[
https://issues.apache.org/jira/browse/IMPALA-10987?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17434491#comment-17434491
]
Vihang Karajgaonkar commented on IMPALA-10987:
----------------------------------------------
Yes, unfortunately, currently we require a global invalidate to reset the
events processor if the events sync is reenabled on a table. My original
thinking behind this design decision was that
1) Events processor cannot just start processing events from that point onwards
because of the fact that it might have missed some create/drop events as well.
This is probably more relevant to database level flag than table level although
a table may also had add/drop partition events which are skipped during this
time window.
2) I did not anticipate re-enabling events sync on a table or database may not
be very common. This would likely be a one-time operation and hence I thought
it was okay to do a catalogd reset.
That said this was always on my to-do list to get rid of this requirement. I
will look into ways to avoid doing global invalidate when the events sync is
turned back on. I don't think this is a bug since it is documented behavior.
See https://impala.apache.org/docs/build/html/topics/impala_metadata.html
> Changing impala.disableHmsSync in Hive can break event processing
> -----------------------------------------------------------------
>
> Key: IMPALA-10987
> URL: https://issues.apache.org/jira/browse/IMPALA-10987
> Project: IMPALA
> Issue Type: Bug
> Components: Frontend
> Reporter: Csaba Ringhofer
> Priority: Major
>
> To reproduce, start Impala with event polling:
> {code}
> bin/start-impala-cluster.py --catalogd_args="--hms_event_polling_interval_s=2
> --catalog_topic_mode=minimal" --impalad_args="--use_local_catalog=1"
> {code}
> From Hive:
> {code}
> CREATE DATABASE temp;
> CREATE EXTERNAL TABLE temp.t (i int) PARTITIONED BY (p int)
> TBLPROPERTIES('impala.disableHmsSync'='true');
> ALTER TABLE temp.t SET TBLPROPERTIES ('impala.disableHmsSync'='false');
> {code}
> From this point event sync will be broken in Impala. It can be fixed only
> with global INVALIDATE METADATA (or restarting catalogd)
> catalogd log will include an exception like this:
> {code}
> E1026 10:30:16.151208 22514 MetastoreEventsProcessor.java:653] Event
> processing needs a invalidate command to resolve the state
> Java exception follows:
> org.apache.impala.catalog.events.MetastoreNotificationNeedsInvalidateException:
> EventId: 15956 EventType: ALTER_TABLE Detected that event sync was tur
> ned on for the table temp.t and the table does not exist. Event processing
> cannot be continued further. Issue a invalidate metadata command to reset
> the event processing state
> at
> org.apache.impala.catalog.events.MetastoreEvents$AlterTableEvent.process(MetastoreEvents.java:992)
> at
> org.apache.impala.catalog.events.MetastoreEvents$MetastoreEvent.processIfEnabled(MetastoreEvents.java:345)
> at
> org.apache.impala.catalog.events.MetastoreEventsProcessor.processEvents(MetastoreEventsProcessor.java:747)
> at
> org.apache.impala.catalog.events.MetastoreEventsProcessor.processEvents(MetastoreEventsProcessor.java:645)
> at
> java.util.concurrent.Executors$RunnableAdapter.call(Executors.java:511)
> at java.util.concurrent.FutureTask.runAndReset(FutureTask.java:308)
> at
> java.util.concurrent.ScheduledThreadPoolExecutor$ScheduledFutureTask.access$301(ScheduledThreadPoolExecutor.java:180)
> at
> java.util.concurrent.ScheduledThreadPoolExecutor$ScheduledFutureTask.run(ScheduledThreadPoolExecutor.java:294)
> at
> java.util.concurrent.ThreadPoolExecutor.runWorker(ThreadPoolExecutor.java:1149)
> at
> java.util.concurrent.ThreadPoolExecutor$Worker.run(ThreadPoolExecutor.java:624)
> {code}
> and future events will be lead to a log like this:
> {code}
> W1026 10:30:18.151962 22514 MetastoreEventsProcessor.java:638] Event
> processing is skipped since status is NEEDS_INVALIDATE. Last synced event id
> is 15955
> {code}
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
