You are viewing a plain text version of this content. The canonical link for it is here.
Posted to issues@ignite.apache.org by "Vladislav Pyatkov (JIRA)" <ji...@apache.org> on 2017/10/31 09:42:00 UTC
[jira] [Created] (IGNITE-6799) Check of starvation in striped
thread pool
Vladislav Pyatkov created IGNITE-6799:
-----------------------------------------
Summary: Check of starvation in striped thread pool
Key: IGNITE-6799
URL: https://issues.apache.org/jira/browse/IGNITE-6799
Project: Ignite
Issue Type: Improvement
Security Level: Public (Viewable by anyone)
Reporter: Vladislav Pyatkov
We have got false alarm like:
{noformat}
2017-10-30 14:01:40.308[WARN ][grid-timeout-worker-#63%DPL_GRID%DplGridNodeName%][o.a.ignite.internal.util.typedef.G] >>> Possible starvation in striped pool.
2017-10-30 13:56:41.538[WARN ][grid-timeout-worker-#63%DPL_GRID%DplGridNodeName%][o.a.ignite.internal.util.typedef.G] >>> Possible starvation in striped pool.
2017-10-30 13:46:40.488[WARN ][grid-timeout-worker-#63%DPL_GRID%DplGridNodeName%][o.a.ignite.internal.util.typedef.G] >>> Possible starvation in striped pool.
2017-10-30 13:37:45.481[WARN ][grid-timeout-worker-#63%DPL_GRID%DplGridNodeName%][o.a.ignite.internal.util.typedef.G] >>> Possible starvation in striped pool.
{noformat}
It will be on checkpoint usually, but that is false triggering. Because thread have not been active long time, but got active recently.
We should save last active state on stripe like it done with completedCntrs and rewrite condition:
{code}
completedCntrs[i] != -1 &&
completedCntrs[i] == completedCnt &&
actives[i] == active &&
active
{code}
--
This message was sent by Atlassian JIRA
(v6.4.14#64029)