Comment 17 for bug 416990

Revision history for this message
Michael Hudson-Doyle (mwhudson) wrote :

We've landed the change to use bzr 2.1b3 so this should improve a bunch after the next rollout.

We can also probably put a cap in place on the consumption of each process like we do for the puller. The user experience won't be great when their pull which will presumably have been running for a long time suddenly croaks because the process at the far end got aborted though.

I don't see the issue that each process takes a noticeable hunk of ram and we run one process per connection going away.

The long term solution still has to be running the bzr lp-serve process on multiple hosts.