Race condition for active/passive pods when no leader is available about fissile

[mook-as]

So we basically need to update that script to call update_readiness false unless it actually passes a readiness probe (or there isn't one, in which case it get unset)…

[qu1queee]

I´not sure how reliable is the readiness probe, while the checks here could be successful even if the job is down. For example, if the bbs job for diego-api is down, the additional readiness will execute the head -c0 </dev/tcp/${HOSTNAME}/8891(see line), returning an exit code 0.(pls correct me if im wrong). I would probably move this block before monit checks, so it will ensure the label of skiff-role-active can be set to false, opinions?

[mook-as]

I was thinking of something like

• Set trap to set active to false on exit
• Current stuff
• Erase trap, set active to true
  Basically, try to ensure that we set the active flag however we exit, instead of just on the happy path. After all, the script (with errexit) is designed to abort early on failure, anyway.

Would that cover everything (and be less likely to be broken as we change the script)?

[qu1queee]

yes, sorry for the late response(I already saw the merged PR). Trapping that method on EXIT was a nice idea.