Failing to recover once checkpoint fails

[Aljoscha Krettek]

Wait a sec, I just checked out the code again and it seems we already do that: https://github.com/apache/flink/blob/9071e3befb8c279f73c3094c9f6bddc0e7cce9e5/flink-runtime/src/main/java/org/apache/flink/runtime/checkpoint/ZooKeeperCompletedCheckpointStore.java#L210

If there were some checkpoints but none could be read we fail recovery.

On 24. Jan 2018, at 11:32, Aljoscha Krettek <[hidden email]> wrote:

That sounds reasonable: We would keep the first fix, i.e. never delete checkpoints if they're "corrupt", only when they're subsumed. Additionally, we fail the job if there are some checkpoints in ZooKeeper but none of them can be restored to prevent the case where a job starts from scratch even though it shouldn't.

Does that sum it up?

On 24. Jan 2018, at 01:19, Vishal Santoshi <[hidden email]> wrote:

If we hit the retry limit, abort the job. In our case we will restart from the last SP ( we as any production pile do it is n time s a day )  and that I would think should be OK for most folks ?

On Tue, Jan 23, 2018 at 11:38 AM, Vishal Santoshi <[hidden email]> wrote:

Thank you for considering this. If I understand you correctly.

* CHK pointer on ZK for a CHK state on hdfs was done successfully.

* Some issue restarted the pipeline.

* The NN was down unfortunately and flink could not retrieve the  CHK state from the CHK pointer on ZK.

Before 

* The CHK pointer was being removed and the job started from a brand new slate.

After ( this fix on 1.4 +) 

* do not delete the CHK pointer ( It has to be subsumed to be deleted ). 

* Flink keeps using this CHK pointer ( if retry is > 0 and till we hit any retry limit ) to restore state

* NN comes back 

* Flink restores state on the next retry.

I would hope that is the sequence to follow.

Regards.

On Tue, Jan 23, 2018 at 7:25 AM, Aljoscha Krettek <[hidden email]> wrote:

Hi Vishal,

I think you might be right. We fixed the problem that checkpoints where dropped via https://issues.apache.org/jira/browse/FLINK-7783. However, we still have the problem that if the DFS is not up at all then it will look as if the job is starting from scratch. However, the alternative is failing the job, in which case you will also never be able to restore from a checkpoint. What do you think?

Best,

Aljoscha

On 23. Jan 2018, at 10:15, Fabian Hueske <[hidden email]> wrote:

Sorry for the late reply.

I created FLINK-8487 [1] to track this problem

@Vishal, can you have a look and check if if forgot some details? I logged the issue for Flink 1.3.2, is that correct?

Please add more information if you think it is relevant.

Thanks,

Fabian

[1] https://issues.apache.org/jira/browse/FLINK-8487

2018-01-18 22:14 GMT+01:00 Vishal Santoshi <[hidden email]>:

Or this one 

https://issues.apache.org/jira/browse/FLINK-4815

On Thu, Jan 18, 2018 at 4:13 PM, Vishal Santoshi <[hidden email]> wrote:

ping. 

    This happened again on production and it seems reasonable to abort when a checkpoint is not found rather than behave as if it is a brand new pipeline.  

On Tue, Jan 16, 2018 at 9:33 AM, Vishal Santoshi <[hidden email]> wrote:

Folks sorry for being late on this. Can some body with the knowledge of this code base create a jira issue for the above ? We have seen this more than once on production.

On Mon, Oct 9, 2017 at 10:21 AM, Aljoscha Krettek <[hidden email]> wrote:

Hi Vishal,

Some relevant Jira issues for you are:

 - https://issues.apache.org/jira/browse/FLINK-4808: Allow skipping failed checkpoints

 - https://issues.apache.org/jira/browse/FLINK-4815: Automatic fallback to earlier checkpoint when checkpoint restore fails

 - https://issues.apache.org/jira/browse/FLINK-7783: Don't always remove checkpoints in ZooKeeperCompletedCheckpointStore#recover()

Best,

Aljoscha

On 9. Oct 2017, at 09:06, Fabian Hueske <[hidden email]> wrote:

Hi Vishal,

it would be great if you could create a JIRA ticket with Blocker priority.
Please add all relevant information of your detailed analysis, add a link to this email thread (see [1] for the web archive of the mailing list), and post the id of the JIRA issue here.

Thanks for looking into this!

Best regards,

Fabian

[1] https://lists.apache.org/list.html?user@...

2017-10-06 15:59 GMT+02:00 Vishal Santoshi <[hidden email]>:

Thank you for confirming. 

       

 I think this is a critical bug. In essence any checkpoint store ( hdfs/S3/File)  will loose state if it is unavailable at resume. This becomes all the more painful with your confirming that  "failed checkpoints killing the job"  b'coz essentially it mean that if remote store in unavailable  during checkpoint than you have lost state ( till of course you have a retry of none or an unbounded retry delay, a delay that you hope the store revives in ) .. Remember  the first retry failure  will cause new state according the code as written iff the remote store is down. We would rather have a configurable property that establishes  our desire to abort something like a "abort_retry_on_chkretrevalfailure"

In our case it is very important that we do not undercount a window, one reason we use flink and it's awesome failure guarantees, as various alarms sound ( we do anomaly detection on the time series ).

Please create a jira ticket for us to follow or we could do it.

PS Not aborting on checkpointing, till a configurable limit is very important too.

On Fri, Oct 6, 2017 at 2:36 AM, Aljoscha Krettek <[hidden email]> wrote:

Hi Vishal,

I think you're right! And thanks for looking into this so deeply. 

With your last mail your basically saying, that the checkpoint could not be restored because your HDFS was temporarily down. If Flink had not deleted that checkpoint it might have been possible to restore it at a later point, right?

Regarding failed checkpoints killing the job: yes, this is currently the expected behaviour but there are plans to change this.

Best,

Aljoscha

On 5. Oct 2017, at 17:40, Vishal Santoshi <[hidden email]> wrote:

I think this is the offending piece. There is a catch all Exception, which IMHO should understand a recoverable exception from an unrecoverable on. 

try {
	completedCheckpoint = retrieveCompletedCheckpoint(checkpointStateHandle);
	if (completedCheckpoint != null) {
	completedCheckpoints.add(completedCheckpoint);
	}
	} catch (Exception e) {
	LOG.warn("Could not retrieve checkpoint. Removing it from the completed " +
	"checkpoint store.", e);
	// remove the checkpoint with broken state handle
	removeBrokenStateHandle(checkpointStateHandle.f1, checkpointStateHandle.f0);
	}

On Thu, Oct 5, 2017 at 10:57 AM, Vishal Santoshi <[hidden email]> wrote:

So this is the issue and tell us that it is wrong. ZK had some state ( backed by hdfs ) that referred to a checkpoint ( the same exact last successful checkpoint that was successful before NN screwed us ). When the JM tried to recreate the state and b'coz NN was down failed to retrieve the CHK handle from hdfs and conveniently ( and I think very wrongly ) removed the CHK from being considered and cleaned the pointer ( though failed as was NN was down and is obvious from the dangling file in recovery ) . The metadata itself was on hdfs and failure in retrieving should have been a stop all, not going to trying doing magic exception rather than starting from a blank state.

org.apache.flink.util.FlinkException: Could not retrieve checkpoint 44286 from state handle under /0000000000000044286. This indicates that the retrieved state handle is broken. Try cleaning the state handle store.

On Thu, Oct 5, 2017 at 10:13 AM, Vishal Santoshi <[hidden email]> wrote:

Also note that  the zookeeper recovery did  ( sadly on the same hdfs cluster ) also showed the same behavior. It had the pointers to the chk point  ( I  think that is what it does, keeps metadata of where the checkpoint etc  ) .  It too decided to keep the recovery file from the failed state.

-rw-r--r--   3 root hadoop       7041 2017-10-04 13:55 /flink-recovery/prod/completedCheckpoint6c9096bb9ed4

-rw-r--r--   3 root hadoop       7044 2017-10-05 10:07 /flink-recovery/prod/completedCheckpoint7c5a19300092

This is getting a little interesting. What say you :)

On Thu, Oct 5, 2017 at 9:26 AM, Vishal Santoshi <[hidden email]> wrote:

Another thing I noted was this thing

drwxr-xr-x   - root hadoop          0 2017-10-04 13:54 /flink-checkpoints/prod/c4af8dfa864e2f9a51764de9f0725b39/chk-44286

drwxr-xr-x   - root hadoop          0 2017-10-05 09:15 /flink-checkpoints/prod/c4af8dfa864e2f9a51764de9f0725b39/chk-45428

Generally what Flink does IMHO is that it replaces the chk point directory with a new one. I see it happening now. Every minute it replaces the old directory.  In this job's case however, it did not delete the 2017-10-04 13:54  and hence the chk-44286 directory.  This was the last chk-44286 (  I think  )  successfully created before NN had issues but as is usual did not delete this  chk-44286. It looks as if it started with a blank slate ???????? Does this strike a chord ?????

On Thu, Oct 5, 2017 at 8:56 AM, Vishal Santoshi <[hidden email]> wrote:

Hello Fabian, 

                      First of all congratulations on this fabulous framework. I have worked with GDF and though GDF has some natural pluses Flink's state management is far more advanced. With kafka as a source it negates issues GDF has ( GDF integration with pub/sub is organic and that is to be expected but non FIFO pub/sub is an issue with windows on event time etc )

                   Coming back to this issue. We have that same kafka topic feeding a streaming druid datasource and we do not see any issue there, so so data loss on the source, kafka is not applicable. I am totally certain that the "retention" time was not an issue. It is 4 days of retention and we fixed this issue within 30 minutes. We could replay kafka with a new consumer group.id and that worked fine. 

Note these properties and see if they strike a chord.

* The setCommitOffsetsOnCheckpoints(boolean) for kafka consumers is the default true. I bring this up to see whether flink will in any circumstance drive consumption on the kafka perceived offset rather than the one in the checkpoint.

* The state.backend.fs.memory-threshold: 0 has not been set.  The state is big enough though therefore IMHO no way the state is stored along with the meta data in JM ( or ZK ? ) . The reason I bring this up is to make sure when you say that the size has to be less than 1024bytes , you are talking about cumulative state of the pipeine.

* We have a good sense of SP ( save point )  and CP ( checkpoint ) and certainly understand that they actually are not dissimilar. However in this case there were multiple attempts to restart the pipe before it finally succeeded. 

* Other hdfs related poperties.

 

 state.backend.fs.checkpointdir: hdfs:///flink-checkpoints/<%= flink_hdfs_root %>

 state.savepoints.dir: hdfs:///flink-savepoints/<%= flink_hdfs_root %>

 recovery.zookeeper.storageDir: hdfs:///flink-recovery/<%= flink_hdfs_root %>

Do these make sense ? Is there anything else I should look at.  Please also note that it is the second time this has happened. The first time I was vacationing and was not privy to the state of the flink pipeline, but the net effect were similar. The counts for the first window after an internal restart dropped. 

Thank you for you patience and regards,

Vishal

On Thu, Oct 5, 2017 at 5:01 AM, Fabian Hueske <[hidden email]> wrote:

Hi Vishal,

window operators are always stateful because the operator needs to remember previously received events (WindowFunction) or intermediate results (ReduceFunction).

Given the program you described, a checkpoint should include the Kafka consumer offset and the state of the window operator. If the program eventually successfully (i.e., without an error) recovered from the last checkpoint, all its state should have been restored. Since the last checkpoint was before HDFS went into safe mode, the program would have been reset to that point. If the Kafka retention time is less than the time it took to fix HDFS you would have lost data because it would have been removed from Kafka. If that's not the case, we need to investigate this further because a checkpoint recovery must not result in state loss.

Restoring from a savepoint is not so much different from automatic checkpoint recovery. Given that you have a completed savepoint, you can restart the job from that point. The main difference is that checkpoints are only used for internal recovery and usually discarded once the job is terminated while savepoints are retained.

Regarding your question if a failed checkpoint should cause the job to fail and recover I'm not sure what the current status is.

Stefan (in CC) should know what happens if a checkpoint fails.

Best, Fabian

2017-10-05 2:20 GMT+02:00 Vishal Santoshi <[hidden email]>:

To add to it, my pipeline is a simple 

keyBy(0)
        .timeWindow(Time.of(window_size, TimeUnit.MINUTES))
        .allowedLateness(Time.of(late_by, TimeUnit.SECONDS))
        .reduce(new ReduceFunction(), new WindowFunction())

On Wed, Oct 4, 2017 at 8:19 PM, Vishal Santoshi <[hidden email]> wrote:

Hello folks,

As far as I know checkpoint failure should be ignored and retried with potentially larger state. I had this situation

* hdfs went into a safe mode b'coz of Name Node issues
* exception was thrown

    org.apache.hadoop.ipc.RemoteException(org.apache.hadoop.ipc.StandbyException): Operation category WRITE is not supported in state standby. Visit https://s.apache.org/sbnn-error
    ..................

    at org.apache.flink.runtime.fs.hdfs.HadoopFileSystem.mkdirs(HadoopFileSystem.java:453)
        at org.apache.flink.core.fs.SafetyNetWrapperFileSystem.mkdirs(SafetyNetWrapperFileSystem.java:111)
        at org.apache.flink.runtime.state.filesystem.FsCheckpointStreamFactory.createBasePath(FsCheckpointStreamFactory.java:132)

* The pipeline came back after a few restarts and checkpoint failures, after the hdfs issues were resolved.

