fix copyRowsFunc hangs bug #611
Conversation
go/logic/migrator.go
Outdated
| @@ -1135,6 +1140,9 @@ func (this *Migrator) iterateChunks() error { | |||
| } | |||
| // Enqueue copy operation; to be executed by executeWriteFuncs() | |||
| this.copyRowsQueue <- copyRowsFunc | |||
| if !hasFurtherRange { | |||
| break | |||
| } | |||
There was a problem hiding this comment.
Since this code is outside copyRowsFunc, then it only skips enqueuing of new copyRowsFunc functions, but still there could be multiple copyRowsFunc instacnes inside this.copyRowsQueue that haven't been evaluated yet.
This is why I believe this PR mitigates the potential for deadlock but does not actually solve it.
I think we should check hasFurtherRange from within copyRowsFunc to solve it completely.
There was a problem hiding this comment.
@shlomi-noach yes, you are right. I will make some modification.
wfxiang08
force-pushed
the
dev_copyrow_hangs
branch
from
2b78c3c
to
c7dff99
Compare
|
@shlomi-noach I have make some modification, will you pls. have a review. |
shlomi-noach
temporarily deployed
to
production/mysql_role=ghost_testing
|
Apologies for the time it took, I was sidetracked. On the flip side, this PR has been actively tested for some 50 days, so it's been well hammered. Merging! |
Related issue: #610
Description
When copyRowsFunc finishes copy rows, the status
hasFurtherRangeis kept across different iterations of copyRowsFunc, and whenhasFurtherRangeis false, never try to callCalculateNextIterationRangeEndValues.When
hasFurtherRangeis false, the original table also might be locked and in this caseCalculateNextIterationRangeEndValueswill be hang forever.