How to cache M2 dependencies between agents?

[ejoncas]

Hi there,

I'm trying to get the .m2/repository to be cached somehow so I don't need to download all the jars every time a new agent comes up. This process is quite lengthy and every morning where the instances gets created we have to wait 30 minutes for the first few builds to succeed.

After that, everything works well as long as the agent does not get killed.

Would it be possible to package the .m2/repository before killing the agents and upload it to S3 for the next morning? Downloading a few GB from S3 is quick, but not so much if download from our artifactory.

Or perhaps we should cache that within the docker image and push it to ECR?

Any suggestion/ideas?

[toolmantim]

Good question! It looks like at the moment this example just covers a single agent, using a host mount:

https://github.com/buildkite/maven-example/blob/master/.buildkite/docker-compose.yml#L7-L8

To get this working across agents, there's basically those two options you mentioned. But perhaps the S3 option might be best, because it'd still work with this host mount setup?

In that case you could add a post-checkout repository hook and a pre-exit repository hook:
https://buildkite.com/docs/agent/v3/hooks#repository-hooks

Perhaps the post-checkout could have:

aws s3 sync "s3://some-bucket/.m2/$BUILDKITE_BRANCH" "$HOME/.m2"

And perhaps the pre-exit could have:

aws s3 sync "$HOME/.m2" s3://some-bucket/.m2

I'm not sure about concurrency issues with it, but perhaps something like that could be a good start?

You could definitely try using a Docker image as well, but given this example just uses the standard maven image you'd have a bit more work to do there. It's definitely possible though.