[asterisk-bugs] [JIRA] (ASTERISK-27285) AstDB Locks taking a while to unlock

[Zach R (JIRA)]

Zach R created ASTERISK-27285:
---------------------------------

             Summary: AstDB Locks taking a while to unlock
                 Key: ASTERISK-27285
                 URL: https://issues.asterisk.org/jira/browse/ASTERISK-27285
             Project: Asterisk
          Issue Type: Bug
      Security Level: None
    Affects Versions: 13.10.0
         Environment: Centos 6.9 Final 64 bit, using chan_pjsip with realtime for AORs, Endpoints, Auths, Global. Contacts are in the ASTDB
            Reporter: Zach R
            Severity: Minor


I've noticed when I use the AstDB for chan_pjsip's contacts for a large amount of registrations per second the sorcery/contacts task processor will eventually overflow and lock up. I first traced what that task processor does in the code and saw it handles callbacks for deleting and creating the contacts.

I then tested to see if it was teh AstDB write time that was causing it to back up, as I read that asterisk and/or sqlite3 lock the file/table to avoid corruption or any race conditions for the sqlite3 database. While using realtime on res_odbc I did not have any issues even when sending 40 registrations/calls per second using sipp. This doesn't count the response to the 401 with the nonce.

After that I added some code using the clock function from C to time how long it was taking in main/db.c for db_puts. I put one set of timestamps/clock objects being set surrounding (on the outside) of the lock done and one on the inside.

I noticed that it seemed it waited quite a while for the lock to be released sometimes, while the inside of the lock took at most 0.01 seconds.

I've attached the scenario file I used for sipp, the logs I printed out in console with the timing, as well as the code I changed to do the time logs.

Some of these took as long as 0.5 seconds around the lock but 0.0 inside it.



--
This message was sent by Atlassian JIRA
(v6.2#6252)