Opened 14 years ago
Closed 14 years ago
#6983 closed defect (duplicate)
deadlock on readReadyLock and socket lock
| Reported by: | Mark Spieth | Owned by: | danielk |
|---|---|---|---|
| Priority: | major | Milestone: | 0.22 |
| Component: | MythTV - General | Version: | head |
| Severity: | high | Keywords: | |
| Cc: | Ticket locked: | no |
Description (last modified by )
there are 2 problems with customevent
there is a deadlock between socket lock and readReadyLock and can happen on multi core machines. 4 cores or more with mythweb scheduling a program causes it to happen most times. doesnt seem to happen on a 2 core.
There may be a better way to fix this, but the attached solution works.
the 2nd is that customevent tries to send messages on sockets which are not connected (thus there are no listeners).
it gives
write -> -1 565 BACKEND_MESSAGE[]:[]ASK_RECORDING 1 0 0 0[]:[]Ten
News[]:[][]:[]A...
writeStringList: Error, called with unconnected socket
type errors. pointless doing this
see also #6894 for this one.
Attachments (1)
Change History (5)
Changed 14 years ago by
| Attachment: | 19-backend-rec-event.patch added |
|---|
comment:2 Changed 14 years ago by
| Milestone: | unknown → 0.22 |
|---|---|
| Owner: | changed from Isaac Richards to danielk |
| Status: | new → assigned |
| Version: | unknown → head |
comment:3 Changed 14 years ago by
Mark, can you provide a backtrace of the process in deadlock + the mythbackend -v socket,network,extra log? If you still have the core around and can't repro now the backtrace is more important.
There is an existing report of mythweb deadlocking the backend, but in that report we were sending data on a dead socket so that was my first concern.
I plan to set up mythweb again next week and investigate, but any info I have before then will be useful.
Refs #6969.