SVN Indexing slow on changesets involving a large number of files

Question

Hi all,

I've just started using Fisheye 3.0.1.

I noticed that SVN repository indexing is slow AND CPU intensive which sounded strange to me, so I did some sampling.

The SVN repository is available via HTTPS which is known to be slow, so I expected bad IO performance but not a steady 100% CPU usage.

I found out that during SVN indexing Fisheye is often executing com.cenqua.fisheye.svn.util.ChangePathUtil.isSubPath(..), which consumes 100% of a CPU.

Typical stack trace:

at java.lang.String.equals(String.java:1017)
	at com.cenqua.fisheye.svn.util.ChangePathUtil.isSubPath(ChangePathUtil.java:18)
	at com.atlassian.fisheye.svn.Svn2MessageContext.isApplicableChangePath(Svn2MessageContext.java:109)
	at com.atlassian.fisheye.svn.Svn2MessageContext.getApplicableChangePaths(Svn2MessageContext.java:153)
	at com.atlassian.fisheye.svn.Svn2MessageContext.isPartOfAdd(Svn2MessageContext.java:428)
	at com.atlassian.fisheye.svn.Svn2Infill2Processor.getInfo(Svn2Infill2Processor.java:363)
	at com.atlassian.fisheye.svn.Svn2Infill2Processor.infillChangeSet(Svn2Infill2Processor.java:187)
	at com.atlassian.fisheye.svn.Svn2Infill2Processor.access$100(Svn2Infill2Processor.java:78)
	at com.atlassian.fisheye.svn.Svn2Infill2Processor$1.perform(Svn2Infill2Processor.java:121)
	at com.atlassian.fisheye.svn.Svn2Infill2Processor$1.perform(Svn2Infill2Processor.java:110)
	at com.cenqua.fisheye.cache.BaseRevisionCache.withDbWriteLock(BaseRevisionCache.java:967)
	at com.atlassian.fisheye.svn.Svn2Infill2Processor.process(Svn2Infill2Processor.java:110)
	at com.atlassian.fisheye.svn.Svn2Scanner.slurpRepository(Svn2Scanner.java:210)
	at com.atlassian.fisheye.svn.Svn2Scanner.doSlurpTransaction(Svn2Scanner.java:177)
	at com.cenqua.fisheye.rep.BaseRepositoryScanner.ping(BaseRepositoryScanner.java:85)
	at com.cenqua.fisheye.rep.BaseRepositoryEngine.doSlurp(BaseRepositoryEngine.java:92)
	at com.cenqua.fisheye.rep.RepositoryEngine.slurp(RepositoryEngine.java:408)
	at com.cenqua.fisheye.rep.ping.IndexingPingRequest.doRequest(IndexingPingRequest.java:25)
	at com.cenqua.fisheye.rep.ping.IncrementalPingRequest.doRequest(IncrementalPingRequest.java:30)
	at com.cenqua.fisheye.rep.ping.PingRequest.process(PingRequest.java:58)
	at com.cenqua.fisheye.rep.RepositoryHandle.processPingRequests(RepositoryHandle.java:228)

Is there any fine tuning that I can do via configuration of the repository?

Thanks in advance.

Best regards,

Marco

Sempla IT · Answer

Hi Nick,

thanks for your reply. I started indexing from a higher changeset and solved the problem in this way.

BTW, changesets that count 10K files, as an order of magnitude, are still slow but Fisheye processes them without major issues.

The problem with 300k files changesets is that CPU is almost constantly at 100%, suggesting me that com.cenqua.fisheye.svn.util.ChangePathUtil.isSubPath(..) is working on in-memory objects. Also, heap occupation grows and exceeds -Xmx1g settings (no user activity). This makes me suspect that com.cenqua.fisheye.svn.util.ChangePathUtil.isSubPath (or the code block that invokes it) could be further optimized.

Slow indexing because of poor IO is not currently a major concern for me, what's more important IMHO is not having a direct dependency between changeset size and heap occupation.

Thanks again, regards.

Marco

Nick · Answer

Hi Marco - 300k files is indeed a huge changeset.

Has FishEye finished processing this changeset now and your CPU cooled down a little ?

Further, indexing will be much faster for you if you can use svnsync to sync your svn repo local to the fisheye server.

Cheers,

Nick

CelsoA · Answer

Hi,

There are some configurations that can be done to tune your FishEye instance.

They are described here:

https://confluence.atlassian.com/display/FISHEYE/Tuning+FishEye+performance

Please give it a try. :D