Scaling up Stash: What are the limits?

[GeorgG]

Hi everyone,

I'm using Stash at the company I work with and I really love it. So far we're about 200 users, about a quarter of which use it regularly. We're wondering what the limits of scaling up Stash to a wider audience are (our company has 300k+ employees). I'd appreciate any experiences you have gained in using Stash in an environment with, say, 30k+ users.

Thanks in advance for your support and cheers,

Georg

[cofarrell]

Hi Georg,

I won't add this as as "answer", because it would be good to hear from other customers, but just my/our 2c.

By far and away the biggest problem we've had in Stash is related to continuous integration. As the number of builds that clone repositories from Stash increases that puts a huge load on Stash due to the CPU intensive operation of Git recalculating a new pack file (every time). To compensate we now cache those pack file, which has increased the number of continuous clones significantly.

The most important document you can read is the following:

https://confluence.atlassian.com/display/STASH/Scaling+Stash

Apologies, this will sound like a marketing slogan, but we've tried to build Stash to scale from the beginning. This has led to do slightly strange sounding things like fork to a git process rather than do the "obvious" thing and use JGit, which actually has memory problems at scale. I wouldn't say we've solved all of the problems yet, but it's always one of the first things we consider when adding features, and we're working to push the limits of Stash/Git as far as we can.

Looking forward to hearing what experiences other users will add.

Charles

[cofarrell]

A related Answer:

https://answers.atlassian.com/questions/163853/speccing-a-stash-server