Dealing with this was delayed due to Canonical's end-of-year holiday period.
Earlier this week we tried a temporary change that reverted the socket timeout added in https://code.launchpad.net/~cjwatson/launchpad/+git/launchpad/+merge/433219 (the timeout value had later been increased to 15 minutes). That seemed to fix translation exports. Now, this is very strange indeed: timestamps in logs indicate that the operation completes comfortably inside the timeout, so just having a timeout on the socket should make no difference to anything, and yet it does.
My plan is to disable this timeout for scripts for now, to get us back to a known state. I'm not really happy about this since it restores our previous awkward handling of internal network outages, but that was easier to cope with than this failure mode.
Dealing with this was delayed due to Canonical's end-of-year holiday period.
Earlier this week we tried a temporary change that reverted the socket timeout added in https:/ /code.launchpad .net/~cjwatson/ launchpad/ +git/launchpad/ +merge/ 433219 (the timeout value had later been increased to 15 minutes). That seemed to fix translation exports. Now, this is very strange indeed: timestamps in logs indicate that the operation completes comfortably inside the timeout, so just having a timeout on the socket should make no difference to anything, and yet it does.
My plan is to disable this timeout for scripts for now, to get us back to a known state. I'm not really happy about this since it restores our previous awkward handling of internal network outages, but that was easier to cope with than this failure mode.