[asterisk-bugs] [JIRA] (ASTERISK-28962) Asterisk Memory Leak after SIP Reply Flood

[Jonas Swiatek (JIRA)]

Jonas Swiatek created ASTERISK-28962:
----------------------------------------

             Summary: Asterisk Memory Leak after SIP Reply Flood
                 Key: ASTERISK-28962
                 URL: https://issues.asterisk.org/jira/browse/ASTERISK-28962
             Project: Asterisk
          Issue Type: Bug
      Security Level: None
          Components: pjproject/pjsip
    Affects Versions: 16.6.1
         Environment: Amazon Linux 2 (CentOS based)
            Reporter: Jonas Swiatek
            Severity: Critical


I've got a user with a SIP Client (MicroSIP) that some times goes haywire, and responds to the initial INVITE from Asterisk by retransmitting it's 180 Ringing on a loop, seemingly unbounded. We're talking thousands of the same reply. I haven't been able to retrieve all of them from Homer as there are simply too many for their UI to cooperate. But the 100 or so I did manage to pull was all sent within a handful of milliseconds.

When Asterisk received these, it's memory usage shots up significantly, and the log file started printing hundreds of these lines:

WARNING[13337] channel.c: Exceptionally long queue length queuing to PJSIP/registrar-0000d96a

The channel noted is the outgoing call to the user with the defective SIP Client.

There was nothing else in the messages log file related to this at all.

Asterisk CPU spiked (thought not to 100%), which is not a surprise as it does need to process all these SIP Replies, however the memory usage did not go down afterwards, even 12 hours after.

Normally it'll sit around 5 mb or something, but after this it had allocated a good 6GB of ram and there was no sign it was about to give any of it up 12 hours later when I restarted it.

I've got a PCAP file, though it contains sensitive information so I'd prefer not to share it publicly

I've also moved this customer to a separate Asterisk instance running 16.11.1, and I can probably enable whatever other logging might be helpful on it, in case this client goes haywire again and takes out that instance.



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