You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Currently we don't make much use of readiness/liveness probe, though there are major cases where if those are to be installed, it would help cluster admins or users recognize and diagnose what is going on when some component goes bad. Let me write some down as my first thought.
1). When vizier-db takes tens of seconds to boot up, which is quite normal especially if the components are cleanly deployed for the first time, vizier-core is unavailable in the meantime. We can tell vizier-core is ready once Katib manager successfully starts to listen.
2). Normal liveness probe against 0.0.0.0:6789 seems the best, though we cannot tell whether or not the backend db is healthy, which would heavily affect much of katib manager's gRPC methods.
The text was updated successfully, but these errors were encountered:
Currently we don't make much use of readiness/liveness probe, though there are major cases where if those are to be installed, it would help cluster admins or users recognize and diagnose what is going on when some component goes bad. Let me write some down as my first thought.
vizier-core
is ready once Katib manager successfully starts to listen.The text was updated successfully, but these errors were encountered: