Performance testing Git

[Lt. Manlove]

I have an instance of Bitbucket Server Data Center with approximately 6000 developers but it is looking to double in the next year. I'm looking to performance test the instance and I realize I need to simulate various Git and application operations to get a realistic idea of how the servers are performing.

The application operations such as creating project, adding users and creating pull requests can be covered with jMeter but that doesn't incorporate the Git operations. Does anybody have any experience or ideas on moving this forward?

 

[Dj_Jestar]

What on earth are you doing with 6,000-12,000 developers?

Go read GitHub's engineering blog. You are extremely unlikely to get any kind of info about git on that kind of scale from this forum.

 

[Lt. Manlove]

Indeed it is a very large company!

I'm thinking Git operations can be simulated using Bash but scaling this out and benchmarking / reporting is the challenge I'm facing.

Maybe this has never been done before but I'm looking for some inspiration to help

 

[Lt. Manlove]

Perhaps I can expose a bash script which would cover various Git operations that can be called with jMeter

 

[Dj_Jestar]

I doubt you're going to be able to test the entire thing. You'll have to fragment and extrapolate results.

Is there anything in particular you want to test? Replication between nodes or something? I mean you say git, but there's multiple functions within git you could be concerned about.

 

[mid_gen]

Yeesh that is big. In the past I've worked on projects that have used HP and CA load testing tools that can be configured to do pretty much whatever kind of load you want. Building the test scripts for that kind of volume to meaningfully replicate a real load is going to be a bit of a mare.

I'd probably be on the phone to Atlassian if you're that big a customer!

 

[planty]

Surely those 6,000 developers are spread across different projects/repositories/etc? Can't you just have multiple, smaller instances?

 

[Etheco]

recent performance on bitbucket alone has been very "slow" and only 2 of us use it and have noticed slowness on commits and merges.

This is for the free tier not sure if there paid service escalates you to better hardware etc though.

 

[Craig321]

I have an instance of Bitbucket Server Data Center with approximately 6000 developers but it is looking to double in the next year. I'm looking to performance test the instance and I realize I need to simulate various Git and application operations to get a realistic idea of how the servers are performing.

The application operations such as creating project, adding users and creating pull requests can be covered with jMeter but that doesn't incorporate the Git operations. Does anybody have any experience or ideas on moving this forward?

Surely that's a question you can approach Atlassian with? They should be able to either assure you their platform will handle it, or give you some stats?

 

[peterwalkley]

I think you might be approaching this the wrong way. Why try and generate artificial load when the performance and metrics on your current system can be monitored with today's developer base - then (simplistically) double all the numbers for the planned increase.

If I've understood this correctly, you have your own version of bitbucket in your data centre: What kind of load is that putting on the current hardware, how often do you need to do server restarts because of user complaints, what rate are you filling up your storage capacity etc ?

Bitbucket runs their own public facing internet service akin to github - and I would suspect the traffic there is higher than your 12K developers figure, so its really a capacity measurement and planning exercise you need to do with your hardware - the same as for any server centric software used by lots of people. If you don't have those skills in your organisation, then its worth finding some specialist consultants to help.

My experience as a developer in a big company is that the network connection to the source code repository is the choke point, not the software itself unless its running out of storage space.