[
https://issues.apache.org/jira/browse/SOLR-12692?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16592720#comment-16592720
]
Jan Høydahl commented on SOLR-12692:
------------------------------------
The patch looks good. A bit hard to test the various error conditions if we
don't mock the response data. Few comments:
* If you get into zk_max_latency issues, the same error may be added once for
each ZK if all are busy. Perhaps it helps to include the host name in the
message to distinguish?
* The "ok" key for each zkhost is "true" if RUOK returns IMOK. Should we flip
that to false if we detect issues with that host?
* Not high prio, but code-wise it would perhaps be cleaner to separate
information fetch phase ({{monitorZookeeper}}) from the inspection and
detection of errors. I.e. keep {{monitorZookeeper}} as-is and add a new method
{{detectIssues(zkStatus, errors)}} where all analysis, both existing and the
new per-host analysis is done? This is less important though.
> Add hints/warnings for the ZK Status Admin UI
> ---------------------------------------------
>
> Key: SOLR-12692
> URL: https://issues.apache.org/jira/browse/SOLR-12692
> Project: Solr
> Issue Type: Improvement
> Security Level: Public(Default Security Level. Issues are Public)
> Components: Admin UI
> Reporter: Varun Thacker
> Priority: Minor
> Attachments: SOLR-12692.patch, wrong_zk_warning.png, zk_ensemble.png
>
>
> Firstly I love the new UI pages ( ZK Status and Nodes ) . Thanks [~janhoy]
> for all the great work!
> I setup a 3 node ZK ensemble to play around with the UI and attaching the
> screenshot for reference.
>
> Here are a few suggestions I had
> # Let’s show Approximate Size in human readable form. We can use
> RamUsageEstimator#humanReadableUnits to calculate it
> # Show warning symbol when Ensemble is standalone
> # If maxSessionTimeout < Solr's ZK_CLIENT_TIMEOUT then ZK will only honor
> up-to the maxSessionTimeout value for the Solr->ZK connection. We could mark
> that as a warning.
> # If maxClientCnxns < live_nodes show this as a red? Each solr node connects
> to all zk nodes so if the number of nodes in the cluster is high one should
> also be increasing maxClientCnxns
>
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscr...@lucene.apache.org
For additional commands, e-mail: dev-h...@lucene.apache.org
