Details
-
Bug
-
Status: Resolved
-
Critical
-
Resolution: Duplicate
-
5.14.5
-
None
-
None
-
Patch
Description
When ActiveMQ is configured with JDBC persistence storage (postgres) and there is many pending events (more than 5 000 000) consumers receive events in a bulk only every several seconds even if a queue contains a lot of events.
To get such state of AMQ I've used activemq-perf-maven-plugin with below settings:
mvn org.apache.activemq.tooling:activemq-perf-maven-plugin:consumer -Dconsumer.destName=queue://TEST.FOO -Dconsumer.recvDuration=300000 -Dconsumer.sessAckMode=tran sacted -DsysTest.reportType=verbose -DsysTest.numClients=3 -Dfactory.asyncSession=false -Dfactory.prefetchQueue=1 -Dconsumer.sessTransacted=true -Dfactory.prefetchPolicy.queuePrefetch=1
Output and result of execution attached.
On database side there are information that execution of query is slow:
2017-07-13 04:37:38 UTC;activemq;11184;5966f908.2bb0;7/2253014;0;1;LOG: duration: 7706.308 ms execute S_3: SELECT ID, MSG FROM ACTIVEMQ_MSGS WHERE CONTAINER=$1 AND ID < $2 AND ID > $3 AND XID IS NULL ORDER BY ID 2017-07-13 04:37:38 UTC;activemq;11184;5966f908.2bb0;7/2253014;0;2;DETAIL: parameters: $1 = 'queue://TEST.FOO', $2 = '11786094', $3 = '5583825' 2017-07-13 04:37:47 UTC;activemq;11184;5966f908.2bb0;7/2255501;0;3;LOG: duration: 7921.731 ms execute S_3: SELECT ID, MSG FROM ACTIVEMQ_MSGS WHERE CONTAINER=$1 AND ID < $2 AND ID > $3 AND XID IS NULL ORDER BY ID 2017-07-13 04:37:47 UTC;activemq;11184;5966f908.2bb0;7/2255501;0;4;DETAIL: parameters: $1 = 'queue://TEST.FOO', $2 = '11786094', $3 = '5586325' 2017-07-13 04:37:56 UTC;activemq;6556;5966f907.199c;6/1900865;0;1;LOG: duration: 8110.328 ms execute S_3: SELECT ID, MSG FROM ACTIVEMQ_MSGS WHERE CONTAINER=$1 AND ID < $2 AND ID > $3 AND XID IS NULL ORDER BY ID 2017-07-13 04:37:56 UTC;activemq;6556;5966f907.199c;6/1900865;0;2;DETAIL: parameters: $1 = 'queue://TEST.FOO', $2 = '11786094', $3 = '5588825' 2017-07-13 04:38:03 UTC;activemq;6556;5966f907.199c;6/1901366;0;3;LOG: duration: 7711.928 ms execute S_3: SELECT ID, MSG FROM ACTIVEMQ_MSGS WHERE CONTAINER=$1 AND ID < $2 AND ID > $3 AND XID IS NULL ORDER BY ID 2017-07-13 04:38:03 UTC;activemq;6556;5966f907.199c;6/1901366;0;4;DETAIL: parameters: $1 = 'queue://TEST.FOO', $2 = '11786094', $3 = '5589325' 2017-07-13 04:38:11 UTC;activemq;2148;5966f359.864;2/2670600;0;3;LOG: duration: 7737.924 ms execute S_5: SELECT ID, MSG FROM ACTIVEMQ_MSGS WHERE CONTAINER=$1 AND ID < $2 AND ID > $3 AND XID IS NULL ORDER BY ID 2017-07-13 04:38:11 UTC;activemq;2148;5966f359.864;2/2670600;0;4;DETAIL: parameters: $1 = 'queue://TEST.FOO', $2 = '11786094', $3 = '5589825' 2017-07-13 04:38:20 UTC;activemq;2148;5966f359.864;2/2671104;0;5;LOG: duration: 7750.932 ms execute S_5: SELECT ID, MSG FROM ACTIVEMQ_MSGS WHERE CONTAINER=$1 AND ID < $2 AND ID > $3 AND XID IS NULL ORDER BY ID 2017-07-13 04:38:20 UTC;activemq;2148;5966f359.864;2/2671104;0;6;DETAIL: parameters: $1 = 'queue://TEST.FOO', $2 = '11786094', $3 = '5590325' 2017-07-13 04:38:28 UTC;activemq;6556;5966f907.199c;6/1902887;0;5;LOG: duration: 7718.725 ms execute S_3: SELECT ID, MSG FROM ACTIVEMQ_MSGS WHERE CONTAINER=$1 AND ID < $2 AND ID > $3 AND XID IS NULL ORDER BY ID 2017-07-13 04:38:28 UTC;activemq;6556;5966f907.199c;6/1902887;0;6;DETAIL: parameters: $1 = 'queue://TEST.FOO', $2 = '11786094', $3 = '5590825' 2017-07-13 04:38:35 UTC;activemq;11184;5966f908.2bb0;7/2259004;0;5;LOG: duration: 7685.510 ms execute S_3: SELECT ID, MSG FROM ACTIVEMQ_MSGS WHERE CONTAINER=$1 AND ID < $2 AND ID > $3 AND XID IS NULL ORDER BY ID 