Repository navigation
UCP/EP/FLUSH: fix too early completion for failed EP - #5952
Conversation
| } | ||
|
|
||
| ASSERT_UCS_OK(UCS_PTR_STATUS(req)); | ||
| /* close request can be completed with any status depends on peer state */ |
There was a problem hiding this comment.
ok for now, in next PR by @Artemy-Mellanox close(FORCE) will always return NULL (success)
There was a problem hiding this comment.
but close(flush) can be completed with err too
There was a problem hiding this comment.
yes, close (flush) can complete with err
can u pls open another PR with doc clarification on this:
e.g close(flush) operation can return failure if the remote side disconnected at same time, but nevertheless all resources associated with the endpoint will be destroyed.
close(force) will always succeed
There was a problem hiding this comment.
IMHO this way we add extra commitment without any benefit.
maybe don't differentiate flush/force return status and just mention that "close request can be completed with error if the remote side disconnected at same time but nevertheless all resources associated with the endpoint will be destroyed"?
|
/azp run |
|
Azure Pipelines successfully started running 1 pipeline(s). |
6960ee4 to
a89ce65
Compare
|
/azp run |
|
Azure Pipelines successfully started running 1 pipeline(s). |
|
failure was #3897 |
What
fix too early completion for failed EP
Why ?
in case if scheduled err handler was not be invoked yet, the EP can be closed with requests in pending
reproducer is implemented in #5931 but that PR has other dependency