A Subversion branch or tag is a copy of a particular repository path at a point in time. This is denoted as the addition of a particular path.
For example:
------------------------------------------------------------------------
r1577 | dgr02 | 2016-06-01 14:13:18 +0100 (Wed, 01 Jun 2016) | 1 line
Changed paths:
A /tags/SVN-HOOKS-1.1 (from /branches/svn-hooks-1.x:1576)
CBST-654 : Added issueType checking functionality.
------------------------------------------------------------------------
The deletion of a branch or tag is the corresponding removal of that particular path:
------------------------------------------------------------------------
r1586 | dgr02 | 2016-06-02 14:55:22 +0100 (Thu, 02 Jun 2016) | 1 line
Changed paths:
D /branches/CBST-417
Branch clean-up
------------------------------------------------------------------------
When users are looking to clean-up old tags they will often perform this in in en masse manner - for example (different repository):
------------------------------------------------------------------------
r42953 | luw07 | 2016-06-02 13:15:31 +0100 (Thu, 02 Jun 2016) | 1 line
Changed paths:
D /Database/DB_RDM/tags/snapshots/DB_RDM-106.0.19
D /Database/DB_RDM/tags/snapshots/DB_RDM-106.0.21
D /Database/DB_RDM/tags/snapshots/DB_RDM-106.0.22
D /Database/DB_RDM/tags/snapshots/DB_RDM-106.0.23
D /Database/DB_RDM/tags/snapshots/DB_RDM-106.0.31
D /Database/DB_RDM/tags/snapshots/DB_RDM-106.0.32
... 480 additional deletions ...
This causes FishEye to get in a mess because it is unable to retrieve all of the changes associated with the revision in a suitable time. In the example above, each tag has 12,000+ paths so the whole changeset (according to FishEye) will amount to over 6,000,000 changes.
This, understandably, can cause FishEye to timeout its processing of the revision as it's SVN Operation Timeout is by default 1 hour.
For a changeset containing 1,000,000s of items it's unlikely to complete the transfer of all of this data within an hour for FishEye to then process; however whilst this would mean that the changelog would stay up to date it means that it would be continually re-attempting index the paths within this one revision.
My Question
Okay - so what I want to know is what are people's recommendations for actions which I can take which will help FishEye to cope with these kind of changesets?
Things I can come up with:
- Increase FishEye's SVN Operation Timeout parameter to a very high value (e.g. 2 weeks) so that it actually manages to retrieve all of the data – this can however have an impact on the heap requirements for FishEye as it then needs to be able to cope with a very large changeset.
- Put a hook on the repository which detects whether a changeset is impacting multiple tags and reject the commit. This however won't cope with if someone (in the example above) were to delete the directory containing the tags – that would appear in the commit log as a single item deletion.
What other suggestions might folks have for how I can protect my FishEye instance / help users to work in a more intelligent manner?