[mythtv-users] Playback of recordings ends a short way into them

[Rich Freeman]

On Wed, Nov 12, 2014 at 1:37 AM, Richard Townsend
<townsend at astro.wisc.edu> wrote:
> 2014-11-11 23:37:15.747353 E [2187/4174] HDHRStreamHandler threadedfilewriter.cpp:239
> (Write) - TFW(/mnt/mythtv/tv/1571_20141112053000.mpg:68): Maximum buffer size exceeded.
>                         file will be truncated, no further writing will be done.
>                         This generally indicates your disk performance
>                         is insufficient to deal with the number of on-going
>                         recordings, or you have a disk failure.
> 2014-11-11 23:40:09.661866 W [2187/4171] TFWWrite threadedfilewriter.cpp:571 (DiskLoop) -
> TFW(/mnt/mythtv/tv/1571_20141112053000.mpg:68): write(58092) cnt 1149 total 67108856
>  -- took a long time, 231384 ms
>
> I’ve been trying to see if my NFS performance is a bottleneck,
> but it doesn’t seem to be. Is this possibly an issue with the HD
> HomeRun?
>

I used to have that kind of problem all the time with multi-disk
arrays and multiple recordings.  I found that disabling mythtv's fsync
calls and increasing the buffer size fixed the problem for me. I did
have a conversation with a developer a while ago and they indicated
that this change could be problematic, so they wouldn't recommend it
for general use.

Writes to RAID tend to be happier with larger write sizes and longer
latencies, since the bandwidth of all those drives is high, but the
cost to seek them all is also high.  I think that all those fsync
calls basically kill the write cache and make it harder for the kernel
to manage the IO queue.  On the other hand, the mythtv devs put in the
fsync calls because their experience was that the kernel doesn't do
such a great job with the queue anyway.

I also run my mythbackend process with realtime IO scheduling class.

If you want to try it, here is my patch, but you're probably on your
own with this one:

--- mythtv/libs/libmythbase/threadedfilewriter.cpp.orig 2014-03-21
18:21:22.067061254 -0400
+++ mythtv/libs/libmythbase/threadedfilewriter.cpp      2014-03-21
18:21:47.325630103 -0400
@@ -41,7 +41,7 @@ void TFWSyncThread::run(void)
     RunEpilog();
 }

-const uint ThreadedFileWriter::kMaxBufferSize   = 8 * 1024 * 1024;
+const uint ThreadedFileWriter::kMaxBufferSize   = 64 * 1024 * 1024;
 const uint ThreadedFileWriter::kMinWriteSize    = 64 * 1024;
 const uint ThreadedFileWriter::kMaxBlockSize    = 1 * 1024 * 1024;

@@ -382,9 +382,9 @@ void ThreadedFileWriter::Sync(void)
         // fdatasync tries to avoid updating metadata, but will in
         // practice always update metadata if any data is written
         // as the file will usually have grown.
-        fdatasync(fd);
+//        fdatasync(fd);
 #else
-        fsync(fd);
+//        fsync(fd);
 #endif
     }
 }

--
Rich