[
https://issues.apache.org/jira/browse/IMPALA-9842?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17138108#comment-17138108
]
ASF subversion and git services commented on IMPALA-9842:
---------------------------------------------------------
Commit aa6d7887eec1efd33c73f77e5346a499569e5b6b in impala's branch
refs/heads/master from Joe McDonnell
[ https://gitbox.apache.org/repos/asf?p=impala.git;h=aa6d788 ]
IMPALA-9842: Fix hang when cancelling BufferedPlanRootSink
In BufferedPlanRootSink::FlushFinal(), if Cancel() runs
before FlushFinal() waits on the consumer_eos_ condition variable,
the thread in FlushFinal() will wait forever. This is because it is
not checking for cancellation or synchronizing with the Cancel()
thread.
Specifically:
Thread A: Calls BufferedPlanRootSink::Cancel(), signalling any
thread currently waiting on the consumer_eos_ condition variable.
Thread B: Enters FlushFinal(). Never tests RuntimeState::is_cancelled()
and calls Wait() on the consumer_eos_ condition variable. This waits
forever.
This changes BufferedPlanRootSink::Cancel() to get the lock_ before
signalling the consumer_eos_ condition variable. It also changes
FlushFinal() to call Wait() in a loop. It breaks out of the loop if
it is cancelled or the batch_queue_ is empty. There are two cases:
1. FlushFinal() gets the lock_ first and only releases it when waiting
on the consumer_eos_ condition variable. It will get signalled by
Cancel().
2. Cancel() gets the lock_ first and FlushFinal() will not wait,
because is_cancelled() is true.
Testing:
- Run core tests
Change-Id: Id6f3fbc05420ca95313fa79ea106547feb92b16b
Reviewed-on: http://gerrit.cloudera.org:8080/16088
Reviewed-by: Tim Armstrong <[email protected]>
Tested-by: Impala Public Jenkins <[email protected]>
> TestValidateMetrics.test_metrics_are_zero fails with num-fragments-in-flight
> not reaching zero
> ----------------------------------------------------------------------------------------------
>
> Key: IMPALA-9842
> URL: https://issues.apache.org/jira/browse/IMPALA-9842
> Project: IMPALA
> Issue Type: Bug
> Components: Backend
> Affects Versions: Impala 4.0
> Reporter: Joe McDonnell
> Assignee: Joe McDonnell
> Priority: Blocker
> Labels: broken-build, flaky
>
> A couple test runs have failed with TestValidateMetrics.test_metrics_are_zero
> failure:
> {noformat}
> AssertionError: Metric value impala-server.num-fragments-in-flight did not
> reach value 0 in 60s Dumping impalad debug pages: ...{noformat}
> The memz shows:
> {noformat}
> Process: Limit=12.00 GB Total=2.93 GB Peak=138.77 GB
> JVM: max heap size: Total=1.78 GB
> JVM: non-heap committed: Total=211.75 MB
> Buffer Pool: Free Buffers: Total=1.90 MB
> Buffer Pool: Clean Pages: Total=0
> Buffer Pool: Unused Reservation: Total=-2.00 MB
> Control Service Queue: Limit=50.00 MB Total=0 Peak=16.47 MB
> Data Stream Service Queue: Limit=614.40 MB Total=0 Peak=636.91 MB
> Data Stream Manager Early RPCs: Total=0 Peak=684.53 KB
> TCMalloc Overhead: Total=347.15 MB
> Spill-to-disk temporary compression buffers: Limit=512.00 MB Total=0
> Peak=110.43 MB
> RequestPool=root.default: Total=4.02 MB Peak=136.23 GB
> Query(074a86ef88e16768:1a7bf9f400000000): Limit=300.00 MB
> Reservation=4.00 MB Reserva$
> Fragment 074a86ef88e16768:1a7bf9f400000000: Reservation=4.00 MB
> OtherMemory=16.00 K$
> HDFS_SCAN_NODE (id=0): Reservation=0 OtherMemory=0 Total=0
> Peak=830.00 KB
> PLAN_ROOT_SINK: Reservation=4.00 MB ReservationLimit=100.00 MB
> OtherMemory=8.00 K$
> RequestPool=fe-eval-exprs: Total=0 Peak=256.00 MB
> RequestPool=root.no-limits: Total=0 Peak=188.88 MB
> Untracked Memory: Total=616.14 MB
> {noformat}
> Query 074a86ef88e16768:1a7bf9f4 is:
> {noformat}
> I0608 19:09:08.410117 25628 Frontend.java:1498]
> 074a86ef88e16768:1a7bf9f400000000] Analyzing query: select * from
> functional.alltypes limit 10 db: default{noformat}
> The interesting query options are debug_action and fetch_rows_timeout_ms.
> These narrow it down to hs2/test_fetch_timeout.py
> ([https://github.com/apache/impala/blob/master/tests/hs2/test_fetch_timeout.py#L207-L215]):
> {noformat}
> 11: debug_action (string) = "CRS_BEFORE_COORD_STARTS:SLEEP@5000",
> ...
