[
https://issues.apache.org/jira/browse/HAWQ-201?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15037041#comment-15037041
]
ASF GitHub Bot commented on HAWQ-201:
-------------------------------------
Github user huor commented on a diff in the pull request:
https://github.com/apache/incubator-hawq/pull/144#discussion_r46503312
--- Diff: src/backend/resourcemanager/resourcemanager.c ---
@@ -2168,6 +2169,37 @@ int generateAllocRequestToBroker(void)
elog(RMLOG, "Resource manager now needs %d GRM containers.", reqcore);
+ /*
+ * Check if should raise water level to deal with resource fragment or
+ * resource uneven problems. We trigger this logic only when no resource
+ * request caused by lack of resource, and no pending resource are
waited
+ * for.
+ */
+ if ( reqcore <= 0 &&
+ mctrack->TotalPending.Core <= 0 &&
+ (PQUEMGR->hasResourceProblem[RESPROBLEM_FRAGMENT] ||
+ PQUEMGR->hasResourceProblem[RESPROBLEM_UNEVEN] ||
+ PQUEMGR->hasResourceProblem[RESPROBLEM_TOOFEWSEG]) )
+ {
+ /* Check if it is possible to raise water level. */
+ if ( mctrack->TotalAllocated.Core + 1 <=
+ PRESPOOL->GRMTotal.Core * PQUEMGR->GRMQueueMaxCapacity
)
+ {
+ /*
+ * We only add one more GRM container to acquire, this
will trigger
+ * the following logic to raise the water level.
+ */
+ reqcore = 1;
+ reqmem = reqcore * mctrack->MemCoreRatio;
+
+ PQUEMGR->hasResourceProblem[RESPROBLEM_FRAGMENT] =
false;
--- End diff --
Here we are marking fragment, or uneven, or too few segments problem false,
is this means that adding one more container would resolve all these three kind
of issue?
> Resource uneven or fragment problem may cause resource queue goes into long
> term pause
> --------------------------------------------------------------------------------------
>
> Key: HAWQ-201
> URL: https://issues.apache.org/jira/browse/HAWQ-201
> Project: Apache HAWQ
> Issue Type: Bug
> Components: Resource Manager
> Reporter: Yi Jin
> Assignee: Yi Jin
>
> This fix is to let resource manager actively raise water level to try to deal
> with possible resource uneven, resource fragment problem.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
