Health-check of connection on check-in about mobc HOT 10 CLOSED

After the check-out of a pool connection, users will directly interact with the DB connection by the Deref mechanism, so it's hard for me to trace and mark what state the original conn is.

And I am turning to the maintainers of Redis-rs for help.

from mobc.

Yes, the point of this issue is that mobc provides appropriate hook to discard connections on check-in.

from mobc.

We also have to investigate how this works for postgres. Without proposed changes, it either has the same bug as redis or will error on the first request after checking out connection (latter is better as it isn't a security issue, but is still suboptimal).

from mobc.

Ok, I added a new hook to do some cleaning works on the check-in, but it defaults to be disabled.

from mobc.

Thanks. I don't understand the motivation for it to be opt-in, though. I believe it should be always enabled (i.e. there should be no configuration to disable it). This check should be cheap and ignoring it only degrades the performance and stability of the application. (Comparing to health check on check-in which might be slow as it sends things to the network and is safe to ignore if your connections are never idle for too long).

from mobc.

Since I chose to use an async hook at last, but I don't want to pay for the async transform cost. Do you think it is necessary to support async in this hook? Or it should just return boolean like what r2d2 does.

from mobc.

I like the idea to restrict it to sync code. Async here is a large monitoring issue, it's harder to account tasks spawned by the destructor in case they are stuck for some reason.

from mobc.

Hey! Sorry for late.

I have made some changes to the code, you may take a look at it.

#33

Suggestions for revision are welcome.

from mobc.

Suggested a different name in PR. Other than that, looks good. Thanks!

from mobc.

Thank you as well.

from mobc.