NITB hangs if many subscribes and/or subsr. exists with empty extension in hlr
keith at rhizomatica.org
Mon Jul 25 12:19:39 UTC 2016
On one of our sites osmo-nitb was hanging a few seconds after startup.
It took me a while to track it down, and I'm sorry I do not have log
output as it has gone to scroll buffer heaven, and I do not want to
break the system again right now to get another output, but I can say
more or less what happens.
If I see it again, I'll be sure to capture full debug log. I also have a
backup of the state of the offending hlr, so I can bring it up locally
and see if I can replicate and send the log.
In fact, there was nothing out of the ordinary in the log anyway before
nitb simply stops responding. (CTRL-C still shuts down cleanly if would
This 'bad' hlr has some 30,815 subscriber entries, one of them has empty
string for extension.
(We normally never get to anything near 30,000 as we purge inactive
subscribers regularly, but the blank extension had broken our purge job)
On this blank extension issue, i'm not sure it coincides with the hang,
but I noted this error followed by DBI traceback from here:
Error was non unique value for column extension, which makes sense of
course, as we don't give it a value.
Also seems we don't recover from this, as |db_subscriber_alloc_exten()
will never be called?
I deleted some 28,000 subscriber entries that were not commissioned
users, and the nitb then functions, so I'm at a loss to know where it
might have been actually stuck, or if it might be related to the blank
extension, or to the large amount of subscriber entries, or something else.
Maybe it's obvious to somebody. I have of course deleted the blank
extension entry now from the hlr.
-------------- next part --------------
An HTML attachment was scrubbed...
More information about the OpenBSC