Improve reliability, add more tracing/debugging
-
Stop prodding OBS backend when state is building
When OBS is told a worker isbuilding
it thinks it’s "building for someone else" and marks in asaway
. Seebs_repserver
for more details. -
Return cached worker info immediately, update in background
Workers already update their state when it changes, and this state propagates through OBS proxies, so we can safely return it immediately when requested by the warden, and schedule an update in the background.This should mitigate timeouts when the warden prods OBS proxy server prods OBS proxy client prods worker, and while the reply climbs back through the interpipes, warden times out and kills the job.
Edited by Emanuele Aina