[mythtv-users] everyday i have to make sure myth worked the day before

[Michael T. Dean]

On 04/19/2011 11:37 PM, Brian J. Murrell wrote:
> Yeah, most definitely back with a vengeance.  Again tonight:
...
> 2011-04-19 22:04:30.209 MythCoreContext: Connecting to backend server:
> 10.75.22.2:6543 (try 1 of 5)
> 2011-04-19 22:04:30.261 MythCoreContext: Connecting to backend server:
> 10.75.22.2:6543 (try 1 of 5)
> 2011-04-19 22:04:37.214 MythSocket(892d960:11): readStringList: Error,
> timed out after 7000 ms.
> 2011-04-19 22:04:37.215 Protocol version check failure.
>                          The response to MYTH_PROTO_VERSION was empty.
>                          This happens when the backend is too busy to
> respond,
>                          or has deadlocked in due to bugs or hardware
> failure.
...
> And so on and so on.  And missed recordings to go along with it, further
> indicting the reliability of this whole setup for the family here.  :-(
>
> Is nobody at else seeing these types of errors?  Are there no developers
> at all interested in what's going on here?  There has got to be a bug
> buried in here somewhere.
>
> I've even opened a ticket (like somebody requested) and put all of this
> detail into it and it's not even seen a peep.

These are the socket errors that have been plaguing some users on some 
systems for many versions, now (since 0.21-fixes), and are likely due to 
changes in libc or the kernel or system configuration or something that 
MythTV code may not have kept up with.  They only seem to occur on some 
systems.  Those that see them, see them a lot.  Others never see them.  
(I'm in the latter group.)  It may well be timing-related (which would 
make sense since you said that use -v most prevents the problem--which 
could be because the additional verbosity changes the timing of 
operations enough to prevent the problem).

Basically, what happens is mythbackend's socket code gets into a bad 
state where nothing can connect, so when new clients (in the case shown 
in your logs, mythfrontned, but sometimes even clients like 
mythpreviewgen--whose logging data is munged into the mythbackend log, 
often making users think that mythbackend is logging those errors) try 
to connect, they fail to connect to the mythbackend.  In some cases the 
mythbackend will continue to operate "internally" (but unable to "hear" 
communications from outside the mythbackend process).  In other cases, 
the mythbackend will stop working properly and fail to record shows,  etc.

When the socket issues first started happening--they seemed to come out 
of nowhere as people started upgrading their systems to newer libraries 
or kernels or something--they were very common on most systems.  Then 
Daniel K rewrote the socket code, and this solved the issue for the 
majority of users.  Unfortunately, the issue still occurs for some 
users.  Because the issue is very system/timing/... dependent, we 
haven't found a way for someone who hasn't been seeing the issues to 
reproduce them.  Therefore, it is going to be /very/ difficult for 
anyone to fix until someone who actually sees the problem can diagnose 
(and/or fix) it.

That said, it's possible that "changing up your system" may allow you to 
run mythbackend without the issue.  (Unfortunately, I don't know what 
all you might need to change up--processor,  distro version, distro, RAM 
(speed or capacity), NIC, ...)

Mike