I would not have worried about the restart, but it was evident that I lost my operator state. Either it was my kafka consumer that kept on advancing it's offset between a start and the next checkpoint failure ( a minute's worth ) or the the operator that had partial aggregates was lost. I have a 15 minute window of counts on a keyed operator

I am using ROCKS DB and of course have checkpointing turned on.

The questions thus are

* Should a pipeline be restarted if checkpoint fails ?
* Why on restart did the operator state did not recreate ?
* Is the nature of the exception thrown have to do with any of this b'coz suspend and resume from a save point work as expected ?
* And though I am pretty sure, are operators like the Window operator stateful by drfault and thus if I have timeWindow(Time.of(window_size, TimeUnit.MINUTES)).reduce(new ReduceFunction(), new WindowFunction()), the state is managed by flink ?

Thanks.

[Vishal Santoshi]

As in, if there are chk handles in zk, there should no reason to start a new job ( bad handle, no hdfs connectivity etc ),

 yes that sums it up.

On Wed, Jan 24, 2018 at 5:35 AM, Aljoscha Krettek <[hidden email]> wrote:

Wait a sec, I just checked out the code again and it seems we already do that: https://github.com/apache/flink/blob/9071e3befb8c279f73c3094c9f6bddc0e7cce9e5/flink-runtime/src/main/java/org/apache/flink/runtime/checkpoint/ZooKeeperCompletedCheckpointStore.java#L210

If there were some checkpoints but none could be read we fail recovery.

On 24. Jan 2018, at 11:32, Aljoscha Krettek <[hidden email]> wrote:

That sounds reasonable: We would keep the first fix, i.e. never delete checkpoints if they're "corrupt", only when they're subsumed. Additionally, we fail the job if there are some checkpoints in ZooKeeper but none of them can be restored to prevent the case where a job starts from scratch even though it shouldn't.

Does that sum it up?

On 24. Jan 2018, at 01:19, Vishal Santoshi <[hidden email]> wrote:

If we hit the retry limit, abort the job. In our case we will restart from the last SP ( we as any production pile do it is n time s a day )  and that I would think should be OK for most folks ?

On Tue, Jan 23, 2018 at 11:38 AM, Vishal Santoshi <[hidden email]> wrote:

Thank you for considering this. If I understand you correctly.

* CHK pointer on ZK for a CHK state on hdfs was done successfully.

* Some issue restarted the pipeline.

* The NN was down unfortunately and flink could not retrieve the  CHK state from the CHK pointer on ZK.

Before 

* The CHK pointer was being removed and the job started from a brand new slate.

After ( this fix on 1.4 +) 

* do not delete the CHK pointer ( It has to be subsumed to be deleted ). 

* Flink keeps using this CHK pointer ( if retry is > 0 and till we hit any retry limit ) to restore state

* NN comes back 

* Flink restores state on the next retry.

I would hope that is the sequence to follow.

Regards.

On Tue, Jan 23, 2018 at 7:25 AM, Aljoscha Krettek <[hidden email]> wrote:

Hi Vishal,

I think you might be right. We fixed the problem that checkpoints where dropped via https://issues.apache.org/jira/browse/FLINK-7783. However, we still have the problem that if the DFS is not up at all then it will look as if the job is starting from scratch. However, the alternative is failing the job, in which case you will also never be able to restore from a checkpoint. What do you think?

Best,

Aljoscha

On 23. Jan 2018, at 10:15, Fabian Hueske <[hidden email]> wrote:

Sorry for the late reply.

I created FLINK-8487 [1] to track this problem

@Vishal, can you have a look and check if if forgot some details? I logged the issue for Flink 1.3.2, is that correct?

Please add more information if you think it is relevant.

Thanks,

Fabian

[1] https://issues.apache.org/jira/browse/FLINK-8487

2018-01-18 22:14 GMT+01:00 Vishal Santoshi <[hidden email]>:

Or this one 

https://issues.apache.org/jira/browse/FLINK-4815

On Thu, Jan 18, 2018 at 4:13 PM, Vishal Santoshi <[hidden email]> wrote:

ping. 

    This happened again on production and it seems reasonable to abort when a checkpoint is not found rather than behave as if it is a brand new pipeline.  

On Tue, Jan 16, 2018 at 9:33 AM, Vishal Santoshi <[hidden email]> wrote:

Folks sorry for being late on this. Can some body with the knowledge of this code base create a jira issue for the above ? We have seen this more than once on production.

On Mon, Oct 9, 2017 at 10:21 AM, Aljoscha Krettek <[hidden email]> wrote:

Hi Vishal,

Some relevant Jira issues for you are:

 - https://issues.apache.org/jira/browse/FLINK-4808: Allow skipping failed checkpoints

 - https://issues.apache.org/jira/browse/FLINK-4815: Automatic fallback to earlier checkpoint when checkpoint restore fails

 - https://issues.apache.org/jira/browse/FLINK-7783: Don't always remove checkpoints in ZooKeeperCompletedCheckpointStore#recover()

Best,

Aljoscha

On 9. Oct 2017, at 09:06, Fabian Hueske <[hidden email]> wrote:

Hi Vishal,

it would be great if you could create a JIRA ticket with Blocker priority.
Please add all relevant information of your detailed analysis, add a link to this email thread (see [1] for the web archive of the mailing list), and post the id of the JIRA issue here.

Thanks for looking into this!

Best regards,

Fabian

[1] https://lists.apache.org/list.html?user@...

2017-10-06 15:59 GMT+02:00 Vishal Santoshi <[hidden email]>:

Thank you for confirming. 

       

 I think this is a critical bug. In essence any checkpoint store ( hdfs/S3/File)  will loose state if it is unavailable at resume. This becomes all the more painful with your confirming that  "failed checkpoints killing the job"  b'coz essentially it mean that if remote store in unavailable  during checkpoint than you have lost state ( till of course you have a retry of none or an unbounded retry delay, a delay that you hope the store revives in ) .. Remember  the first retry failure  will cause new state according the code as written iff the remote store is down. We would rather have a configurable property that establishes  our desire to abort something like a "abort_retry_on_chkretrevalfailure"

In our case it is very important that we do not undercount a window, one reason we use flink and it's awesome failure guarantees, as various alarms sound ( we do anomaly detection on the time series ).

Please create a jira ticket for us to follow or we could do it.

PS Not aborting on checkpointing, till a configurable limit is very important too.

On Fri, Oct 6, 2017 at 2:36 AM, Aljoscha Krettek <[hidden email]> wrote:

Hi Vishal,

I think you're right! And thanks for looking into this so deeply. 

With your last mail your basically saying, that the checkpoint could not be restored because your HDFS was temporarily down. If Flink had not deleted that checkpoint it might have been possible to restore it at a later point, right?

Regarding failed checkpoints killing the job: yes, this is currently the expected behaviour but there are plans to change this.

Best,

Aljoscha

On 5. Oct 2017, at 17:40, Vishal Santoshi <[hidden email]> wrote:

I think this is the offending piece. There is a catch all Exception, which IMHO should understand a recoverable exception from an unrecoverable on. 

try {
	completedCheckpoint = retrieveCompletedCheckpoint(checkpointStateHandle);
	if (completedCheckpoint != null) {
	completedCheckpoints.add(completedCheckpoint);
	}
	} catch (Exception e) {
	LOG.warn("Could not retrieve checkpoint. Removing it from the completed " +
	"checkpoint store.", e);
	// remove the checkpoint with broken state handle
	removeBrokenStateHandle(checkpointStateHandle.f1, checkpointStateHandle.f0);
	}

On Thu, Oct 5, 2017 at 10:57 AM, Vishal Santoshi <[hidden email]> wrote:

So this is the issue and tell us that it is wrong. ZK had some state ( backed by hdfs ) that referred to a checkpoint ( the same exact last successful checkpoint that was successful before NN screwed us ). When the JM tried to recreate the state and b'coz NN was down failed to retrieve the CHK handle from hdfs and conveniently ( and I think very wrongly ) removed the CHK from being considered and cleaned the pointer ( though failed as was NN was down and is obvious from the dangling file in recovery ) . The metadata itself was on hdfs and failure in retrieving should have been a stop all, not going to trying doing magic exception rather than starting from a blank state.

org.apache.flink.util.FlinkException: Could not retrieve checkpoint 44286 from state handle under /0000000000000044286. This indicates that the retrieved state handle is broken. Try cleaning the state handle store.

On Thu, Oct 5, 2017 at 10:13 AM, Vishal Santoshi <[hidden email]> wrote:

Also note that  the zookeeper recovery did  ( sadly on the same hdfs cluster ) also showed the same behavior. It had the pointers to the chk point  ( I  think that is what it does, keeps metadata of where the checkpoint etc  ) .  It too decided to keep the recovery file from the failed state.

-rw-r--r--   3 root hadoop       7041 2017-10-04 13:55 /flink-recovery/prod/completedCheckpoint6c9096bb9ed4

-rw-r--r--   3 root hadoop       7044 2017-10-05 10:07 /flink-recovery/prod/completedCheckpoint7c5a19300092

This is getting a little interesting. What say you :)

On Thu, Oct 5, 2017 at 9:26 AM, Vishal Santoshi <[hidden email]> wrote:

Another thing I noted was this thing

drwxr-xr-x   - root hadoop          0 2017-10-04 13:54 /flink-checkpoints/prod/c4af8dfa864e2f9a51764de9f0725b39/chk-44286

drwxr-xr-x   - root hadoop          0 2017-10-05 09:15 /flink-checkpoints/prod/c4af8dfa864e2f9a51764de9f0725b39/chk-45428

Generally what Flink does IMHO is that it replaces the chk point directory with a new one. I see it happening now. Every minute it replaces the old directory.  In this job's case however, it did not delete the 2017-10-04 13:54  and hence the chk-44286 directory.  This was the last chk-44286 (  I think  )  successfully created before NN had issues but as is usual did not delete this  chk-44286. It looks as if it started with a blank slate ???????? Does this strike a chord ?????

On Thu, Oct 5, 2017 at 8:56 AM, Vishal Santoshi <[hidden email]> wrote:

Hello Fabian, 

                      First of all congratulations on this fabulous framework. I have worked with GDF and though GDF has some natural pluses Flink's state management is far more advanced. With kafka as a source it negates issues GDF has ( GDF integration with pub/sub is organic and that is to be expected but non FIFO pub/sub is an issue with windows on event time etc )

                   Coming back to this issue. We have that same kafka topic feeding a streaming druid datasource and we do not see any issue there, so so data loss on the source, kafka is not applicable. I am totally certain that the "retention" time was not an issue. It is 4 days of retention and we fixed this issue within 30 minutes. We could replay kafka with a new consumer group.id and that worked fine. 

Note these properties and see if they strike a chord.

* The setCommitOffsetsOnCheckpoints(boolean) for kafka consumers is the default true. I bring this up to see whether flink will in any circumstance drive consumption on the kafka perceived offset rather than the one in the checkpoint.

* The state.backend.fs.memory-threshold: 0 has not been set.  The state is big enough though therefore IMHO no way the state is stored along with the meta data in JM ( or ZK ? ) . The reason I bring this up is to make sure when you say that the size has to be less than 1024bytes , you are talking about cumulative state of the pipeine.

* We have a good sense of SP ( save point )  and CP ( checkpoint ) and certainly understand that they actually are not dissimilar. However in this case there were multiple attempts to restart the pipe before it finally succeeded. 

* Other hdfs related poperties.

 

 state.backend.fs.checkpointdir: hdfs:///flink-checkpoints/<%= flink_hdfs_root %>

 state.savepoints.dir: hdfs:///flink-savepoints/<%= flink_hdfs_root %>

 recovery.zookeeper.storageDir: hdfs:///flink-recovery/<%= flink_hdfs_root %>

Do these make sense ? Is there anything else I should look at.  Please also note that it is the second time this has happened. The first time I was vacationing and was not privy to the state of the flink pipeline, but the net effect were similar. The counts for the first window after an internal restart dropped. 

Thank you for you patience and regards,

Vishal

On Thu, Oct 5, 2017 at 5:01 AM, Fabian Hueske <[hidden email]> wrote:

Hi Vishal,

window operators are always stateful because the operator needs to remember previously received events (WindowFunction) or intermediate results (ReduceFunction).

Given the program you described, a checkpoint should include the Kafka consumer offset and the state of the window operator. If the program eventually successfully (i.e., without an error) recovered from the last checkpoint, all its state should have been restored. Since the last checkpoint was before HDFS went into safe mode, the program would have been reset to that point. If the Kafka retention time is less than the time it took to fix HDFS you would have lost data because it would have been removed from Kafka. If that's not the case, we need to investigate this further because a checkpoint recovery must not result in state loss.

Restoring from a savepoint is not so much different from automatic checkpoint recovery. Given that you have a completed savepoint, you can restart the job from that point. The main difference is that checkpoints are only used for internal recovery and usually discarded once the job is terminated while savepoints are retained.

Regarding your question if a failed checkpoint should cause the job to fail and recover I'm not sure what the current status is.

Stefan (in CC) should know what happens if a checkpoint fails.

