Improve FishEye repository scanning performance using Hadoop or something else

[Justin Johnson]

We have almost 600 Subversion repositories, some that are 30 GB or more in size. The last time we tried to support FishEye the scanning was taking over a month and we decided it wasn't viable.

I'm wondering if Atlassian has considered supporting the use of Hadoop to improve scanning performance? Revisions seem like a unit of work that could be distributed to various nodes for processing.

Other than Hadoop, does anyone have other suggestions for improving the performance?

[Justin Johnson · Accepted solution]

Atlassian has not responded so presumably it isn't possible.

[JamieA]

Bah, accidentally deleted my comment. Be sure that all your repos are structured in the way it likes: https://answers.atlassian.com/questions/19281/how-can-i-reduce-the-size-of-the-fisheye-indexes

I ended up writing something that automatically generates the exclusion rules.

30Gb repos doesn't tell us much - if it's binary files fisheye doesn't care, if it's metadata it does.

[Renjith Pillai]

Some minor hints (svn)

• http over https improved the performance a lot
• svn also has a direct file access mechanism, if you can afford to access the svn server's disks directly from the fisheye servers

[Justin Johnson]

We were using file:// URLs and the repositories were stored on SAN, so that wasn't the issue. Thanks though.

[JamieA]

FWIW - anything other than file:// access is pretty much a non-starter for real life svn repos.

[JamieA]

Be sure all your repos are structured in the way that fisheye likes: https://answers.atlassian.com/questions/19281/how-can-i-reduce-the-size-of-the-fisheye-indexes . I ended up writing something to automatically generate the exclusions.

30Gb repos doesn't really tell us anything useful. It could be binary files, in which case it makes no odds to FE, or metadata, in which case it will kill it.

[Raju Kumar]

Execellent ! I am truly impressed that there is so much about this subject that has been revealed and you did it so nicely with so considerably class.and visit more http://hadooptraininginhyderabad.co.in/

[Sergey Svishchev]

There's feature request for 'scanning agent' -- https://jira.atlassian.com/browse/FE-1988 -- vote for it if you like it.

So far the only 'distributed scanning solution' is to spin up an aux instance and do the initial scanning there.

[mahaao]

Hi Justin,

We also met some problem with Big repositories.

We have over 300 repos, and each one is bigger than 10G.

We are also interesting about your idea of "using Hadoop".

Do you know how to use Haddoop on Fisheye/Crucible?

How Hadoop will improve the scanning performance?

[Justin Johnson]

As far as I know it isn't possible. I was hoping Atlassian would weigh in on the possibility.

[Justin Johnson]

I found http://confluence.atlassian.com/display/FISHEYE/Best+Practices+for+FishEye+Configuration for general performance tips. I'm still interested in the idea of using Hadoop though. Atlassian?