kubernetes: Delay health check for 180 seconds to account for long re…

[tgraf]

…store

If many endpoints have been running before restart, endpoint restore can
eventually take a while due to contacting the kvstore and compilation of
endpoints. Bump the initial delay befor health checks occurs to 180
seconds to avoid unneeded pod restarts.

Signed-off-by: Thomas Graf thomas@cilium.io

[ianvernon]

It makes sense to have a longer time between health checks to avoid unnecessary restarts, but a jump from 10-180 seconds is quite a big one. Does it take almost three minutes to get endpoints restored after restarting ? This means that a user can have Cilium in a bad state for almost three minutes before it restarts.

[tgraf]

It makes sense to have a longer time between health checks to avoid unnecessary restarts, but a jump from 10-180 seconds is quite a big one. Does it take almost three minutes to get endpoints restored after restarting ? This means that a user can have Cilium in a bad state for almost three minutes before it restarts.

This is only the initial delay before the first ever health check is being performed. Cilium should error out with a fatal error message if there is any error during bootstrap.

[ianvernon]

This is only the initial delay before the first ever health check is being performed. Cilium should error out with a fatal error message if there is any error during bootstrap.

OK makes sense, thanks !