Best, Fabian

2017-10-05 2:20 GMT+02:00 Vishal Santoshi <[hidden email]>:

To add to it, my pipeline is a simple 

keyBy(0)
        .timeWindow(Time.of(window_size, TimeUnit.MINUTES))
        .allowedLateness(Time.of(late_by, TimeUnit.SECONDS))
        .reduce(new ReduceFunction(), new WindowFunction())

On Wed, Oct 4, 2017 at 8:19 PM, Vishal Santoshi <[hidden email]> wrote:

Hello folks,

As far as I know checkpoint failure should be ignored and retried with potentially larger state. I had this situation

* hdfs went into a safe mode b'coz of Name Node issues
* exception was thrown

    org.apache.hadoop.ipc.RemoteException(org.apache.hadoop.ipc.StandbyException): Operation category WRITE is not supported in state standby. Visit https://s.apache.org/sbnn-error
    ..................

    at org.apache.flink.runtime.fs.hdfs.HadoopFileSystem.mkdirs(HadoopFileSystem.java:453)
        at org.apache.flink.core.fs.SafetyNetWrapperFileSystem.mkdirs(SafetyNetWrapperFileSystem.java:111)
        at org.apache.flink.runtime.state.filesystem.FsCheckpointStreamFactory.createBasePath(FsCheckpointStreamFactory.java:132)

* The pipeline came back after a few restarts and checkpoint failures, after the hdfs issues were resolved.

I would not have worried about the restart, but it was evident that I lost my operator state. Either it was my kafka consumer that kept on advancing it's offset between a start and the next checkpoint failure ( a minute's worth ) or the the operator that had partial aggregates was lost. I have a 15 minute window of counts on a keyed operator

I am using ROCKS DB and of course have checkpointing turned on.

The questions thus are

* Should a pipeline be restarted if checkpoint fails ?
* Why on restart did the operator state did not recreate ?
* Is the nature of the exception thrown have to do with any of this b'coz suspend and resume from a save point work as expected ?
* And though I am pretty sure, are operators like the Window operator stateful by drfault and thus if I have timeWindow(Time.of(window_size, TimeUnit.MINUTES)).reduce(new ReduceFunction(), new WindowFunction()), the state is managed by flink ?

Thanks.

[Aljoscha Krettek]

Did you see my second mail?

On 24. Jan 2018, at 12:50, Vishal Santoshi <[hidden email]> wrote:

As in, if there are chk handles in zk, there should no reason to start a new job ( bad handle, no hdfs connectivity etc ),

 yes that sums it up.

On Wed, Jan 24, 2018 at 5:35 AM, Aljoscha Krettek <[hidden email]> wrote:

Wait a sec, I just checked out the code again and it seems we already do that: https://github.com/apache/flink/blob/9071e3befb8c279f73c3094c9f6bddc0e7cce9e5/flink-runtime/src/main/java/org/apache/flink/runtime/checkpoint/ZooKeeperCompletedCheckpointStore.java#L210

If there were some checkpoints but none could be read we fail recovery.

On 24. Jan 2018, at 11:32, Aljoscha Krettek <[hidden email]> wrote:

That sounds reasonable: We would keep the first fix, i.e. never delete checkpoints if they're "corrupt", only when they're subsumed. Additionally, we fail the job if there are some checkpoints in ZooKeeper but none of them can be restored to prevent the case where a job starts from scratch even though it shouldn't.

Does that sum it up?

On 24. Jan 2018, at 01:19, Vishal Santoshi <[hidden email]> wrote:

If we hit the retry limit, abort the job. In our case we will restart from the last SP ( we as any production pile do it is n time s a day )  and that I would think should be OK for most folks ?

On Tue, Jan 23, 2018 at 11:38 AM, Vishal Santoshi <[hidden email]> wrote:

Thank you for considering this. If I understand you correctly.

* CHK pointer on ZK for a CHK state on hdfs was done successfully.

* Some issue restarted the pipeline.

* The NN was down unfortunately and flink could not retrieve the  CHK state from the CHK pointer on ZK.

Before 

* The CHK pointer was being removed and the job started from a brand new slate.

After ( this fix on 1.4 +) 

* do not delete the CHK pointer ( It has to be subsumed to be deleted ). 

* Flink keeps using this CHK pointer ( if retry is > 0 and till we hit any retry limit ) to restore state

* NN comes back 

* Flink restores state on the next retry.

I would hope that is the sequence to follow.

Regards.

On Tue, Jan 23, 2018 at 7:25 AM, Aljoscha Krettek <[hidden email]> wrote:

Hi Vishal,

I think you might be right. We fixed the problem that checkpoints where dropped via https://issues.apache.org/jira/browse/FLINK-7783. However, we still have the problem that if the DFS is not up at all then it will look as if the job is starting from scratch. However, the alternative is failing the job, in which case you will also never be able to restore from a checkpoint. What do you think?

Best,

Aljoscha

On 23. Jan 2018, at 10:15, Fabian Hueske <[hidden email]> wrote:

Sorry for the late reply.

I created FLINK-8487 [1] to track this problem

@Vishal, can you have a look and check if if forgot some details? I logged the issue for Flink 1.3.2, is that correct?

Please add more information if you think it is relevant.

Thanks,

Fabian

[1] https://issues.apache.org/jira/browse/FLINK-8487

2018-01-18 22:14 GMT+01:00 Vishal Santoshi <[hidden email]>:

Or this one 

https://issues.apache.org/jira/browse/FLINK-4815

On Thu, Jan 18, 2018 at 4:13 PM, Vishal Santoshi <[hidden email]> wrote:

ping. 

    This happened again on production and it seems reasonable to abort when a checkpoint is not found rather than behave as if it is a brand new pipeline.  

On Tue, Jan 16, 2018 at 9:33 AM, Vishal Santoshi <[hidden email]> wrote:

Folks sorry for being late on this. Can some body with the knowledge of this code base create a jira issue for the above ? We have seen this more than once on production.

On Mon, Oct 9, 2017 at 10:21 AM, Aljoscha Krettek <[hidden email]> wrote:

Hi Vishal,

Some relevant Jira issues for you are:

 - https://issues.apache.org/jira/browse/FLINK-4808: Allow skipping failed checkpoints

 - https://issues.apache.org/jira/browse/FLINK-4815: Automatic fallback to earlier checkpoint when checkpoint restore fails

 - https://issues.apache.org/jira/browse/FLINK-7783: Don't always remove checkpoints in ZooKeeperCompletedCheckpointStore#recover()

Best,

Aljoscha

On 9. Oct 2017, at 09:06, Fabian Hueske <[hidden email]> wrote:

Hi Vishal,

it would be great if you could create a JIRA ticket with Blocker priority.
Please add all relevant information of your detailed analysis, add a link to this email thread (see [1] for the web archive of the mailing list), and post the id of the JIRA issue here.

Thanks for looking into this!

Best regards,

Fabian

[1] https://lists.apache.org/list.html?user@...

2017-10-06 15:59 GMT+02:00 Vishal Santoshi <[hidden email]>:

Thank you for confirming. 

       

 I think this is a critical bug. In essence any checkpoint store ( hdfs/S3/File)  will loose state if it is unavailable at resume. This becomes all the more painful with your confirming that  "failed checkpoints killing the job"  b'coz essentially it mean that if remote store in unavailable  during checkpoint than you have lost state ( till of course you have a retry of none or an unbounded retry delay, a delay that you hope the store revives in ) .. Remember  the first retry failure  will cause new state according the code as written iff the remote store is down. We would rather have a configurable property that establishes  our desire to abort something like a "abort_retry_on_chkretrevalfailure"

In our case it is very important that we do not undercount a window, one reason we use flink and it's awesome failure guarantees, as various alarms sound ( we do anomaly detection on the time series ).

Please create a jira ticket for us to follow or we could do it.

PS Not aborting on checkpointing, till a configurable limit is very important too.

On Fri, Oct 6, 2017 at 2:36 AM, Aljoscha Krettek <[hidden email]> wrote:

Hi Vishal,

I think you're right! And thanks for looking into this so deeply. 

With your last mail your basically saying, that the checkpoint could not be restored because your HDFS was temporarily down. If Flink had not deleted that checkpoint it might have been possible to restore it at a later point, right?

Regarding failed checkpoints killing the job: yes, this is currently the expected behaviour but there are plans to change this.

Best,

Aljoscha

On 5. Oct 2017, at 17:40, Vishal Santoshi <[hidden email]> wrote:

I think this is the offending piece. There is a catch all Exception, which IMHO should understand a recoverable exception from an unrecoverable on. 

try {
	completedCheckpoint = retrieveCompletedCheckpoint(checkpointStateHandle);
	if (completedCheckpoint != null) {
	completedCheckpoints.add(completedCheckpoint);
	}
	} catch (Exception e) {
	LOG.warn("Could not retrieve checkpoint. Removing it from the completed " +
	"checkpoint store.", e);
	// remove the checkpoint with broken state handle
	removeBrokenStateHandle(checkpointStateHandle.f1, checkpointStateHandle.f0);
	}

On Thu, Oct 5, 2017 at 10:57 AM, Vishal Santoshi <[hidden email]> wrote:

So this is the issue and tell us that it is wrong. ZK had some state ( backed by hdfs ) that referred to a checkpoint ( the same exact last successful checkpoint that was successful before NN screwed us ). When the JM tried to recreate the state and b'coz NN was down failed to retrieve the CHK handle from hdfs and conveniently ( and I think very wrongly ) removed the CHK from being considered and cleaned the pointer ( though failed as was NN was down and is obvious from the dangling file in recovery ) . The metadata itself was on hdfs and failure in retrieving should have been a stop all, not going to trying doing magic exception rather than starting from a blank state.

org.apache.flink.util.FlinkException: Could not retrieve checkpoint 44286 from state handle under /0000000000000044286. This indicates that the retrieved state handle is broken. Try cleaning the state handle store.

On Thu, Oct 5, 2017 at 10:13 AM, Vishal Santoshi <[hidden email]> wrote:

Also note that  the zookeeper recovery did  ( sadly on the same hdfs cluster ) also showed the same behavior. It had the pointers to the chk point  ( I  think that is what it does, keeps metadata of where the checkpoint etc  ) .  It too decided to keep the recovery file from the failed state.

-rw-r--r--   3 root hadoop       7041 2017-10-04 13:55 /flink-recovery/prod/completedCheckpoint6c9096bb9ed4

-rw-r--r--   3 root hadoop       7044 2017-10-05 10:07 /flink-recovery/prod/completedCheckpoint7c5a19300092

This is getting a little interesting. What say you :)

On Thu, Oct 5, 2017 at 9:26 AM, Vishal Santoshi <[hidden email]> wrote:

Another thing I noted was this thing

drwxr-xr-x   - root hadoop          0 2017-10-04 13:54 /flink-checkpoints/prod/c4af8dfa864e2f9a51764de9f0725b39/chk-44286

drwxr-xr-x   - root hadoop          0 2017-10-05 09:15 /flink-checkpoints/prod/c4af8dfa864e2f9a51764de9f0725b39/chk-45428

Generally what Flink does IMHO is that it replaces the chk point directory with a new one. I see it happening now. Every minute it replaces the old directory.  In this job's case however, it did not delete the 2017-10-04 13:54  and hence the chk-44286 directory.  This was the last chk-44286 (  I think  )  successfully created before NN had issues but as is usual did not delete this  chk-44286. It looks as if it started with a blank slate ???????? Does this strike a chord ?????

On Thu, Oct 5, 2017 at 8:56 AM, Vishal Santoshi <[hidden email]> wrote:

Hello Fabian, 

                      First of all congratulations on this fabulous framework. I have worked with GDF and though GDF has some natural pluses Flink's state management is far more advanced. With kafka as a source it negates issues GDF has ( GDF integration with pub/sub is organic and that is to be expected but non FIFO pub/sub is an issue with windows on event time etc )

                   Coming back to this issue. We have that same kafka topic feeding a streaming druid datasource and we do not see any issue there, so so data loss on the source, kafka is not applicable. I am totally certain that the "retention" time was not an issue. It is 4 days of retention and we fixed this issue within 30 minutes. We could replay kafka with a new consumer group.id and that worked fine. 

Note these properties and see if they strike a chord.

* The setCommitOffsetsOnCheckpoints(boolean) for kafka consumers is the default true. I bring this up to see whether flink will in any circumstance drive consumption on the kafka perceived offset rather than the one in the checkpoint.

* The state.backend.fs.memory-threshold: 0 has not been set.  The state is big enough though therefore IMHO no way the state is stored along with the meta data in JM ( or ZK ? ) . The reason I bring this up is to make sure when you say that the size has to be less than 1024bytes , you are talking about cumulative state of the pipeine.

* We have a good sense of SP ( save point )  and CP ( checkpoint ) and certainly understand that they actually are not dissimilar. However in this case there were multiple attempts to restart the pipe before it finally succeeded. 

* Other hdfs related poperties.

 

 state.backend.fs.checkpointdir: hdfs:///flink-checkpoints/<%= flink_hdfs_root %>

 state.savepoints.dir: hdfs:///flink-savepoints/<%= flink_hdfs_root %>

 recovery.zookeeper.storageDir: hdfs:///flink-recovery/<%= flink_hdfs_root %>

Do these make sense ? Is there anything else I should look at.  Please also note that it is the second time this has happened. The first time I was vacationing and was not privy to the state of the flink pipeline, but the net effect were similar. The counts for the first window after an internal restart dropped. 

Thank you for you patience and regards,

Vishal

On Thu, Oct 5, 2017 at 5:01 AM, Fabian Hueske <[hidden email]> wrote:

Hi Vishal,

window operators are always stateful because the operator needs to remember previously received events (WindowFunction) or intermediate results (ReduceFunction).

Given the program you described, a checkpoint should include the Kafka consumer offset and the state of the window operator. If the program eventually successfully (i.e., without an error) recovered from the last checkpoint, all its state should have been restored. Since the last checkpoint was before HDFS went into safe mode, the program would have been reset to that point. If the Kafka retention time is less than the time it took to fix HDFS you would have lost data because it would have been removed from Kafka. If that's not the case, we need to investigate this further because a checkpoint recovery must not result in state loss.

Restoring from a savepoint is not so much different from automatic checkpoint recovery. Given that you have a completed savepoint, you can restart the job from that point. The main difference is that checkpoints are only used for internal recovery and usually discarded once the job is terminated while savepoints are retained.

Regarding your question if a failed checkpoint should cause the job to fail and recover I'm not sure what the current status is.

Stefan (in CC) should know what happens if a checkpoint fails.

Best, Fabian

2017-10-05 2:20 GMT+02:00 Vishal Santoshi <[hidden email]>:

To add to it, my pipeline is a simple 

keyBy(0)
        .timeWindow(Time.of(window_size, TimeUnit.MINUTES))
        .allowedLateness(Time.of(late_by, TimeUnit.SECONDS))
        .reduce(new ReduceFunction(), new WindowFunction())

On Wed, Oct 4, 2017 at 8:19 PM, Vishal Santoshi <[hidden email]> wrote:

Hello folks,

As far as I know checkpoint failure should be ignored and retried with potentially larger state. I had this situation

* hdfs went into a safe mode b'coz of Name Node issues
* exception was thrown

    org.apache.hadoop.ipc.RemoteException(org.apache.hadoop.ipc.StandbyException): Operation category WRITE is not supported in state standby. Visit https://s.apache.org/sbnn-error
    ..................

    at org.apache.flink.runtime.fs.hdfs.HadoopFileSystem.mkdirs(HadoopFileSystem.java:453)
        at org.apache.flink.core.fs.SafetyNetWrapperFileSystem.mkdirs(SafetyNetWrapperFileSystem.java:111)
        at org.apache.flink.runtime.state.filesystem.FsCheckpointStreamFactory.createBasePath(FsCheckpointStreamFactory.java:132)

* The pipeline came back after a few restarts and checkpoint failures, after the hdfs issues were resolved.

I would not have worried about the restart, but it was evident that I lost my operator state. Either it was my kafka consumer that kept on advancing it's offset between a start and the next checkpoint failure ( a minute's worth ) or the the operator that had partial aggregates was lost. I have a 15 minute window of counts on a keyed operator

I am using ROCKS DB and of course have checkpointing turned on.

The questions thus are

* Should a pipeline be restarted if checkpoint fails ?
* Why on restart did the operator state did not recreate ?
* Is the nature of the exception thrown have to do with any of this b'coz suspend and resume from a save point work as expected ?
* And though I am pretty sure, are operators like the Window operator stateful by drfault and thus if I have timeWindow(Time.of(window_size, TimeUnit.MINUTES)).reduce(new ReduceFunction(), new WindowFunction()), the state is managed by flink ?

Thanks.

[Vishal Santoshi]

Sorry.

There are 2 scenerios

  * Idempotent Sinks Use Case where we would want to restore from the latest valid checkpoint.  If I understand the code correctly we try to  retrieve all completed checkpoints  for all handles in ZK and abort ( throw an exception ) if there are handles but no corresponding complete checkpoints in hdfs,  else we use the latest valid checkpoint state.  On abort a restart  and thus restore of the  pipe  is issued repeating the above execution. If the failure in hdfs was transient a retry will succeed else when the  retry limit is reached the pipeline is aborted for good.

* Non Idempotent Sinks where we have no retries. We do not want to recover from the last available checkpoint as the above code will do as the more  into history we go the more duplicates will be delivered. The only solution is use exactly once semantics of the source and sinks if possible.

On Wed, Jan 24, 2018 at 7:20 AM, Aljoscha Krettek <[hidden email]> wrote:

Did you see my second mail?

On 24. Jan 2018, at 12:50, Vishal Santoshi <[hidden email]> wrote:

As in, if there are chk handles in zk, there should no reason to start a new job ( bad handle, no hdfs connectivity etc ),

 yes that sums it up.

On Wed, Jan 24, 2018 at 5:35 AM, Aljoscha Krettek <[hidden email]> wrote:

Wait a sec, I just checked out the code again and it seems we already do that: https://github.com/apache/flink/blob/9071e3befb8c279f73c3094c9f6bddc0e7cce9e5/flink-runtime/src/main/java/org/apache/flink/runtime/checkpoint/ZooKeeperCompletedCheckpointStore.java#L210

If there were some checkpoints but none could be read we fail recovery.

On 24. Jan 2018, at 11:32, Aljoscha Krettek <[hidden email]> wrote:

That sounds reasonable: We would keep the first fix, i.e. never delete checkpoints if they're "corrupt", only when they're subsumed. Additionally, we fail the job if there are some checkpoints in ZooKeeper but none of them can be restored to prevent the case where a job starts from scratch even though it shouldn't.

Does that sum it up?

On 24. Jan 2018, at 01:19, Vishal Santoshi <[hidden email]> wrote:

If we hit the retry limit, abort the job. In our case we will restart from the last SP ( we as any production pile do it is n time s a day )  and that I would think should be OK for most folks ?

On Tue, Jan 23, 2018 at 11:38 AM, Vishal Santoshi <[hidden email]> wrote:

Thank you for considering this. If I understand you correctly.

* CHK pointer on ZK for a CHK state on hdfs was done successfully.

* Some issue restarted the pipeline.

* The NN was down unfortunately and flink could not retrieve the  CHK state from the CHK pointer on ZK.

Before 

* The CHK pointer was being removed and the job started from a brand new slate.

After ( this fix on 1.4 +) 

* do not delete the CHK pointer ( It has to be subsumed to be deleted ). 

* Flink keeps using this CHK pointer ( if retry is > 0 and till we hit any retry limit ) to restore state

* NN comes back 

* Flink restores state on the next retry.

I would hope that is the sequence to follow.

Regards.

On Tue, Jan 23, 2018 at 7:25 AM, Aljoscha Krettek <[hidden email]> wrote:

Hi Vishal,

I think you might be right. We fixed the problem that checkpoints where dropped via https://issues.apache.org/jira/browse/FLINK-7783. However, we still have the problem that if the DFS is not up at all then it will look as if the job is starting from scratch. However, the alternative is failing the job, in which case you will also never be able to restore from a checkpoint. What do you think?

Best,

Aljoscha

On 23. Jan 2018, at 10:15, Fabian Hueske <[hidden email]> wrote:

Sorry for the late reply.

I created FLINK-8487 [1] to track this problem

@Vishal, can you have a look and check if if forgot some details? I logged the issue for Flink 1.3.2, is that correct?

Please add more information if you think it is relevant.

Thanks,

Fabian

[1] https://issues.apache.org/jira/browse/FLINK-8487

2018-01-18 22:14 GMT+01:00 Vishal Santoshi <[hidden email]>:

Or this one 

https://issues.apache.org/jira/browse/FLINK-4815

On Thu, Jan 18, 2018 at 4:13 PM, Vishal Santoshi <[hidden email]> wrote:

ping. 

    This happened again on production and it seems reasonable to abort when a checkpoint is not found rather than behave as if it is a brand new pipeline.  

On Tue, Jan 16, 2018 at 9:33 AM, Vishal Santoshi <[hidden email]> wrote:

Folks sorry for being late on this. Can some body with the knowledge of this code base create a jira issue for the above ? We have seen this more than once on production.

On Mon, Oct 9, 2017 at 10:21 AM, Aljoscha Krettek <[hidden email]> wrote:

Hi Vishal,

Some relevant Jira issues for you are:

 - https://issues.apache.org/jira/browse/FLINK-4808: Allow skipping failed checkpoints

 - https://issues.apache.org/jira/browse/FLINK-4815: Automatic fallback to earlier checkpoint when checkpoint restore fails

 - https://issues.apache.org/jira/browse/FLINK-7783: Don't always remove checkpoints in ZooKeeperCompletedCheckpointStore#recover()

Best,

Aljoscha

On 9. Oct 2017, at 09:06, Fabian Hueske <[hidden email]> wrote:

Hi Vishal,

it would be great if you could create a JIRA ticket with Blocker priority.
Please add all relevant information of your detailed analysis, add a link to this email thread (see [1] for the web archive of the mailing list), and post the id of the JIRA issue here.

Thanks for looking into this!

Best regards,

Fabian

[1] https://lists.apache.org/list.html?user@...

2017-10-06 15:59 GMT+02:00 Vishal Santoshi <[hidden email]>:

Thank you for confirming. 

       

 I think this is a critical bug. In essence any checkpoint store ( hdfs/S3/File)  will loose state if it is unavailable at resume. This becomes all the more painful with your confirming that  "failed checkpoints killing the job"  b'coz essentially it mean that if remote store in unavailable  during checkpoint than you have lost state ( till of course you have a retry of none or an unbounded retry delay, a delay that you hope the store revives in ) .. Remember  the first retry failure  will cause new state according the code as written iff the remote store is down. We would rather have a configurable property that establishes  our desire to abort something like a "abort_retry_on_chkretrevalfailure"

In our case it is very important that we do not undercount a window, one reason we use flink and it's awesome failure guarantees, as various alarms sound ( we do anomaly detection on the time series ).

Please create a jira ticket for us to follow or we could do it.

PS Not aborting on checkpointing, till a configurable limit is very important too.

On Fri, Oct 6, 2017 at 2:36 AM, Aljoscha Krettek <[hidden email]> wrote:

Hi Vishal,

I think you're right! And thanks for looking into this so deeply. 

With your last mail your basically saying, that the checkpoint could not be restored because your HDFS was temporarily down. If Flink had not deleted that checkpoint it might have been possible to restore it at a later point, right?

Regarding failed checkpoints killing the job: yes, this is currently the expected behaviour but there are plans to change this.

Best,

Aljoscha

On 5. Oct 2017, at 17:40, Vishal Santoshi <[hidden email]> wrote:

I think this is the offending piece. There is a catch all Exception, which IMHO should understand a recoverable exception from an unrecoverable on. 

try {
	completedCheckpoint = retrieveCompletedCheckpoint(checkpointStateHandle);
	if (completedCheckpoint != null) {
	completedCheckpoints.add(completedCheckpoint);
	}
	} catch (Exception e) {
	LOG.warn("Could not retrieve checkpoint. Removing it from the completed " +
	"checkpoint store.", e);
	// remove the checkpoint with broken state handle
	removeBrokenStateHandle(checkpointStateHandle.f1, checkpointStateHandle.f0);
	}

On Thu, Oct 5, 2017 at 10:57 AM, Vishal Santoshi <[hidden email]> wrote:

So this is the issue and tell us that it is wrong. ZK had some state ( backed by hdfs ) that referred to a checkpoint ( the same exact last successful checkpoint that was successful before NN screwed us ). When the JM tried to recreate the state and b'coz NN was down failed to retrieve the CHK handle from hdfs and conveniently ( and I think very wrongly ) removed the CHK from being considered and cleaned the pointer ( though failed as was NN was down and is obvious from the dangling file in recovery ) . The metadata itself was on hdfs and failure in retrieving should have been a stop all, not going to trying doing magic exception rather than starting from a blank state.

org.apache.flink.util.FlinkException: Could not retrieve checkpoint 44286 from state handle under /0000000000000044286. This indicates that the retrieved state handle is broken. Try cleaning the state handle store.

On Thu, Oct 5, 2017 at 10:13 AM, Vishal Santoshi <[hidden email]> wrote:

Also note that  the zookeeper recovery did  ( sadly on the same hdfs cluster ) also showed the same behavior. It had the pointers to the chk point  ( I  think that is what it does, keeps metadata of where the checkpoint etc  ) .  It too decided to keep the recovery file from the failed state.

-rw-r--r--   3 root hadoop       7041 2017-10-04 13:55 /flink-recovery/prod/completedCheckpoint6c9096bb9ed4

-rw-r--r--   3 root hadoop       7044 2017-10-05 10:07 /flink-recovery/prod/completedCheckpoint7c5a19300092

This is getting a little interesting. What say you :)

On Thu, Oct 5, 2017 at 9:26 AM, Vishal Santoshi <[hidden email]> wrote:

Another thing I noted was this thing

drwxr-xr-x   - root hadoop          0 2017-10-04 13:54 /flink-checkpoints/prod/c4af8dfa864e2f9a51764de9f0725b39/chk-44286

drwxr-xr-x   - root hadoop          0 2017-10-05 09:15 /flink-checkpoints/prod/c4af8dfa864e2f9a51764de9f0725b39/chk-45428

Generally what Flink does IMHO is that it replaces the chk point directory with a new one. I see it happening now. Every minute it replaces the old directory.  In this job's case however, it did not delete the 2017-10-04 13:54  and hence the chk-44286 directory.  This was the last chk-44286 (  I think  )  successfully created before NN had issues but as is usual did not delete this  chk-44286. It looks as if it started with a blank slate ???????? Does this strike a chord ?????

