You are viewing a plain text version of this content. The canonical link for it is here.
Posted to issues@geode.apache.org by "ASF subversion and git services (JIRA)" <ji...@apache.org> on 2019/02/02 01:16:01 UTC

[jira] [Commented] (GEODE-6329) gfsh show missing-disk-stores shows disk stores that aren't missing

    [ https://issues.apache.org/jira/browse/GEODE-6329?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16758802#comment-16758802 ] 

ASF subversion and git services commented on GEODE-6329:
--------------------------------------------------------

Commit 8e32605e8ca026eb28f4c6095dfbcaca5cbd09a1 in geode's branch refs/heads/feature/GEODE-6291 from Dan Smith
[ https://gitbox.apache.org/repos/asf?p=geode.git;h=8e32605 ]

GEODE-6329, GEODE-6328: Fix issues with show missing-disk-stores (#3136)

* Using consistent logic for finding missing disk stores

The gfsh command for finding missing disk stores was not ignoring
present disk stores. However, the JMX command was. The gfsh command was also not deduplicating the missing disk stores. Changing the gfsh command to use the JMX logic.

* Making sure member is removed from persistent view

When we start up a member that has persistent data that it should delete
(due to already established redundancy), the member was not correctly
removing members from it's persistent view in
PersistenceAdvisorImpl.updateMembershipView. This was due to the fact
that the destroyOfflineData method was clearing the recoveredMembers
set, but not the underlying persistentMemberView.

The persistentMemberView probably should be preserved, in case the
member is killed and restarted before it gets a new view. So we should
reset the recoveredMembers set to the correct value.

* Simplifying the logic for reporting present disk stores

The logic for reporting what disk stores were actually running on a
node, for the purposes of not reporting those disk stores as missing,
was overly complicated. Due to that, it only included disk stores that
actually had any persistent regions/buckets on that node, rather than
all disk stores.

Refactoring the code to just list all disk stores.

> gfsh show missing-disk-stores shows disk stores that aren't missing
> -------------------------------------------------------------------
>
>                 Key: GEODE-6329
>                 URL: https://issues.apache.org/jira/browse/GEODE-6329
>             Project: Geode
>          Issue Type: Bug
>          Components: management
>            Reporter: Dan Smith
>            Assignee: Dan Smith
>            Priority: Major
>              Labels: pull-request-available
>             Fix For: 1.9.0
>
>          Time Spent: 40m
>  Remaining Estimate: 0h
>
> The gfsh show missing-disk-stores command is showing disk stores that aren't actually missing.
> To reproduce 
>  - Start 4 servers with a persistent PR with redundancy 1
>  - Stop 2 servers, rebalancing in between
>  - Restart the 2 servers and wait for recovery to complete
> Gfsh reports one of the disk stores of the first stopped server as missing.
> There appear to be few problems here. The JMX listMissingDiskStores removes disk stores that are reported as present by other members, however  the gfsh command is not using the same logic, so it's reporting the same disk stores as missing even if they are reported as present.
> The JMX command also can fail here, however, because if a member has a disk store but 0 buckets, the member is not reporting the disk store as present.
> Finally, the missing disk stores reported by individual nodes come from the persistent view of each bucket. Geode should be updating the persistent view when each member restarts. Therefore the "missing" disk store should have been removed from the persistent view.



--
This message was sent by Atlassian JIRA
(v7.6.3#76005)