafka with Keyed topics). This would help in making the job > graph embarrassingly parallel while leveraging rocksdb state backend and also > the fine grained recovery semantics. -- This message was sent by Atlassian JIRA (v7.6.3#76005)
[jira] [Commented] (FLINK-8042) Retry individual failover-strategy for some time first before reverting to full job restart
[ https://issues.apache.org/jira/browse/FLINK-8042?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=16354124#comment-16354124 ] Monal Daxini commented on FLINK-8042: - This bug is quite detrimental to workloads where there is large state and is massively parallel. This is potential blocker for one of our use cases. Causing a full job restart takes while to recover. With local state recovery (potentially targeted for 1.5) and running the job with 1 or 2 extra TaskManagers might mitigate this to an extent. However, does not solve the problem. It would be good to address this sooner than later. Do we have a design doc that outlines the interface changes, if not can we please start one. > Retry individual failover-strategy for some time first before reverting to > full job restart > --- > > Key: FLINK-8042 > URL: https://issues.apache.org/jira/browse/FLINK-8042 > Project: Flink > Issue Type: Bug > Components: ResourceManager, State Backends, Checkpointing >Affects Versions: 1.3.2 >Reporter: Steven Zhen Wu >Priority: Blocker > Fix For: 1.5.0 > > > Let's we will a taskmanager node. When Flink tries to attempt fine grained > recovery and fails replacement taskmanager node didn't come back in time, it > reverts to full job restart. > Stephan and Till was suggesting that Flink can/should retry fine grained > recovery for some time before giving up and reverting full job restart -- This message was sent by Atlassian JIRA (v7.6.3#76005)
[jira] [Commented] (FLINK-4596) RESTART_STRATEGY is not really pluggable
[ https://issues.apache.org/jira/browse/FLINK-4596?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=15513734#comment-15513734 ] Monal Daxini commented on FLINK-4596: - Hi [~StephanEwen] what are your thoughts on this? > RESTART_STRATEGY is not really pluggable > > > Key: FLINK-4596 > URL: https://issues.apache.org/jira/browse/FLINK-4596 > Project: Flink > Issue Type: Bug >Reporter: Nagarjun Guraja > > Standalone cluster config accepts an implementation(class) as part of the > yaml config file but that does not work either as cluster level restart > strategy or streaming job level restart strategy > CLUSTER LEVEL CAUSE: createRestartStrategyFactory converts configured value > of strategyname to lowercase and searches for class name using lowercased > string. > JOB LEVEL CAUSE: Checkpointed streams have specific code to add > fixeddelayrestartconfiguration if no RestartConfiguration is specified in > the job env. Also, jobs cannot provide their own custom restart strategy > implementation and are constrained to pick up one of the three restart > strategies provided by flink. > FIX: Do not lower case the strategy config value, support a new > restartconfiguration to fallback to cluster level restart strategy and > support jobs to provide custom implementation of the strategy class itself. -- This message was sent by Atlassian JIRA (v6.3.4#6332)