[
https://issues.apache.org/jira/browse/PHOENIX-2692?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Gabriel Reid updated PHOENIX-2692:
----------------------------------
Attachment: PHOENIX-2692.patch
Patch to add a {{phoenix.stats.enabled}} config setting, which is true by
default, so that statistics can be disabled.
The configuration value is read via the configuration in the
RegionCoprocessorEnvironment in UngroupedAggregateRegionObserver, which makes
it technically possible to disable stats collection for a single table by
setting this setting as a coprocessor parameter.
This is a patch against the HBase 1.0 main line, and will need a little bit of
work to rebase it on master, but the full integration test suite runs correctly
with it applied.
> Provide flag for disabling statistics collection
> ------------------------------------------------
>
> Key: PHOENIX-2692
> URL: https://issues.apache.org/jira/browse/PHOENIX-2692
> Project: Phoenix
> Issue Type: New Feature
> Reporter: Gabriel Reid
> Attachments: PHOENIX-2692.patch
>
>
> It would be useful to be able to completely disable statistics collection in
> some situations. This is typically done for debugging performance issues that
> have come up as part of statistics collection.
> Disabling statistics collection is typically done by setting the
> {{phoenix.stats.guidepost.width}} config setting to a very large number, but
> this has the problem that it is not 100% reliable (some guidepost data is
> still written to the SYSTEM.STATS table if this is done), and it's just kind
> of a work-around that currently (mostly) works, but this method of disabling
> stats collection isn't guaranteed to work in the future (for example, if new
> types of statistics are also collected, or if they're based on something
> other than file size).
> Our specific use case for disabling statistics collection is based on the way
> that we're typically using Phoenix, which is more as a library which
> facilitates querying a combination of data that is managed by Phoenix along
> with data that is not managed by Phoenix. In this case, because the
> statistics collection only looks at the data managed by Phoenix, it doesn't
> function as needed (as well as having the additional issue of leading to
> problems like LP-2680).
> Ideally we'll move towards using statistics as we iron out these issues, but
> having the ability to disable statistics (at least for now) would also be
> very useful.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
