[jira] [Created] (OAK-2667) Add new tooling to troubleshot mongodb slow queries

["Thierry Ygé (JIRA)" <ji...@apache.org>]

Thierry Ygé created OAK-2667:
--------------------------------

             Summary: Add new tooling to troubleshot mongodb slow queries
                 Key: OAK-2667
                 URL: https://issues.apache.org/jira/browse/OAK-2667
             Project: Jackrabbit Oak
          Issue Type: New Feature
          Components: oak-core
    Affects Versions: 1.0.12
            Reporter: Thierry Ygé
            Priority: Minor


currently some customer reported Hugely slow query at MongoDB level.

    2015-03-16T20:01:24.870+0100 [conn12] query aem-author.nodes query: { $query: { id:
    Unknown macro: { $gt}

    , modified:
    Unknown macro: { $gte}

    }, $orderby:
    Unknown macro: { id}

    , $hint:
    Unknown macro: { id}

    } planSummary: IXSCAN
    Unknown macro: { _id}

    ntoreturn:0 ntoskip:0 nscanned:9021956 nscannedObjects:9021956 keyUpdates:0 numYields:251409 locks(micros) r:553377233 nreturned:13 reslen:5737 1513722ms
    2015-03-18T14:14:15.903+0100 [conn52] query aem-author.nodes query: { $query: { id:
    Unknown macro: { $gt}

    , modified:
    Unknown macro: { $gte}

    }, $orderby:
    Unknown macro: { id}

    , $hint:
    Unknown macro: { id}

    } planSummary: IXSCAN
    Unknown macro: { _id}

    ntoreturn:0 ntoskip:0 nscanned:9047318 nscannedObjects:9047318 keyUpdates:0 numYields:223663 locks(micros) r:390010493 nreturned:73 reslen:34275 1229400ms

When this is happening it has a business impact because during the 30 minutes of execution of the slow query the author instance is unusable.

It would be nice to have a similar jmx mbean to track slow query in jcr but here to track the one performed on mongodb and keep a stacktrace then for those long query to find the culprit code that triggered this query. A threshold parameter could be defined to fix a limit response time for the query so that it get tracked.

On log level I think there are some "perflogger" existing, but not sure if those can include a stacktrace (maybe in TRACE level mode) for such purpose. That would help to quickly identify the thread and code that is behind the long running query in mongodb.




--
This message was sent by Atlassian JIRA
(v6.3.4#6332)