TNS Listener got high & wandered off...

[DBAwhosaysNIE]

o.k., explain this one:

@05:00 EDT this morning I got a page saying the TNS Listener wasn't running on a pre-production (3rd party app being implemented)server. I attempted to restart it but got an error stack that ended (or began depending on how you look at it) with Linux Error: 32: broken pipe. mind you, the instance never reported any sort of error and worked fine using SQL*Plus from a local shell. I checked various config files, network stuff (the obvious) and even rebooted the box but continued getting same error even after reboot.

what fixed it?

I finally tweaked my .bash_profile (SuSE 7.1, BTW) and did a reinstall to a new ORACLE_HOME, mounted the SAME database (although, again instance never seemed to have a problem) & started the listener with the SAME listener.ora file (from old ORACLE_HOME) and it's working FINE!

it's kind of a post-mortum question but what in the $YOUR_FAVORITE_EXPLATIVE could have caused this? vendor doesn't have shell access to db server and system logs show no logins since last Friday anyway (which is what I'd expect).

 

[karluk]

I got a few hits on Metalink on Linux error 32. The most promising was note 1071646.6, which describes a problem with the file $ORACLE_HOME/network/names/.sdns.ora.

If that's the problem, you have to delete the line that reads

\ = (ADDRESS=(PROTOCOL=IPC)(KEY=ONAMES))

and restart your names server.

 

[DBAwhosaysNIE]

MYSTERY SOLVED! (>1 yr later)

vendor in their infinite wisdom doen't believe in using any sort of connection pooling mechanism (like concurrent managers in Oracle Applications). therefore anytime an end user logs into app (which is frequent since it's a CRM/POS-ish app for field sales folk) it spawns a session, forks a process (if you're not running MTS) and (here's the really cool part):

logs the connection in $ORACLE_HOME/network/listener.log

do this enough and listener.log eventually hits that majic 2GB barrier (though the error suggests anything but).

now that I know what I'm looking for it's easy to work around (crontab, FIFOs w/split, gzip, etc.) but has ANYONE out there EVER had a listener.log hit 2GB? I've run Oracle for some pretty big ERPs (Fortune 20 teleco) and currently a top 20 website but NEVER have I heard of such a thing (but then we've always been smart enough to connection pool.

anyway, one more thing for the thorough DBA to keep an eye on...