Ensure context isn't exhausted via concurrent query as opposed to sentinel query

[gnpaone]

Switch from sequential sentinel query to concurrent query to avoid context exhaustion and connect to new master redis after redis failover

[gnpaone]

Hey @ndyakov tried to fix your review points in #3174

[ndyakov]

Please add a test , it may be a setup of three sentinels, two of which are unreachable. Define the context in such a way, to get context.DeadlineExceeded error on GetMasterAddrByName. Such setup should fail prior to your PR and succeeded with the introduced change, correct?

[kwenzh]

Please add a test , it may be a setup of three sentinels, two of which are unreachable. Define the context in such a way, to get context.DeadlineExceeded error on GetMasterAddrByName. Such setup should fail prior to your PR and succeeded with the introduced change, correct?

The current problem is that the faulty sentinel node is in the front, which will cause context exhaustion. Even if the last sentinel can correctly provide the master address, don't we consider this situation?

[kwenzh]

I think that once any sentinel returns the master address correctly, it should be considered finished and other coroutines should be notified of the end. Even if the context timeout occurs afterwards, we should return the master address that has been queried, right?

[kwenzh]

I think that once any sentinel returns the master address correctly, it should be considered finished and other coroutines should be notified of the end. Even if the context timeout occurs afterwards, we should return the master address that has been queried, right?

with issue： #3172

[gnpaone]

I think that once any sentinel returns the master address correctly, it should be considered finished and other coroutines should be notified of the end. Even if the context timeout occurs afterwards, we should return the master address that has been queried, right?

with issue： #3172

Maybe should I use something like

ctx, cancel := context.WithCancel(ctx)
defer cancel()

...

cancel()

to stop goroutines early, once success is achieved ie sentinel returns the master address correctly?

[kwenzh]

once success is achieved ie sentinel returns the master address correctly?

yeah，once success returns the master address correctly，That's exactly what I meant！

[gnpaone]

Done

[ndyakov]

@gnpaone can you please add the issue in the description of this PR. I will review the latest changes later today and get back to you if there are any comments.

[gnpaone]

Updated the description @ndyakov