On Thu, Oct 5, 2017 at 8:56 AM, Vishal Santoshi <[hidden email]> wrote:

Hello Fabian, 

                      First of all congratulations on this fabulous framework. I have worked with GDF and though GDF has some natural pluses Flink's state management is far more advanced. With kafka as a source it negates issues GDF has ( GDF integration with pub/sub is organic and that is to be expected but non FIFO pub/sub is an issue with windows on event time etc )

                   Coming back to this issue. We have that same kafka topic feeding a streaming druid datasource and we do not see any issue there, so so data loss on the source, kafka is not applicable. I am totally certain that the "retention" time was not an issue. It is 4 days of retention and we fixed this issue within 30 minutes. We could replay kafka with a new consumer group.id and that worked fine. 

Note these properties and see if they strike a chord.

* The setCommitOffsetsOnCheckpoints(boolean) for kafka consumers is the default true. I bring this up to see whether flink will in any circumstance drive consumption on the kafka perceived offset rather than the one in the checkpoint.

* The state.backend.fs.memory-threshold: 0 has not been set.  The state is big enough though therefore IMHO no way the state is stored along with the meta data in JM ( or ZK ? ) . The reason I bring this up is to make sure when you say that the size has to be less than 1024bytes , you are talking about cumulative state of the pipeine.

* We have a good sense of SP ( save point )  and CP ( checkpoint ) and certainly understand that they actually are not dissimilar. However in this case there were multiple attempts to restart the pipe before it finally succeeded. 

* Other hdfs related poperties.

 

 state.backend.fs.checkpointdir: hdfs:///flink-checkpoints/<%= flink_hdfs_root %>

 state.savepoints.dir: hdfs:///flink-savepoints/<%= flink_hdfs_root %>

 recovery.zookeeper.storageDir: hdfs:///flink-recovery/<%= flink_hdfs_root %>

Do these make sense ? Is there anything else I should look at.  Please also note that it is the second time this has happened. The first time I was vacationing and was not privy to the state of the flink pipeline, but the net effect were similar. The counts for the first window after an internal restart dropped. 

Thank you for you patience and regards,

Vishal

On Thu, Oct 5, 2017 at 5:01 AM, Fabian Hueske <[hidden email]> wrote:

Hi Vishal,

window operators are always stateful because the operator needs to remember previously received events (WindowFunction) or intermediate results (ReduceFunction).

Given the program you described, a checkpoint should include the Kafka consumer offset and the state of the window operator. If the program eventually successfully (i.e., without an error) recovered from the last checkpoint, all its state should have been restored. Since the last checkpoint was before HDFS went into safe mode, the program would have been reset to that point. If the Kafka retention time is less than the time it took to fix HDFS you would have lost data because it would have been removed from Kafka. If that's not the case, we need to investigate this further because a checkpoint recovery must not result in state loss.

Restoring from a savepoint is not so much different from automatic checkpoint recovery. Given that you have a completed savepoint, you can restart the job from that point. The main difference is that checkpoints are only used for internal recovery and usually discarded once the job is terminated while savepoints are retained.

Regarding your question if a failed checkpoint should cause the job to fail and recover I'm not sure what the current status is.

Stefan (in CC) should know what happens if a checkpoint fails.

Best, Fabian

2017-10-05 2:20 GMT+02:00 Vishal Santoshi <[hidden email]>:

To add to it, my pipeline is a simple 

keyBy(0)
        .timeWindow(Time.of(window_size, TimeUnit.MINUTES))
        .allowedLateness(Time.of(late_by, TimeUnit.SECONDS))
        .reduce(new ReduceFunction(), new WindowFunction())

On Wed, Oct 4, 2017 at 8:19 PM, Vishal Santoshi <[hidden email]> wrote:

Hello folks,

As far as I know checkpoint failure should be ignored and retried with potentially larger state. I had this situation

* hdfs went into a safe mode b'coz of Name Node issues
* exception was thrown

    org.apache.hadoop.ipc.RemoteException(org.apache.hadoop.ipc.StandbyException): Operation category WRITE is not supported in state standby. Visit https://s.apache.org/sbnn-error
    ..................

    at org.apache.flink.runtime.fs.hdfs.HadoopFileSystem.mkdirs(HadoopFileSystem.java:453)
        at org.apache.flink.core.fs.SafetyNetWrapperFileSystem.mkdirs(SafetyNetWrapperFileSystem.java:111)
        at org.apache.flink.runtime.state.filesystem.FsCheckpointStreamFactory.createBasePath(FsCheckpointStreamFactory.java:132)

* The pipeline came back after a few restarts and checkpoint failures, after the hdfs issues were resolved.

I would not have worried about the restart, but it was evident that I lost my operator state. Either it was my kafka consumer that kept on advancing it's offset between a start and the next checkpoint failure ( a minute's worth ) or the the operator that had partial aggregates was lost. I have a 15 minute window of counts on a keyed operator

I am using ROCKS DB and of course have checkpointing turned on.

The questions thus are

* Should a pipeline be restarted if checkpoint fails ?
* Why on restart did the operator state did not recreate ?
* Is the nature of the exception thrown have to do with any of this b'coz suspend and resume from a save point work as expected ?
* And though I am pretty sure, are operators like the Window operator stateful by drfault and thus if I have timeWindow(Time.of(window_size, TimeUnit.MINUTES)).reduce(new ReduceFunction(), new WindowFunction()), the state is managed by flink ?

Thanks.

[Vishal Santoshi]

To add to this, we are assuming that the default configuration will fail a pipeline if  a checkpoint fails and will hit the recover loop only and only if the retry limit is not reached 

On Thu, Jan 25, 2018 at 7:00 AM, Vishal Santoshi <[hidden email]> wrote:

Sorry.

There are 2 scenerios

  * Idempotent Sinks Use Case where we would want to restore from the latest valid checkpoint.  If I understand the code correctly we try to  retrieve all completed checkpoints  for all handles in ZK and abort ( throw an exception ) if there are handles but no corresponding complete checkpoints in hdfs,  else we use the latest valid checkpoint state.  On abort a restart  and thus restore of the  pipe  is issued repeating the above execution. If the failure in hdfs was transient a retry will succeed else when the  retry limit is reached the pipeline is aborted for good.

* Non Idempotent Sinks where we have no retries. We do not want to recover from the last available checkpoint as the above code will do as the more  into history we go the more duplicates will be delivered. The only solution is use exactly once semantics of the source and sinks if possible.

On Wed, Jan 24, 2018 at 7:20 AM, Aljoscha Krettek <[hidden email]> wrote:

Did you see my second mail?

On 24. Jan 2018, at 12:50, Vishal Santoshi <[hidden email]> wrote:

As in, if there are chk handles in zk, there should no reason to start a new job ( bad handle, no hdfs connectivity etc ),

 yes that sums it up.

On Wed, Jan 24, 2018 at 5:35 AM, Aljoscha Krettek <[hidden email]> wrote:

Wait a sec, I just checked out the code again and it seems we already do that: https://github.com/apache/flink/blob/9071e3befb8c279f73c3094c9f6bddc0e7cce9e5/flink-runtime/src/main/java/org/apache/flink/runtime/checkpoint/ZooKeeperCompletedCheckpointStore.java#L210

If there were some checkpoints but none could be read we fail recovery.

On 24. Jan 2018, at 11:32, Aljoscha Krettek <[hidden email]> wrote:

That sounds reasonable: We would keep the first fix, i.e. never delete checkpoints if they're "corrupt", only when they're subsumed. Additionally, we fail the job if there are some checkpoints in ZooKeeper but none of them can be restored to prevent the case where a job starts from scratch even though it shouldn't.

Does that sum it up?

On 24. Jan 2018, at 01:19, Vishal Santoshi <[hidden email]> wrote:

If we hit the retry limit, abort the job. In our case we will restart from the last SP ( we as any production pile do it is n time s a day )  and that I would think should be OK for most folks ?

On Tue, Jan 23, 2018 at 11:38 AM, Vishal Santoshi <[hidden email]> wrote:

Thank you for considering this. If I understand you correctly.

* CHK pointer on ZK for a CHK state on hdfs was done successfully.

* Some issue restarted the pipeline.

* The NN was down unfortunately and flink could not retrieve the  CHK state from the CHK pointer on ZK.

Before 

* The CHK pointer was being removed and the job started from a brand new slate.

After ( this fix on 1.4 +) 

* do not delete the CHK pointer ( It has to be subsumed to be deleted ). 

* Flink keeps using this CHK pointer ( if retry is > 0 and till we hit any retry limit ) to restore state

* NN comes back 

* Flink restores state on the next retry.

I would hope that is the sequence to follow.

Regards.

On Tue, Jan 23, 2018 at 7:25 AM, Aljoscha Krettek <[hidden email]> wrote:

Hi Vishal,

I think you might be right. We fixed the problem that checkpoints where dropped via https://issues.apache.org/jira/browse/FLINK-7783. However, we still have the problem that if the DFS is not up at all then it will look as if the job is starting from scratch. However, the alternative is failing the job, in which case you will also never be able to restore from a checkpoint. What do you think?

Best,

Aljoscha

On 23. Jan 2018, at 10:15, Fabian Hueske <[hidden email]> wrote:

Sorry for the late reply.

I created FLINK-8487 [1] to track this problem

@Vishal, can you have a look and check if if forgot some details? I logged the issue for Flink 1.3.2, is that correct?

Please add more information if you think it is relevant.

Thanks,

Fabian

[1] https://issues.apache.org/jira/browse/FLINK-8487

2018-01-18 22:14 GMT+01:00 Vishal Santoshi <[hidden email]>:

Or this one 

https://issues.apache.org/jira/browse/FLINK-4815

On Thu, Jan 18, 2018 at 4:13 PM, Vishal Santoshi <[hidden email]> wrote:

ping. 

    This happened again on production and it seems reasonable to abort when a checkpoint is not found rather than behave as if it is a brand new pipeline.  

On Tue, Jan 16, 2018 at 9:33 AM, Vishal Santoshi <[hidden email]> wrote:

Folks sorry for being late on this. Can some body with the knowledge of this code base create a jira issue for the above ? We have seen this more than once on production.

On Mon, Oct 9, 2017 at 10:21 AM, Aljoscha Krettek <[hidden email]> wrote:

Hi Vishal,

Some relevant Jira issues for you are:

 - https://issues.apache.org/jira/browse/FLINK-4808: Allow skipping failed checkpoints

 - https://issues.apache.org/jira/browse/FLINK-4815: Automatic fallback to earlier checkpoint when checkpoint restore fails

 - https://issues.apache.org/jira/browse/FLINK-7783: Don't always remove checkpoints in ZooKeeperCompletedCheckpointStore#recover()

Best,

Aljoscha

On 9. Oct 2017, at 09:06, Fabian Hueske <[hidden email]> wrote:

Hi Vishal,

it would be great if you could create a JIRA ticket with Blocker priority.
Please add all relevant information of your detailed analysis, add a link to this email thread (see [1] for the web archive of the mailing list), and post the id of the JIRA issue here.

Thanks for looking into this!

Best regards,

Fabian

[1] https://lists.apache.org/list.html?user@...

2017-10-06 15:59 GMT+02:00 Vishal Santoshi <[hidden email]>:

Thank you for confirming. 

       

 I think this is a critical bug. In essence any checkpoint store ( hdfs/S3/File)  will loose state if it is unavailable at resume. This becomes all the more painful with your confirming that  "failed checkpoints killing the job"  b'coz essentially it mean that if remote store in unavailable  during checkpoint than you have lost state ( till of course you have a retry of none or an unbounded retry delay, a delay that you hope the store revives in ) .. Remember  the first retry failure  will cause new state according the code as written iff the remote store is down. We would rather have a configurable property that establishes  our desire to abort something like a "abort_retry_on_chkretrevalfailure"

In our case it is very important that we do not undercount a window, one reason we use flink and it's awesome failure guarantees, as various alarms sound ( we do anomaly detection on the time series ).

Please create a jira ticket for us to follow or we could do it.

PS Not aborting on checkpointing, till a configurable limit is very important too.

On Fri, Oct 6, 2017 at 2:36 AM, Aljoscha Krettek <[hidden email]> wrote:

Hi Vishal,

I think you're right! And thanks for looking into this so deeply. 

With your last mail your basically saying, that the checkpoint could not be restored because your HDFS was temporarily down. If Flink had not deleted that checkpoint it might have been possible to restore it at a later point, right?

Regarding failed checkpoints killing the job: yes, this is currently the expected behaviour but there are plans to change this.

Best,

Aljoscha

On 5. Oct 2017, at 17:40, Vishal Santoshi <[hidden email]> wrote:

I think this is the offending piece. There is a catch all Exception, which IMHO should understand a recoverable exception from an unrecoverable on. 

try {
	completedCheckpoint = retrieveCompletedCheckpoint(checkpointStateHandle);
	if (completedCheckpoint != null) {
	completedCheckpoints.add(completedCheckpoint);
	}
	} catch (Exception e) {
	LOG.warn("Could not retrieve checkpoint. Removing it from the completed " +
	"checkpoint store.", e);
	// remove the checkpoint with broken state handle
	removeBrokenStateHandle(checkpointStateHandle.f1, checkpointStateHandle.f0);
	}

On Thu, Oct 5, 2017 at 10:57 AM, Vishal Santoshi <[hidden email]> wrote:

So this is the issue and tell us that it is wrong. ZK had some state ( backed by hdfs ) that referred to a checkpoint ( the same exact last successful checkpoint that was successful before NN screwed us ). When the JM tried to recreate the state and b'coz NN was down failed to retrieve the CHK handle from hdfs and conveniently ( and I think very wrongly ) removed the CHK from being considered and cleaned the pointer ( though failed as was NN was down and is obvious from the dangling file in recovery ) . The metadata itself was on hdfs and failure in retrieving should have been a stop all, not going to trying doing magic exception rather than starting from a blank state.

org.apache.flink.util.FlinkException: Could not retrieve checkpoint 44286 from state handle under /0000000000000044286. This indicates that the retrieved state handle is broken. Try cleaning the state handle store.

