Stefan Egli created OAK-2359:
--------------------------------
Summary: diffImpl is inefficient when there are many split
documents
Key: OAK-2359
URL: https://issues.apache.org/jira/browse/OAK-2359
Project: Jackrabbit Oak
Issue Type: Bug
Components: core
Affects Versions: 1.0.8
Environment: 1.0.8.r1644758
Reporter: Stefan Egli
Priority: Critical
As reported in OAK-2358 there is a potential problem with revisionGC not
cleaning up split documents properly (in 1.0.8.r1644758 at least).
As a side-effect, having many garbage-revisions renders the diffImpl algorithm
to become very slow - normally it would take only a few millis, but with nodes
that have many split-documents I can see diffImpl take hundres of millis,
sometimes up to a few seconds. Which causes the observation dequeuing to be
slower than the rate in which observation events are enqueued, which results in
observation queue never being cleaned up and event handling being delayed more
and more.
Adding some logging showed that diffImpl would often read many split-documents,
which supports the assumption that the revisionGC not cleaning up revisions has
the diffImpl-slowness as a side-effect. Having said that - diffImpl should
probably still be able to run fast, since all the revisions it should look at
should be in the main document, not in split documents.
I dont have a test case handy for this at the moment unfortunately - if more is
coming up, I'll add more details here.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
