phoenix-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "James Taylor (JIRA)" <j...@apache.org>
Subject [jira] [Commented] (PHOENIX-4701) Write client-side metrics asynchronously to SYSTEM.LOG
Date Fri, 11 May 2018 01:30:00 GMT

    [ https://issues.apache.org/jira/browse/PHOENIX-4701?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16471363#comment-16471363
] 

James Taylor commented on PHOENIX-4701:
---------------------------------------

Nice, [~ankit@apache.org]. This is fantastic. Do the test in QueryLoggerIT pass? Would be
good to have more too, but I think this is a big step forward.

Would still be good to confirm the places we emit these metrics to see if we're missing anything.
Hopefully Thomas or Karan know already.

If you're ok with it, I think we should get this into 4.14. We can mark it as beta while
we get feedback. It can be disabled completely, correct?

> Write client-side metrics asynchronously to SYSTEM.LOG
> ------------------------------------------------------
>
>                 Key: PHOENIX-4701
>                 URL: https://issues.apache.org/jira/browse/PHOENIX-4701
>             Project: Phoenix
>          Issue Type: Bug
>            Reporter: James Taylor
>            Assignee: James Taylor
>            Priority: Major
>             Fix For: 4.15.0
>
>         Attachments: PHOENIX-4701_wip1.patch, PHOENIX-4701_wip2.patch, PHOENIX-4701_wip3.patch
>
>
> Rather than inventing a new, different set of client-side metrics to persist, we should
just persist our [client metrics|http://phoenix.apache.org/metrics.html] in the SYSTEM.LOG.
The metrics captures all the same information as your QueryLogInfo (and much more), rolls
all the information up to a single set of metrics for each Phoenix statement (aggregating/merging
parallel scans, etc), and can emits a single log line (which could be written in a single
upsert statement). At SFDC, we emit this information into a file system log in a layer above
(and use Splunk to produce nifty dashboard for monitoring), but this could easily be emitted
directly in Phoenix and go through your asynchronous write path (and then use Phoenix queries
to produce the same kind of dashboards). The only piece would be to add the concept of a log
level to each metric to enable statically controlling which metrics are output.
> With this approach, the SYSTEM.LOG table could be declared immutable and use our dense
storage format with a single byte for column encoding and get a 3-5x perf gain. This would
also open the door for users to potentially add secondary indexes on the table. See schema
identified in the wip2 patch.



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

Mime
View raw message