[
https://issues.apache.org/jira/browse/OAK-8648?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16939481#comment-16939481
]
Julian Reschke edited comment on OAK-8648 at 9/27/19 2:09 PM:
--------------------------------------------------------------
As far as I can tell, there are two reasons why a sweep would be started:
1. Because there is no sweep revision yet on the root document (for that
cluster id). So this would be expected once for any existing cluster id
migrating to 1.8.*, or a a new cluster id.
2. Presence of "garbage commits" in the DocumentNodeStore instance.
Case 2) doesn't seem to be possible upon startup, which leaves us with 1).
[~mreutegg], does this sound right?
So for {{backgroundSweep()}}, I'd suggest generating a string that describes
the reason for the sweep, to be passed to {{forceBackgroundSweep()}}. The
latter already logs on startup, but only on DEBUG level; this is a bit weird,
as later on pogress reporting happens on INFO level. If
{{forceBackgroundSweep()}} indeed happens seldomly, maybe it would be best to
log uncondtionally on INFO level here?
was (Author: reschke):
As far as I can tell, there are two reasons why a sweep would be started:
1. Because there is no sweep revision yet on the root document (for that
cluster id). So this would be expected once for any existing cluster id
migrating to 1.8.*, or a a new cluster id.
2. Presence of "in doubt commits" in the DocumentNodeStore instance.
Case 2) doesn't seem to be possible upon startup, which leaves us with 1).
[~mreutegg], does this sound right?
So for {{backgroundSweep()}}, I'd suggest generating a string that describes
the reason for the sweep, to be passed to {{forceBackgroundSweep()}}. The
latter already logs on startup, but only on DEBUG level; this is a bit weird,
as later on pogress reporting happens on INFO level. If
{{forceBackgroundSweep()}} indeed happens seldomly, maybe it would be best to
log uncondtionally on INFO level here?
> Log reason why NodeDocumentSweeper.sweep is called
> --------------------------------------------------
>
> Key: OAK-8648
> URL: https://issues.apache.org/jira/browse/OAK-8648
> Project: Jackrabbit Oak
> Issue Type: Improvement
> Components: documentmk
> Affects Versions: 1.8.9
> Reporter: Vincent Frey
> Assignee: Julian Reschke
> Priority: Minor
>
> It can happen during a cluster node restart, that a
> NodeDocumentSweeper.sweep is triggered.
> Depending on the number of nodes, this can delay the start process for
> several hours.
> In order to understand what caused the sweep, it could be interesting to log
> (INFO level) the reason why it has been triggered.
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