On Thu, Oct 5, 2017 at 10:13 AM, Vishal Santoshi <[hidden email]> wrote:

Also note that  the zookeeper recovery did  ( sadly on the same hdfs cluster ) also showed the same behavior. It had the pointers to the chk point  ( I  think that is what it does, keeps metadata of where the checkpoint etc  ) .  It too decided to keep the recovery file from the failed state.

-rw-r--r--   3 root hadoop       7041 2017-10-04 13:55 /flink-recovery/prod/completedCheckpoint6c9096bb9ed4

-rw-r--r--   3 root hadoop       7044 2017-10-05 10:07 /flink-recovery/prod/completedCheckpoint7c5a19300092

This is getting a little interesting. What say you :)

On Thu, Oct 5, 2017 at 9:26 AM, Vishal Santoshi <[hidden email]> wrote:

Another thing I noted was this thing

drwxr-xr-x   - root hadoop          0 2017-10-04 13:54 /flink-checkpoints/prod/c4af8dfa864e2f9a51764de9f0725b39/chk-44286

drwxr-xr-x   - root hadoop          0 2017-10-05 09:15 /flink-checkpoints/prod/c4af8dfa864e2f9a51764de9f0725b39/chk-45428

Generally what Flink does IMHO is that it replaces the chk point directory with a new one. I see it happening now. Every minute it replaces the old directory.  In this job's case however, it did not delete the 2017-10-04 13:54  and hence the chk-44286 directory.  This was the last chk-44286 (  I think  )  successfully created before NN had issues but as is usual did not delete this  chk-44286. It looks as if it started with a blank slate ???????? Does this strike a chord ?????

On Thu, Oct 5, 2017 at 8:56 AM, Vishal Santoshi <[hidden email]> wrote:

Hello Fabian, 

                      First of all congratulations on this fabulous framework. I have worked with GDF and though GDF has some natural pluses Flink's state management is far more advanced. With kafka as a source it negates issues GDF has ( GDF integration with pub/sub is organic and that is to be expected but non FIFO pub/sub is an issue with windows on event time etc )

                   Coming back to this issue. We have that same kafka topic feeding a streaming druid datasource and we do not see any issue there, so so data loss on the source, kafka is not applicable. I am totally certain that the "retention" time was not an issue. It is 4 days of retention and we fixed this issue within 30 minutes. We could replay kafka with a new consumer group.id and that worked fine. 

Note these properties and see if they strike a chord.

* The setCommitOffsetsOnCheckpoints(boolean) for kafka consumers is the default true. I bring this up to see whether flink will in any circumstance drive consumption on the kafka perceived offset rather than the one in the checkpoint.

* The state.backend.fs.memory-threshold: 0 has not been set.  The state is big enough though therefore IMHO no way the state is stored along with the meta data in JM ( or ZK ? ) . The reason I bring this up is to make sure when you say that the size has to be less than 1024bytes , you are talking about cumulative state of the pipeine.

* We have a good sense of SP ( save point )  and CP ( checkpoint ) and certainly understand that they actually are not dissimilar. However in this case there were multiple attempts to restart the pipe before it finally succeeded. 

* Other hdfs related poperties.

 

 state.backend.fs.checkpointdir: hdfs:///flink-checkpoints/<%= flink_hdfs_root %>

 state.savepoints.dir: hdfs:///flink-savepoints/<%= flink_hdfs_root %>

 recovery.zookeeper.storageDir: hdfs:///flink-recovery/<%= flink_hdfs_root %>

Do these make sense ? Is there anything else I should look at.  Please also note that it is the second time this has happened. The first time I was vacationing and was not privy to the state of the flink pipeline, but the net effect were similar. The counts for the first window after an internal restart dropped. 

Thank you for you patience and regards,

Vishal

On Thu, Oct 5, 2017 at 5:01 AM, Fabian Hueske <[hidden email]> wrote:

Hi Vishal,

window operators are always stateful because the operator needs to remember previously received events (WindowFunction) or intermediate results (ReduceFunction).

Given the program you described, a checkpoint should include the Kafka consumer offset and the state of the window operator. If the program eventually successfully (i.e., without an error) recovered from the last checkpoint, all its state should have been restored. Since the last checkpoint was before HDFS went into safe mode, the program would have been reset to that point. If the Kafka retention time is less than the time it took to fix HDFS you would have lost data because it would have been removed from Kafka. If that's not the case, we need to investigate this further because a checkpoint recovery must not result in state loss.

Restoring from a savepoint is not so much different from automatic checkpoint recovery. Given that you have a completed savepoint, you can restart the job from that point. The main difference is that checkpoints are only used for internal recovery and usually discarded once the job is terminated while savepoints are retained.

Regarding your question if a failed checkpoint should cause the job to fail and recover I'm not sure what the current status is.

Stefan (in CC) should know what happens if a checkpoint fails.

Best, Fabian

2017-10-05 2:20 GMT+02:00 Vishal Santoshi <[hidden email]>:

To add to it, my pipeline is a simple 

keyBy(0)
        .timeWindow(Time.of(window_size, TimeUnit.MINUTES))
        .allowedLateness(Time.of(late_by, TimeUnit.SECONDS))
        .reduce(new ReduceFunction(), new WindowFunction())

On Wed, Oct 4, 2017 at 8:19 PM, Vishal Santoshi <[hidden email]> wrote:

Hello folks,

As far as I know checkpoint failure should be ignored and retried with potentially larger state. I had this situation

* hdfs went into a safe mode b'coz of Name Node issues
* exception was thrown

    org.apache.hadoop.ipc.RemoteException(org.apache.hadoop.ipc.StandbyException): Operation category WRITE is not supported in state standby. Visit https://s.apache.org/sbnn-error
    ..................

    at org.apache.flink.runtime.fs.hdfs.HadoopFileSystem.mkdirs(HadoopFileSystem.java:453)
        at org.apache.flink.core.fs.SafetyNetWrapperFileSystem.mkdirs(SafetyNetWrapperFileSystem.java:111)
        at org.apache.flink.runtime.state.filesystem.FsCheckpointStreamFactory.createBasePath(FsCheckpointStreamFactory.java:132)

* The pipeline came back after a few restarts and checkpoint failures, after the hdfs issues were resolved.

I would not have worried about the restart, but it was evident that I lost my operator state. Either it was my kafka consumer that kept on advancing it's offset between a start and the next checkpoint failure ( a minute's worth ) or the the operator that had partial aggregates was lost. I have a 15 minute window of counts on a keyed operator

I am using ROCKS DB and of course have checkpointing turned on.

The questions thus are

* Should a pipeline be restarted if checkpoint fails ?
* Why on restart did the operator state did not recreate ?
* Is the nature of the exception thrown have to do with any of this b'coz suspend and resume from a save point work as expected ?
* And though I am pretty sure, are operators like the Window operator stateful by drfault and thus if I have timeWindow(Time.of(window_size, TimeUnit.MINUTES)).reduce(new ReduceFunction(), new WindowFunction()), the state is managed by flink ?

Thanks.

[Stephan Ewen]

The assumption in your previous mail is correct.

Just to double check:

  - The initially affected version you were running was 1.3.2, correct?

The issue should be fixed in all active branches (1.4, 1.5, 1.6) and additional in 1.3.

Currently released versions with this fix: 1.4.0, 1.4.1

1.5.0 is in the makings.

We are looking to create a dedicated 1.3.3 for this fix. 

On Thu, Jan 25, 2018 at 5:13 PM, Vishal Santoshi <[hidden email]> wrote:

To add to this, we are assuming that the default configuration will fail a pipeline if  a checkpoint fails and will hit the recover loop only and only if the retry limit is not reached 

On Thu, Jan 25, 2018 at 7:00 AM, Vishal Santoshi <[hidden email]> wrote:

Sorry.

There are 2 scenerios

  * Idempotent Sinks Use Case where we would want to restore from the latest valid checkpoint.  If I understand the code correctly we try to  retrieve all completed checkpoints  for all handles in ZK and abort ( throw an exception ) if there are handles but no corresponding complete checkpoints in hdfs,  else we use the latest valid checkpoint state.  On abort a restart  and thus restore of the  pipe  is issued repeating the above execution. If the failure in hdfs was transient a retry will succeed else when the  retry limit is reached the pipeline is aborted for good.

* Non Idempotent Sinks where we have no retries. We do not want to recover from the last available checkpoint as the above code will do as the more  into history we go the more duplicates will be delivered. The only solution is use exactly once semantics of the source and sinks if possible.

On Wed, Jan 24, 2018 at 7:20 AM, Aljoscha Krettek <[hidden email]> wrote:

Did you see my second mail?

On 24. Jan 2018, at 12:50, Vishal Santoshi <[hidden email]> wrote:

As in, if there are chk handles in zk, there should no reason to start a new job ( bad handle, no hdfs connectivity etc ),

 yes that sums it up.

On Wed, Jan 24, 2018 at 5:35 AM, Aljoscha Krettek <[hidden email]> wrote:

Wait a sec, I just checked out the code again and it seems we already do that: https://github.com/apache/flink/blob/9071e3befb8c279f73c3094c9f6bddc0e7cce9e5/flink-runtime/src/main/java/org/apache/flink/runtime/checkpoint/ZooKeeperCompletedCheckpointStore.java#L210

If there were some checkpoints but none could be read we fail recovery.

On 24. Jan 2018, at 11:32, Aljoscha Krettek <[hidden email]> wrote:

That sounds reasonable: We would keep the first fix, i.e. never delete checkpoints if they're "corrupt", only when they're subsumed. Additionally, we fail the job if there are some checkpoints in ZooKeeper but none of them can be restored to prevent the case where a job starts from scratch even though it shouldn't.

Does that sum it up?

On 24. Jan 2018, at 01:19, Vishal Santoshi <[hidden email]> wrote:

If we hit the retry limit, abort the job. In our case we will restart from the last SP ( we as any production pile do it is n time s a day )  and that I would think should be OK for most folks ?

On Tue, Jan 23, 2018 at 11:38 AM, Vishal Santoshi <[hidden email]> wrote:

Thank you for considering this. If I understand you correctly.

* CHK pointer on ZK for a CHK state on hdfs was done successfully.

* Some issue restarted the pipeline.

* The NN was down unfortunately and flink could not retrieve the  CHK state from the CHK pointer on ZK.

Before 

* The CHK pointer was being removed and the job started from a brand new slate.

After ( this fix on 1.4 +) 

* do not delete the CHK pointer ( It has to be subsumed to be deleted ). 

* Flink keeps using this CHK pointer ( if retry is > 0 and till we hit any retry limit ) to restore state

* NN comes back 

* Flink restores state on the next retry.

I would hope that is the sequence to follow.

Regards.

On Tue, Jan 23, 2018 at 7:25 AM, Aljoscha Krettek <[hidden email]> wrote:

Hi Vishal,

I think you might be right. We fixed the problem that checkpoints where dropped via https://issues.apache.org/jira/browse/FLINK-7783. However, we still have the problem that if the DFS is not up at all then it will look as if the job is starting from scratch. However, the alternative is failing the job, in which case you will also never be able to restore from a checkpoint. What do you think?

Best,

Aljoscha

On 23. Jan 2018, at 10:15, Fabian Hueske <[hidden email]> wrote:

Sorry for the late reply.

I created FLINK-8487 [1] to track this problem

@Vishal, can you have a look and check if if forgot some details? I logged the issue for Flink 1.3.2, is that correct?

Please add more information if you think it is relevant.

Thanks,

Fabian

[1] https://issues.apache.org/jira/browse/FLINK-8487

2018-01-18 22:14 GMT+01:00 Vishal Santoshi <[hidden email]>:

Or this one 

https://issues.apache.org/jira/browse/FLINK-4815

On Thu, Jan 18, 2018 at 4:13 PM, Vishal Santoshi <[hidden email]> wrote:

ping. 

    This happened again on production and it seems reasonable to abort when a checkpoint is not found rather than behave as if it is a brand new pipeline.  

On Tue, Jan 16, 2018 at 9:33 AM, Vishal Santoshi <[hidden email]> wrote:

Folks sorry for being late on this. Can some body with the knowledge of this code base create a jira issue for the above ? We have seen this more than once on production.

On Mon, Oct 9, 2017 at 10:21 AM, Aljoscha Krettek <[hidden email]> wrote:

Hi Vishal,

Some relevant Jira issues for you are:

 - https://issues.apache.org/jira/browse/FLINK-4808: Allow skipping failed checkpoints

 - https://issues.apache.org/jira/browse/FLINK-4815: Automatic fallback to earlier checkpoint when checkpoint restore fails

 - https://issues.apache.org/jira/browse/FLINK-7783: Don't always remove checkpoints in ZooKeeperCompletedCheckpointStore#recover()

Best,

Aljoscha

On 9. Oct 2017, at 09:06, Fabian Hueske <[hidden email]> wrote:

Hi Vishal,

it would be great if you could create a JIRA ticket with Blocker priority.
Please add all relevant information of your detailed analysis, add a link to this email thread (see [1] for the web archive of the mailing list), and post the id of the JIRA issue here.

