search

ClusterLabs / fence-agents

Fence agents
104 stars 160 forks source link

fencing: add stonith_status_sleep parameter for sleep between status calls during a STONITH action #402

Closed oalbrigt closed 3 years ago

wenningerk commented 3 years ago

Is the intention of this to prevent bugging a physical fencing device / cloud api with status checks so that it gets issues or the status checks are considered dos? If yes do we have this issue as well when actions are retried by pacemaker? Would it thus be interesting to store that somehow? Maybe even have something provided by fenced so that fence-agents can memorize data?

oalbrigt commented 3 years ago

It is intended to avoid getting throttled by some cloud APIs by sending fewer requests.

wenningerk commented 3 years ago

It is intended to avoid getting throttled by some cloud APIs by sending fewer requests.

Aah ... so 'just' as to make a single stonith action quicker and not to avoid triggering some mitigation/locking mechanisms to kick in that would have larger impact ...

oalbrigt commented 3 years ago

Throttling might make it fail in some situations I guess, so it's to avoid any issues that might happen if you hit request quota or similar.

wenningerk commented 3 years ago

Throttling might make it fail in some situations I guess, so it's to avoid any issues that might happen if you hit request quota or similar.

Guess I got it now - you mean explicit throttling by the api - just as I thought initially but less harsh. Then it might be useful to check also over multiple actions. RAs are using private node attributes to memorize things - right? What are cloud APIs checking? Really the time between 2 accesses or rather something like not more than 5 within 10s ...