[Buildbot-devel] Unreliable reconfig?

[David Coppit]

Hi all,

We're running a heavily modified 0.8.1 version. We deploy new or modified builds on a daily basis.

We've noticed that after a few reconfigs, things start to go wonky:

- Buildslaves may go idle with pending jobs. Restarting the buildslave process on the slave does not help. The master must be restarted before the slave will pick up jobs.
- Buildslaves may get into a state where they report that they are performing jobs, but are not actually doing anything.

The latter is particularly painful because it's hard to see that the slaves are in trouble.

We've been relying on restarts instead of reconfigs, but that causes all the in-flight builds to be rebuilt.

Has anyone else run into problems with reconfig?

Should we employ some idling process before a reconfig to reduce the chance of buildslave problems?

Are there certain kinds of master.cfg changes that could cause these symptoms?