[
https://issues.apache.org/jira/browse/OAK-11246?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Rishabh Daim resolved OAK-11246.
--------------------------------
Fix Version/s: 1.74.0
Resolution: Fixed
I have merged the PR into the trunk. Thanks [~ionutpirlogea] for your
contribution.
> Introduce config for perflogger to log very slow operations at info
> -------------------------------------------------------------------
>
> Key: OAK-11246
> URL: https://issues.apache.org/jira/browse/OAK-11246
> Project: Jackrabbit Oak
> Issue Type: Task
> Components: mongomk
> Reporter: Ionut Pirlogea
> Assignee: Rishabh Daim
> Priority: Minor
> Fix For: 1.74.0
>
>
> In oak we have a
> [PerfLogger|https://github.com/apache/jackrabbit-oak/blob/trunk/oak-commons/src/main/java/org/apache/jackrabbit/oak/commons/PerfLogger.java#L48]
> which can be used to log slow operations. It is used in a variety of places,
> most prominent:
> * DocumentNodeStore
> * MongoDocumentStore
> * EventGenerator
> * DocumentNodeStoreBranch
> * AbstractDocumentNodeState
> * NodeObserver
> * and in several places in indexing code
> * as well as in RDB case
> Currently we use only the DEBUG thresholds, we're not using the INFO
> thresholds. That means, we only would see logs when the level is at DEBUG, we
> see nothing by default at INFO.
> We could however consider logging very slow operations at INFO by default.
> Much like mongo logs requests slower than 100ms always, by default.
> In AEM's case we might have many slow operations though, so choosing a low
> threshold would potentially flood the log and cause troubles that way.
> Additionally, there isn't a (global) config that can be tweaked to set this
> threshold. So far it is all hard-coded - and the threshold is actually
> Long.MAX_VALUE (so essentially never).
> Suggesting hereby to introduce such a config - perhaps it doens't have to be
> global, it could be for example covering a few classes. But having such a
> config, then having that tweakable via skyline-ops, could be a useful
> feature. We could then set for example a threshold of 60 seconds and monitor
> the amount of logs generated by that throughout Skyline. If the number is low
> enough, we can go lower to eg 30 sec or ideally eventually 10 sec. Always
> given we are not flooding splunk that way. But the prerequisite for something
> like that is. a skyline-ops configurable perflogger threshold (and wire that
> to the corresponding {{Perflogger.end()}} methods at places we want it to be.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