2017-07-13 04:38:35 UTC;activemq;11184;5966f908.2bb0;7/2259004;0;6;DETAIL: parameters: $1 = 'queue://TEST.FOO', $2 = '11786094', $3 = '5591325' 2017-07-13 04:38:43 UTC;activemq;2148;5966f359.864;2/2672124;0;7;LOG: duration: 7717.112 ms execute S_5: SELECT ID, MSG FROM ACTIVEMQ_MSGS WHERE CONTAINER=$1 AND ID < $2 AND ID > $3 AND XID IS NULL ORDER BY ID 2017-07-13 04:38:43 UTC;activemq;2148;5966f359.864;2/2672124;0;8;DETAIL: parameters: $1 = 'queue://TEST.FOO', $2 = '11786094', $3 = '5591825' 2017-07-13 04:38:51 UTC;activemq;11184;5966f908.2bb0;7/2259996;0;7;LOG: duration: 7717.112 ms execute S_3: SELECT ID, MSG FROM ACTIVEMQ_MSGS WHERE CONTAINER=$1 AND ID < $2 AND ID > $3 AND XID IS NULL ORDER BY ID 2017-07-13 04:38:51 UTC;activemq;11184;5966f908.2bb0;7/2259996;0;8;DETAIL: parameters: $1 = 'queue://TEST.FOO', $2 = '11786094', $3 = '5592325' 2017-07-13 04:38:59 UTC;activemq;11184;5966f908.2bb0;7/2260488;0;9;LOG: duration: 7682.116 ms execute S_3: SELECT ID, MSG FROM ACTIVEMQ_MSGS WHERE CONTAINER=$1 AND ID < $2 AND ID > $3 AND XID IS NULL ORDER BY ID 2017-07-13 04:38:59 UTC;activemq;11184;5966f908.2bb0;7/2260488;0;10;DETAIL: parameters: $1 = 'queue://TEST.FOO', $2 = '11786094', $3 = '5592825' 2017-07-13 04:39:07 UTC;activemq;11184;5966f908.2bb0;7/2260989;0;11;LOG: duration: 7718.309 ms execute S_3: SELECT ID, MSG FROM ACTIVEMQ_MSGS WHERE CONTAINER=$1 AND ID < $2 AND ID > $3 AND XID IS NULL ORDER BY ID 2017-07-13 04:39:07 UTC;activemq;11184;5966f908.2bb0;7/2260989;0;12;DETAIL: parameters: $1 = 'queue://TEST.FOO', $2 = '11786094', $3 = '5593325' 2017-07-13 04:39:15 UTC;activemq;7472;5966e588.1d30;5/1987914;0;3;LOG: duration: 7779.711 ms execute S_5: SELECT ID, MSG FROM ACTIVEMQ_MSGS WHERE CONTAINER=$1 AND ID < $2 AND ID > $3 AND XID IS NULL ORDER BY ID 2017-07-13 04:39:15 UTC;activemq;7472;5966e588.1d30;5/1987914;0;4;DETAIL: parameters: $1 = 'queue://TEST.FOO', $2 = '11786094', $3 = '5594325' 2017-07-13 04:39:23 UTC;activemq;7472;5966e588.1d30;5/1988403;0;5;LOG: duration: 7789.711 ms execute S_5: SELECT ID, MSG FROM ACTIVEMQ_MSGS WHERE CONTAINER=$1 AND ID < $2 AND ID > $3 AND XID IS NULL ORDER BY ID 2017-07-13 04:39:23 UTC;activemq;7472;5966e588.1d30;5/1988403;0;6;DETAIL: parameters: $1 = 'queue://TEST.FOO', $2 = '11786094', $3 = '5594825'
I've tried to improve performance on database side by creating new indexes but it hasn't helped. The problem is in number of rows that the query returns (in may case more than 5 000 000).
After analyzing AMQ code a fix for it seams easy. Only first maxPageSize rows (in my case 500) are used so rest of rows can be skipped and even not processed by database.
I think that the findNextMessagesStatement should be changed to:
findNextMessagesStatement = "SELECT ID, MSG FROM " + getFullMessageTableName() + " WHERE CONTAINER=? AND ID < ? AND ID > ? AND XID IS NULL ORDER BY ID LIMIT ?";
where the LIMIT is set to maxPageSize.
What do you think about such change? Is it safe in context of data consistency and events processing in ActiveMQ? I know that LIMIT may not be a solution for everybody
as it isn't supported by all databases.
There is org.apache.activemq.store.jdbc.JDBCAdapter.limitQuery method that should modify query to limit amount of rows but it has valuable implementation only in OracleJDBCAdapter, for other databases the method doesn't modify a query so limit isn't added.
I also wonder why the range in query ID < $2 AND ID > $3 is so wide?
Attachments
Attachments
Issue Links
- is duplicated by
-
AMQ-7368 Performance issue on PostgreSQL when we have lot of pending messages
- Resolved