InfluxDB builds its indexes during startup and that might take a while. With the default configuration the liveness probe check is restarting the pod before it is ready.

Increasing the liveness probe initial delay to 120s solved this issue. I've rolled out the change to all EFD instances.