Hi Marc, On Fr, 2016-12-30 at 13:57 +0100, Marc Haber wrote: > On Fri, Dec 30, 2016 at 01:26:09PM +0100, Stephan Sürken wrote:
(...) > That would be: > > > > Dec 28 13:44:29 spinturn mini_buildd.call (0114): > > WARNING : ? gpg.. (stderr): gpg: can't connect to the agent: IPC > > connect call failed > I don't have a solution for that other than to retry at least once. > Maybe too many operations in too short time confusing the agent? coincidentally doing some testing on another platform, I also experienced this behaviour (again), so I tried to do some debugging. Bottom line 1st: I could not definitely pin it down. However, at least on the system I experience this, a "just try again after some time"- style workaround seems to work. This workaround will be in 1.0.29 (now in current snapshots from hellfield). With this workaround being applied, look for "Retrying" in the log to actually still see the bug. On the system where this is happening, this is currently the simplest way for me to reproduce this: 1. Run attached ./signtest as user mini-buildd. 2. Rebuild any package. "signtest" will then eventually error out with the described error, at least most of the times. Without any building action, signtest will just run successfully forever. I am guessing this might be some sort of race condition with the gpg stuff sbuild is doing; however, it could also be some esoteric/kernel thing, as I cannot for the life of me get the bug occur on any other systems yet. [The system I actually can reproduce the bug is container-based (openvz), with the host using some older patched kernel.] Hth, S
signtest
Description: application/shellscript