[
https://issues.apache.org/jira/browse/RATIS-533?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16824635#comment-16824635
]
Josh Elser commented on RATIS-533:
----------------------------------
RaftLogCache will hold onto up to {{MaxCachedSegmentNum}} segments in memory
(by default, that'd be 2 * 8MB). The number {{2}} here is what Ozone overrides
to be {{6}}.
{{StateMachineData.setCachingEnabled}} seems to remove {{StateMachineData}}
from the LogEntryProto that is heading to the (Segmented)RaftLog. It seems to
read as though queries would go directly to the StateMachine to answer
questions instead of hitting the cached RaftLog.
I think for our case, we would want the RaftLogCache to hold the Segments in
memory and have the data we're writing since we are not holding on to any state
in the statemachine (to avoid the RaftLog read). In other words, I think
cachingEnabled is an optimization which we can't benefit from, but increasing
the the number of cached segments would help optimizing reads.
> Set important server config options for logservice
> --------------------------------------------------
>
> Key: RATIS-533
> URL: https://issues.apache.org/jira/browse/RATIS-533
> Project: Ratis
> Issue Type: Improvement
> Components: LogService
> Reporter: Josh Elser
> Assignee: Josh Elser
> Priority: Major
>
> Mukul was kind enough to link to
> [https://github.com/apache/hadoop/blob/d31c86892e0ceec5d642f76fc9123fac4fd80db8/hadoop-hdds/container-service/src/main/java/org/apache/hadoop/ozone/container/common/transport/server/ratis/XceiverServerRatis.java]
> in Ozone which lists a bunch of RatisServer configuration keys that Ozone
> sets.
> We should go through this list and set the ones that make sense for the
> LogService too. For example, initial testing with a longer leader election
> timeout helps significantly.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