Thanks for looking into this!

Best regards,

Fabian

[1] https://lists.apache.org/list.html?user@...

2017-10-06 15:59 GMT+02:00 Vishal Santoshi <[hidden email]>:

Thank you for confirming. 

       

 I think this is a critical bug. In essence any checkpoint store ( hdfs/S3/File)  will loose state if it is unavailable at resume. This becomes all the more painful with your confirming that  "failed checkpoints killing the job"  b'coz essentially it mean that if remote store in unavailable  during checkpoint than you have lost state ( till of course you have a retry of none or an unbounded retry delay, a delay that you hope the store revives in ) .. Remember  the first retry failure  will cause new state according the code as written iff the remote store is down. We would rather have a configurable property that establishes  our desire to abort something like a "abort_retry_on_chkretrevalfailure"

In our case it is very important that we do not undercount a window, one reason we use flink and it's awesome failure guarantees, as various alarms sound ( we do anomaly detection on the time series ).

Please create a jira ticket for us to follow or we could do it.

PS Not aborting on checkpointing, till a configurable limit is very important too.

On Fri, Oct 6, 2017 at 2:36 AM, Aljoscha Krettek <[hidden email]> wrote:

Hi Vishal,

I think you're right! And thanks for looking into this so deeply. 

With your last mail your basically saying, that the checkpoint could not be restored because your HDFS was temporarily down. If Flink had not deleted that checkpoint it might have been possible to restore it at a later point, right?

Regarding failed checkpoints killing the job: yes, this is currently the expected behaviour but there are plans to change this.

Best,

Aljoscha

On 5. Oct 2017, at 17:40, Vishal Santoshi <[hidden email]> wrote:

I think this is the offending piece. There is a catch all Exception, which IMHO should understand a recoverable exception from an unrecoverable on. 

try {
	completedCheckpoint = retrieveCompletedCheckpoint(checkpointStateHandle);
	if (completedCheckpoint != null) {
	completedCheckpoints.add(completedCheckpoint);
	}
	} catch (Exception e) {
	LOG.warn("Could not retrieve checkpoint. Removing it from the completed " +
	"checkpoint store.", e);
	// remove the checkpoint with broken state handle
	removeBrokenStateHandle(checkpointStateHandle.f1, checkpointStateHandle.f0);
	}

On Thu, Oct 5, 2017 at 10:57 AM, Vishal Santoshi <[hidden email]> wrote:

So this is the issue and tell us that it is wrong. ZK had some state ( backed by hdfs ) that referred to a checkpoint ( the same exact last successful checkpoint that was successful before NN screwed us ). When the JM tried to recreate the state and b'coz NN was down failed to retrieve the CHK handle from hdfs and conveniently ( and I think very wrongly ) removed the CHK from being considered and cleaned the pointer ( though failed as was NN was down and is obvious from the dangling file in recovery ) . The metadata itself was on hdfs and failure in retrieving should have been a stop all, not going to trying doing magic exception rather than starting from a blank state.

org.apache.flink.util.FlinkException: Could not retrieve checkpoint 44286 from state handle under /0000000000000044286. This indicates that the retrieved state handle is broken. Try cleaning the state handle store.

On Thu, Oct 5, 2017 at 10:13 AM, Vishal Santoshi <[hidden email]> wrote:

Also note that  the zookeeper recovery did  ( sadly on the same hdfs cluster ) also showed the same behavior. It had the pointers to the chk point  ( I  think that is what it does, keeps metadata of where the checkpoint etc  ) .  It too decided to keep the recovery file from the failed state.

-rw-r--r--   3 root hadoop       7041 2017-10-04 13:55 /flink-recovery/prod/completedCheckpoint6c9096bb9ed4

-rw-r--r--   3 root hadoop       7044 2017-10-05 10:07 /flink-recovery/prod/completedCheckpoint7c5a19300092

This is getting a little interesting. What say you :)

On Thu, Oct 5, 2017 at 9:26 AM, Vishal Santoshi <[hidden email]> wrote:

Another thing I noted was this thing

drwxr-xr-x   - root hadoop          0 2017-10-04 13:54 /flink-checkpoints/prod/c4af8dfa864e2f9a51764de9f0725b39/chk-44286

drwxr-xr-x   - root hadoop          0 2017-10-05 09:15 /flink-checkpoints/prod/c4af8dfa864e2f9a51764de9f0725b39/chk-45428

Generally what Flink does IMHO is that it replaces the chk point directory with a new one. I see it happening now. Every minute it replaces the old directory.  In this job's case however, it did not delete the 2017-10-04 13:54  and hence the chk-44286 directory.  This was the last chk-44286 (  I think  )  successfully created before NN had issues but as is usual did not delete this  chk-44286. It looks as if it started with a blank slate ???????? Does this strike a chord ?????

On Thu, Oct 5, 2017 at 8:56 AM, Vishal Santoshi <[hidden email]> wrote:

Hello Fabian, 

                      First of all congratulations on this fabulous framework. I have worked with GDF and though GDF has some natural pluses Flink's state management is far more advanced. With kafka as a source it negates issues GDF has ( GDF integration with pub/sub is organic and that is to be expected but non FIFO pub/sub is an issue with windows on event time etc )

                   Coming back to this issue. We have that same kafka topic feeding a streaming druid datasource and we do not see any issue there, so so data loss on the source, kafka is not applicable. I am totally certain that the "retention" time was not an issue. It is 4 days of retention and we fixed this issue within 30 minutes. We could replay kafka with a new consumer group.id and that worked fine. 

Note these properties and see if they strike a chord.

* The setCommitOffsetsOnCheckpoints(boolean) for kafka consumers is the default true. I bring this up to see whether flink will in any circumstance drive consumption on the kafka perceived offset rather than the one in the checkpoint.

* The state.backend.fs.memory-threshold: 0 has not been set.  The state is big enough though therefore IMHO no way the state is stored along with the meta data in JM ( or ZK ? ) . The reason I bring this up is to make sure when you say that the size has to be less than 1024bytes , you are talking about cumulative state of the pipeine.

* We have a good sense of SP ( save point )  and CP ( checkpoint ) and certainly understand that they actually are not dissimilar. However in this case there were multiple attempts to restart the pipe before it finally succeeded. 

* Other hdfs related poperties.

 

 state.backend.fs.checkpointdir: hdfs:///flink-checkpoints/<%= flink_hdfs_root %>

 state.savepoints.dir: hdfs:///flink-savepoints/<%= flink_hdfs_root %>

 recovery.zookeeper.storageDir: hdfs:///flink-recovery/<%= flink_hdfs_root %>

Do these make sense ? Is there anything else I should look at.  Please also note that it is the second time this has happened. The first time I was vacationing and was not privy to the state of the flink pipeline, but the net effect were similar. The counts for the first window after an internal restart dropped. 

Thank you for you patience and regards,

Vishal

On Thu, Oct 5, 2017 at 5:01 AM, Fabian Hueske <[hidden email]> wrote:

Hi Vishal,

window operators are always stateful because the operator needs to remember previously received events (WindowFunction) or intermediate results (ReduceFunction).

Given the program you described, a checkpoint should include the Kafka consumer offset and the state of the window operator. If the program eventually successfully (i.e., without an error) recovered from the last checkpoint, all its state should have been restored. Since the last checkpoint was before HDFS went into safe mode, the program would have been reset to that point. If the Kafka retention time is less than the time it took to fix HDFS you would have lost data because it would have been removed from Kafka. If that's not the case, we need to investigate this further because a checkpoint recovery must not result in state loss.

Restoring from a savepoint is not so much different from automatic checkpoint recovery. Given that you have a completed savepoint, you can restart the job from that point. The main difference is that checkpoints are only used for internal recovery and usually discarded once the job is terminated while savepoints are retained.

Regarding your question if a failed checkpoint should cause the job to fail and recover I'm not sure what the current status is.

Stefan (in CC) should know what happens if a checkpoint fails.

Best, Fabian

2017-10-05 2:20 GMT+02:00 Vishal Santoshi <[hidden email]>:

To add to it, my pipeline is a simple 

keyBy(0)
        .timeWindow(Time.of(window_size, TimeUnit.MINUTES))
        .allowedLateness(Time.of(late_by, TimeUnit.SECONDS))
        .reduce(new ReduceFunction(), new WindowFunction())

On Wed, Oct 4, 2017 at 8:19 PM, Vishal Santoshi <[hidden email]> wrote:

Hello folks,

As far as I know checkpoint failure should be ignored and retried with potentially larger state. I had this situation

* hdfs went into a safe mode b'coz of Name Node issues
* exception was thrown

    org.apache.hadoop.ipc.RemoteException(org.apache.hadoop.ipc.StandbyException): Operation category WRITE is not supported in state standby. Visit https://s.apache.org/sbnn-error
    ..................

    at org.apache.flink.runtime.fs.hdfs.HadoopFileSystem.mkdirs(HadoopFileSystem.java:453)
        at org.apache.flink.core.fs.SafetyNetWrapperFileSystem.mkdirs(SafetyNetWrapperFileSystem.java:111)
        at org.apache.flink.runtime.state.filesystem.FsCheckpointStreamFactory.createBasePath(FsCheckpointStreamFactory.java:132)

* The pipeline came back after a few restarts and checkpoint failures, after the hdfs issues were resolved.

I would not have worried about the restart, but it was evident that I lost my operator state. Either it was my kafka consumer that kept on advancing it's offset between a start and the next checkpoint failure ( a minute's worth ) or the the operator that had partial aggregates was lost. I have a 15 minute window of counts on a keyed operator

I am using ROCKS DB and of course have checkpointing turned on.

The questions thus are

* Should a pipeline be restarted if checkpoint fails ?
* Why on restart did the operator state did not recreate ?
* Is the nature of the exception thrown have to do with any of this b'coz suspend and resume from a save point work as expected ?
* And though I am pretty sure, are operators like the Window operator stateful by drfault and thus if I have timeWindow(Time.of(window_size, TimeUnit.MINUTES)).reduce(new ReduceFunction(), new WindowFunction()), the state is managed by flink ?

Thanks.

[Vishal Santoshi]

Yes. We have not hit the snag in 1.4.0 ( our current version ).  Again though this occurs  under sustained down time on hadoop and it has been more stable lately :) 

On Wed, Mar 7, 2018 at 4:09 PM, Stephan Ewen <[hidden email]> wrote:

The assumption in your previous mail is correct.

Just to double check:

  - The initially affected version you were running was 1.3.2, correct?

The issue should be fixed in all active branches (1.4, 1.5, 1.6) and additional in 1.3.

Currently released versions with this fix: 1.4.0, 1.4.1

1.5.0 is in the makings.

We are looking to create a dedicated 1.3.3 for this fix. 

On Thu, Jan 25, 2018 at 5:13 PM, Vishal Santoshi <[hidden email]> wrote:

To add to this, we are assuming that the default configuration will fail a pipeline if  a checkpoint fails and will hit the recover loop only and only if the retry limit is not reached 

On Thu, Jan 25, 2018 at 7:00 AM, Vishal Santoshi <[hidden email]> wrote:

Sorry.

There are 2 scenerios

  * Idempotent Sinks Use Case where we would want to restore from the latest valid checkpoint.  If I understand the code correctly we try to  retrieve all completed checkpoints  for all handles in ZK and abort ( throw an exception ) if there are handles but no corresponding complete checkpoints in hdfs,  else we use the latest valid checkpoint state.  On abort a restart  and thus restore of the  pipe  is issued repeating the above execution. If the failure in hdfs was transient a retry will succeed else when the  retry limit is reached the pipeline is aborted for good.

* Non Idempotent Sinks where we have no retries. We do not want to recover from the last available checkpoint as the above code will do as the more  into history we go the more duplicates will be delivered. The only solution is use exactly once semantics of the source and sinks if possible.

On Wed, Jan 24, 2018 at 7:20 AM, Aljoscha Krettek <[hidden email]> wrote:

Did you see my second mail?

On 24. Jan 2018, at 12:50, Vishal Santoshi <[hidden email]> wrote:

As in, if there are chk handles in zk, there should no reason to start a new job ( bad handle, no hdfs connectivity etc ),

 yes that sums it up.

On Wed, Jan 24, 2018 at 5:35 AM, Aljoscha Krettek <[hidden email]> wrote:

Wait a sec, I just checked out the code again and it seems we already do that: https://github.com/apache/flink/blob/9071e3befb8c279f73c3094c9f6bddc0e7cce9e5/flink-runtime/src/main/java/org/apache/flink/runtime/checkpoint/ZooKeeperCompletedCheckpointStore.java#L210

If there were some checkpoints but none could be read we fail recovery.

On 24. Jan 2018, at 11:32, Aljoscha Krettek <[hidden email]> wrote:

That sounds reasonable: We would keep the first fix, i.e. never delete checkpoints if they're "corrupt", only when they're subsumed. Additionally, we fail the job if there are some checkpoints in ZooKeeper but none of them can be restored to prevent the case where a job starts from scratch even though it shouldn't.

Does that sum it up?

