We find this happens often with steps like: 

sh "docker-compose -f docker-compose-ci.yml logs containername"

Or test steps that produce a lot of output.

The outcome is often that a job uses ~5GB of space until it hits our global 1h timeout, even though that step does not have that much output. It's definitely Jenkins spinning on repeating the last few ~100 lines, although the process has already returned.

As a workaround, if the above line is replaced with:

sh "docker-compose logs containername | gzip > containername.log.gz"
archive("containername.log.gz")

then we don't ever experience the hang.

But for test output that isn't ideal.

Note that this issue also happens if the command itself has a timeout:

go test -timeout 10m

It can still hang for an hour, spewing tons of logs that just repeat the last few pages.