Closed duncdrum closed 5 years ago
Regards query-timeout="-1"
I think we need to be clear what it actually means.
If will only abort a query that is in a healthy and running state, there are other reasons why a query may lock up, and in those instances query-timeout
won't have any effect.
Actually query-timeout
is rather dangerous as you can end up with unexpected inconsistencies if a query running an XQuery Update or XUpdate is aborted. The database itself won't be corrupted, but only some of the changes will have been applied, which means that your data may not be consistent.
Also if a database has locked up, you should not terminate it and try and continue. That will likely just result in a corrupted database.
Regards sync-on-commit
I think its description is misleading. The sync happens asynchronously to the commit. So there is no guarantee of anything.
Regards force-restart
... it should never be the default!
If will only abort a query that is in a healthy and running state
that settles query-timeout
then, let's keep it as is.
as for the others thats why I was looking for feedback thanks @adamretter
let's keep them as well.
The following line is giving me pause:
query-timeout
while convenient, it means that orchestrators will probably keep a dead node around potentially indefinitely. I think in a container context this should have high positive value.Similarly,
sync-on-commit
My intuition (need to do some testing on this) is that sync-on-commit might actually speed up things in containerland, but I'm not sure how it would play out in different orchestration scenarios, what if a worker node never reaches the threshold?Enabling
force-restart
on the other hand might make sense, since new instances are created and killed much more frequently compared to traditional server contexts.