Ticket #233 (closed: knowledgebase)
|Reported by:||moorman||Owned by:||moorman|
What we're doing about Subversion statistics
Framing the problem
We have recently been receiving inquiries about a low count on Subversion read statistics. In examining this problem, we found a design flaw that started with the datacenter migration. Work is being scheduled to augment the statistics system to work around this deficiency. Historical numbers will remain in their lower count state, but future stats will be correct.
Subversion stats offline
We've taken the Subversion statistics processing offline in the mean time, to ensure that correct numbers are imported in to the stats system. We'll bring Subversion stats back online after the necessary code changes have been made to generate proper read activity counts.
The underlying issue
The statistics system tracks read activity through log analysis. With the datacenter migration, logging of read operations were split between http and https logs. The stats system has been reading only one of these two logs. Since write ops only occur from authenticated users (https), these stats are not impacted.
Changes will be made to the stats system such that the two logs are processed to populate the single Subversion read activity field. Once the change has been deployed, we'll process the stats for the time period since we turned Subversion stats off. Historical data won't be corrected, but the read activity numbers will be correct moving forward.