On 24. Jan 2018, at 01:19, Vishal Santoshi <[hidden email]> wrote:

If we hit the retry limit, abort the job. In our case we will restart from the last SP ( we as any production pile do it is n time s a day )  and that I would think should be OK for most folks ?

On Tue, Jan 23, 2018 at 11:38 AM, Vishal Santoshi <[hidden email]> wrote:

Thank you for considering this. If I understand you correctly.

* CHK pointer on ZK for a CHK state on hdfs was done successfully.

* Some issue restarted the pipeline.

* The NN was down unfortunately and flink could not retrieve the  CHK state from the CHK pointer on ZK.

Before 

* The CHK pointer was being removed and the job started from a brand new slate.

After ( this fix on 1.4 +) 

* do not delete the CHK pointer ( It has to be subsumed to be deleted ). 

* Flink keeps using this CHK pointer ( if retry is > 0 and till we hit any retry limit ) to restore state

* NN comes back 

* Flink restores state on the next retry.

I would hope that is the sequence to follow.

Regards.

On Tue, Jan 23, 2018 at 7:25 AM, Aljoscha Krettek <[hidden email]> wrote:

Hi Vishal,

I think you might be right. We fixed the problem that checkpoints where dropped via https://issues.apache.org/jira/browse/FLINK-7783. However, we still have the problem that if the DFS is not up at all then it will look as if the job is starting from scratch. However, the alternative is failing the job, in which case you will also never be able to restore from a checkpoint. What do you think?

Best,

Aljoscha

On 23. Jan 2018, at 10:15, Fabian Hueske <[hidden email]> wrote:

Sorry for the late reply.

I created FLINK-8487 [1] to track this problem

@Vishal, can you have a look and check if if forgot some details? I logged the issue for Flink 1.3.2, is that correct?

Please add more information if you think it is relevant.

Thanks,

Fabian

[1] https://issues.apache.org/jira/browse/FLINK-8487

2018-01-18 22:14 GMT+01:00 Vishal Santoshi <[hidden email]>:

Or this one 

https://issues.apache.org/jira/browse/FLINK-4815

On Thu, Jan 18, 2018 at 4:13 PM, Vishal Santoshi <[hidden email]> wrote:

ping. 

    This happened again on production and it seems reasonable to abort when a checkpoint is not found rather than behave as if it is a brand new pipeline.  

On Tue, Jan 16, 2018 at 9:33 AM, Vishal Santoshi <[hidden email]> wrote:

Folks sorry for being late on this. Can some body with the knowledge of this code base create a jira issue for the above ? We have seen this more than once on production.

On Mon, Oct 9, 2017 at 10:21 AM, Aljoscha Krettek <[hidden email]> wrote:

Hi Vishal,

Some relevant Jira issues for you are:

 - https://issues.apache.org/jira/browse/FLINK-4808: Allow skipping failed checkpoints

 - https://issues.apache.org/jira/browse/FLINK-4815: Automatic fallback to earlier checkpoint when checkpoint restore fails

 - https://issues.apache.org/jira/browse/FLINK-7783: Don't always remove checkpoints in ZooKeeperCompletedCheckpointStore#recover()

Best,

Aljoscha

On 9. Oct 2017, at 09:06, Fabian Hueske <[hidden email]> wrote:

Hi Vishal,

it would be great if you could create a JIRA ticket with Blocker priority.
Please add all relevant information of your detailed analysis, add a link to this email thread (see [1] for the web archive of the mailing list), and post the id of the JIRA issue here.

Thanks for looking into this!

Best regards,

Fabian

[1] https://lists.apache.org/list.html?user@...

2017-10-06 15:59 GMT+02:00 Vishal Santoshi <[hidden email]>:

Thank you for confirming. 

       

 I think this is a critical bug. In essence any checkpoint store ( hdfs/S3/File)  will loose state if it is unavailable at resume. This becomes all the more painful with your confirming that  "failed checkpoints killing the job"  b'coz essentially it mean that if remote store in unavailable  during checkpoint than you have lost state ( till of course you have a retry of none or an unbounded retry delay, a delay that you hope the store revives in ) .. Remember  the first retry failure  will cause new state according the code as written iff the remote store is down. We would rather have a configurable property that establishes  our desire to abort something like a "abort_retry_on_chkretrevalfailure"

In our case it is very important that we do not undercount a window, one reason we use flink and it's awesome failure guarantees, as various alarms sound ( we do anomaly detection on the time series ).

Please create a jira ticket for us to follow or we could do it.

PS Not aborting on checkpointing, till a configurable limit is very important too.

On Fri, Oct 6, 2017 at 2:36 AM, Aljoscha Krettek <[hidden email]> wrote:

Hi Vishal,

I think you're right! And thanks for looking into this so deeply. 

With your last mail your basically saying, that the checkpoint could not be restored because your HDFS was temporarily down. If Flink had not deleted that checkpoint it might have been possible to restore it at a later point, right?

Regarding failed checkpoints killing the job: yes, this is currently the expected behaviour but there are plans to change this.

Best,

Aljoscha

On 5. Oct 2017, at 17:40, Vishal Santoshi <[hidden email]> wrote:

I think this is the offending piece. There is a catch all Exception, which IMHO should understand a recoverable exception from an unrecoverable on. 

try {
	completedCheckpoint = retrieveCompletedCheckpoint(checkpointStateHandle);
	if (completedCheckpoint != null) {
	completedCheckpoints.add(completedCheckpoint);
	}
	} catch (Exception e) {
	LOG.warn("Could not retrieve checkpoint. Removing it from the completed " +
	"checkpoint store.", e);
	// remove the checkpoint with broken state handle
	removeBrokenStateHandle(checkpointStateHandle.f1, checkpointStateHandle.f0);
	}

On Thu, Oct 5, 2017 at 10:57 AM, Vishal Santoshi <[hidden email]> wrote:

So this is the issue and tell us that it is wrong. ZK had some state ( backed by hdfs ) that referred to a checkpoint ( the same exact last successful checkpoint that was successful before NN screwed us ). When the JM tried to recreate the state and b'coz NN was down failed to retrieve the CHK handle from hdfs and conveniently ( and I think very wrongly ) removed the CHK from being considered and cleaned the pointer ( though failed as was NN was down and is obvious from the dangling file in recovery ) . The metadata itself was on hdfs and failure in retrieving should have been a stop all, not going to trying doing magic exception rather than starting from a blank state.

org.apache.flink.util.FlinkException: Could not retrieve checkpoint 44286 from state handle under /0000000000000044286. This indicates that the retrieved state handle is broken. Try cleaning the state handle store.

On Thu, Oct 5, 2017 at 10:13 AM, Vishal Santoshi <[hidden email]> wrote:

Also note that  the zookeeper recovery did  ( sadly on the same hdfs cluster ) also showed the same behavior. It had the pointers to the chk point  ( I  think that is what it does, keeps metadata of where the checkpoint etc  ) .  It too decided to keep the recovery file from the failed state.

-rw-r--r--   3 root hadoop       7041 2017-10-04 13:55 /flink-recovery/prod/completedCheckpoint6c9096bb9ed4

-rw-r--r--   3 root hadoop       7044 2017-10-05 10:07 /flink-recovery/prod/completedCheckpoint7c5a19300092

This is getting a little interesting. What say you :)

On Thu, Oct 5, 2017 at 9:26 AM, Vishal Santoshi <[hidden email]> wrote:

Another thing I noted was this thing

drwxr-xr-x   - root hadoop          0 2017-10-04 13:54 /flink-checkpoints/prod/c4af8dfa864e2f9a51764de9f0725b39/chk-44286

drwxr-xr-x   - root hadoop          0 2017-10-05 09:15 /flink-checkpoints/prod/c4af8dfa864e2f9a51764de9f0725b39/chk-45428

Generally what Flink does IMHO is that it replaces the chk point directory with a new one. I see it happening now. Every minute it replaces the old directory.  In this job's case however, it did not delete the 2017-10-04 13:54  and hence the chk-44286 directory.  This was the last chk-44286 (  I think  )  successfully created before NN had issues but as is usual did not delete this  chk-44286. It looks as if it started with a blank slate ???????? Does this strike a chord ?????

On Thu, Oct 5, 2017 at 8:56 AM, Vishal Santoshi <[hidden email]> wrote:

Hello Fabian, 

                      First of all congratulations on this fabulous framework. I have worked with GDF and though GDF has some natural pluses Flink's state management is far more advanced. With kafka as a source it negates issues GDF has ( GDF integration with pub/sub is organic and that is to be expected but non FIFO pub/sub is an issue with windows on event time etc )

                   Coming back to this issue. We have that same kafka topic feeding a streaming druid datasource and we do not see any issue there, so so data loss on the source, kafka is not applicable. I am totally certain that the "retention" time was not an issue. It is 4 days of retention and we fixed this issue within 30 minutes. We could replay kafka with a new consumer group.id and that worked fine. 

Note these properties and see if they strike a chord.

* The setCommitOffsetsOnCheckpoints(boolean) for kafka consumers is the default true. I bring this up to see whether flink will in any circumstance drive consumption on the kafka perceived offset rather than the one in the checkpoint.

* The state.backend.fs.memory-threshold: 0 has not been set.  The state is big enough though therefore IMHO no way the state is stored along with the meta data in JM ( or ZK ? ) . The reason I bring this up is to make sure when you say that the size has to be less than 1024bytes , you are talking about cumulative state of the pipeine.

* We have a good sense of SP ( save point )  and CP ( checkpoint ) and certainly understand that they actually are not dissimilar. However in this case there were multiple attempts to restart the pipe before it finally succeeded. 

* Other hdfs related poperties.

 

 state.backend.fs.checkpointdir: hdfs:///flink-checkpoints/<%= flink_hdfs_root %>

 state.savepoints.dir: hdfs:///flink-savepoints/<%= flink_hdfs_root %>

 recovery.zookeeper.storageDir: hdfs:///flink-recovery/<%= flink_hdfs_root %>

Do these make sense ? Is there anything else I should look at.  Please also note that it is the second time this has happened. The first time I was vacationing and was not privy to the state of the flink pipeline, but the net effect were similar. The counts for the first window after an internal restart dropped. 

Thank you for you patience and regards,

Vishal

On Thu, Oct 5, 2017 at 5:01 AM, Fabian Hueske <[hidden email]> wrote:

Hi Vishal,

window operators are always stateful because the operator needs to remember previously received events (WindowFunction) or intermediate results (ReduceFunction).

Given the program you described, a checkpoint should include the Kafka consumer offset and the state of the window operator. If the program eventually successfully (i.e., without an error) recovered from the last checkpoint, all its state should have been restored. Since the last checkpoint was before HDFS went into safe mode, the program would have been reset to that point. If the Kafka retention time is less than the time it took to fix HDFS you would have lost data because it would have been removed from Kafka. If that's not the case, we need to investigate this further because a checkpoint recovery must not result in state loss.

Restoring from a savepoint is not so much different from automatic checkpoint recovery. Given that you have a completed savepoint, you can restart the job from that point. The main difference is that checkpoints are only used for internal recovery and usually discarded once the job is terminated while savepoints are retained.

Regarding your question if a failed checkpoint should cause the job to fail and recover I'm not sure what the current status is.

Stefan (in CC) should know what happens if a checkpoint fails.

Best, Fabian

2017-10-05 2:20 GMT+02:00 Vishal Santoshi <[hidden email]>:

To add to it, my pipeline is a simple 

keyBy(0)
        .timeWindow(Time.of(window_size, TimeUnit.MINUTES))
        .allowedLateness(Time.of(late_by, TimeUnit.SECONDS))
        .reduce(new ReduceFunction(), new WindowFunction())

On Wed, Oct 4, 2017 at 8:19 PM, Vishal Santoshi <[hidden email]> wrote:

Hello folks,

As far as I know checkpoint failure should be ignored and retried with potentially larger state. I had this situation

* hdfs went into a safe mode b'coz of Name Node issues
* exception was thrown

    org.apache.hadoop.ipc.RemoteException(org.apache.hadoop.ipc.StandbyException): Operation category WRITE is not supported in state standby. Visit https://s.apache.org/sbnn-error
    ..................

    at org.apache.flink.runtime.fs.hdfs.HadoopFileSystem.mkdirs(HadoopFileSystem.java:453)
        at org.apache.flink.core.fs.SafetyNetWrapperFileSystem.mkdirs(SafetyNetWrapperFileSystem.java:111)
        at org.apache.flink.runtime.state.filesystem.FsCheckpointStreamFactory.createBasePath(FsCheckpointStreamFactory.java:132)

* The pipeline came back after a few restarts and checkpoint failures, after the hdfs issues were resolved.

I would not have worried about the restart, but it was evident that I lost my operator state. Either it was my kafka consumer that kept on advancing it's offset between a start and the next checkpoint failure ( a minute's worth ) or the the operator that had partial aggregates was lost. I have a 15 minute window of counts on a keyed operator

I am using ROCKS DB and of course have checkpointing turned on.

The questions thus are

* Should a pipeline be restarted if checkpoint fails ?
* Why on restart did the operator state did not recreate ?
* Is the nature of the exception thrown have to do with any of this b'coz suspend and resume from a save point work as expected ?
* And though I am pretty sure, are operators like the Window operator stateful by drfault and thus if I have timeWindow(Time.of(window_size, TimeUnit.MINUTES)).reduce(new ReduceFunction(), new WindowFunction()), the state is managed by flink ?

Thanks.