[asterisk-users] 1.8.32.3 - no timing indicated, tens of thousands of __sip_autodestruct error messages

[Stefan Viljoen]

Hi list

Just to let everybody know I think I've got to the bottom of the above
problem / error.

Turns out that the issues described in my previous post were caused by
problems in an MSSQL database that the Asterisk 1.8.32.3 instance was
writing to...!

Once I dropped the FreeTDS driver (while trying to diagnose what was going
on) the Asterisk instance immediately started performing normally and the
__sip_autodestruct error messages disappeared.

Channels linked to calls that were hung up closed normally once again and
everything was golden.

The problem on MSSQL turned out to be that some of the indices on the table
to which the Asterisk instance was writing required rebuilding as the table
had grown immensely since the Asterisk instance was created. The issue was
that an insert that would have normally taken about 500 milliseconds was now
taking up to 30 seconds to a minute. Therefore as a call is finished
Asterisk attempted to write the call into the CDR database but was extremely
delayed due to slow database performance. The __sip_autodestruct error
messages apparently is a watchdog thread / process that then detects that
channels that should be closed and gone have not been closed / disposed - it
then posts a warning and then later tries to close the channels involved,
which are still hanging around after the call they are linked to has
finished.

This had the effect that users could only make one call, then had to wait
until the channels disappeared (sometimes up to a minute later as the DB
managed to finish inserting that call's details in to the table with the
index problems) before they could make a second call, while the above
watchdog process complained loudly about channels that were hanging around
that should not be as Asterisk waited for MSSQL.

So the fix was simply to rebuild the indexes on the relevant tabe in MSSQL -
this definitively solved the problem and tens of thousands of calls have now
been made after the reindex in MSSQL and everything is fine.

This is interesting as it seems that Asterisk does NOT use FreeTDS
asynchronously, but synchronously as calls progress...

Anyway, hope this helps someone who runs into something similar - the
databases you touch from the CDR subsystem in Asterisk must be -FAST- and
able to be inserted into in milliseconds. Apparently anything taking a
second or more in a FreeTDS accessed DB linked to Asterisk can start causing
problems in Asterisk itself.

Regards

Stefan