> 94: fetch_rows_timeout_ms (i64) = 0,{noformat}
> It is executing only at the coordinator:
> {noformat}
> I0608 19:09:08.430388 25637 scheduler.cc:549]
> 074a86ef88e16768:1a7bf9f400000000] Exec at coord is true
> I0608 19:09:08.430755 25637 admission-controller.cc:1296]
> 074a86ef88e16768:1a7bf9f400000000] Trying to admit
> id=074a86ef88e16768:1a7bf9f400000000 in pool_name=root.default
> executor_group_name=default per_host_mem_estimate=20.00 MB
> dedicated_coord_mem_estimate=120.00 MB max_requests=-1 (configured
> statically) max_queued=200 (configured statically) max_mem=29.30 GB
> (configured statically)
> I0608 19:09:08.430793 25637 admission-controller.cc:1308]
> 074a86ef88e16768:1a7bf9f400000000] Stats: agg_num_running=10,
> agg_num_queued=0, agg_mem_reserved=3.79 GB,
> local_host(local_mem_admitted=4.67 GB, num_admitted_running=10, num_queued=0,
> backend_mem_reserved=2.24 GB)
> I0608 19:09:08.430811 25637 admission-controller.cc:894]
> 074a86ef88e16768:1a7bf9f400000000] Admitting query
> id=074a86ef88e16768:1a7bf9f400000000
> I0608 19:09:08.430881 25637 debug-util.cc:427]
> 074a86ef88e16768:1a7bf9f400000000] Debug Action:
> CRS_BEFORE_COORD_STARTS:SLEEP@5000 sleeping for 5000 ms{noformat}
> The query starts execution:
> {noformat}
> I0608 19:09:13.431097 25637 coordinator.cc:143]
> 074a86ef88e16768:1a7bf9f400000000] Exec()
> query_id=074a86ef88e16768:1a7bf9f400000000 stmt=select * from
> functional.alltypes limit 10
> I0608 19:09:13.431520 25637 coordinator.cc:463]
> 074a86ef88e16768:1a7bf9f400000000] starting execution on 1 backends for
> query_id=074a86ef88e16768:1a7bf9f400000000
> I0608 19:09:13.432674 114 control-service.cc:153]
> 074a86ef88e16768:1a7bf9f400000000] ExecQueryFInstances():
> query_id=074a86ef88e16768:1a7bf9f400000000 coord=f0873f7637c8:22000
> #instances=1
> ...
> I0608 19:09:13.433480 25637 coordinator.cc:520]
> 074a86ef88e16768:1a7bf9f400000000] started execution on 1 backends for
> query_id=074a86ef88e16768:1a7bf9f400000000
> I0608 19:09:13.434790 26254 query-state.cc:756]
> 074a86ef88e16768:1a7bf9f400000000] Executing instance.
> instance_id=074a86ef88e16768:1a7bf9f400000000 fragment_idx=0
> per_fragment_instance_idx=0 coord_state_idx=0 #in-flight=11{noformat}
> The query is unregistered:
> {noformat}
> I0608 19:09:13.448494 25628 impala-server.cc:1173] UnregisterQuery():
> query_id=074a86ef88e16768:1a7bf9f400000000
> I0608 19:09:13.448557 25628 coordinator.cc:644] ExecState: query
> id=074a86ef88e16768:1a7bf9f400000000 execution cancelled
> I0608 19:09:13.448580 25628 coordinator-backend-state.cc:943]
> query_id=074a86ef88e16768:1a7bf9f400000000 target backend=172.18.0.4:27000:
> Sending CancelQueryFInstances rpc
> I0608 19:09:13.448971 114 control-service.cc:223] CancelQueryFInstances():
> query_id=074a86ef88e16768:1a7bf9f400000000
> I0608 19:09:13.448995 114 query-exec-mgr.cc:105] QueryState:
> query_id=074a86ef88e16768:1a7bf9f400000000 refcnt=4
> I0608 19:09:13.449007 114 query-state.cc:782] Cancel:
> query_id=074a86ef88e16768:1a7bf9f400000000
> I0608 19:09:13.449023 114 krpc-data-stream-mgr.cc:337] cancelling active
> streams for fragment_instance_id=074a86ef88e16768:1a7bf9f400000000
> I0608 19:09:13.449226 25628 coordinator.cc:889] CancelBackends()
> query_id=074a86ef88e16768:1a7bf9f400000000, tried to cancel 1 backends
> I0608 19:09:13.449291 25628 coordinator.cc:1225] Release admission control
> resources for query_id=074a86ef88e16768:1a7bf9f400000000{noformat}
> After that, there are periodic reports coming in for
> 074a86ef88e16768:1a7bf9f400000000 which all return that it is an invalid
> query.
> The original query is starts at 19:09:08, while the
> TestValidateMetrics.test_metrics_are_zero check for remaining fragments is
> happening over an hour later at 20:14:08.**
> Interestingly, this has happened on different runs *with the exact same
> query*.
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
