[asterisk-dev] 1.2.14: DUNDi ignoring incoming replies

[Gavin Hamill]

Apologies for the post to -dev - I believe I've discovered a
malfunction and was looking for feedback before I post a bug.

I have two machines, star (10.0.0.239) and hash (10.0.0.235), both with one E1 telco uplink. star is almost idle, whilst hash is quite busy (~25 Zap channels + MixMonitor + 30 SIP clients). I do DUNDi lookups between them to find an available Zap channel.

Lookups originating from star complete almost instantly ( < 10ms).
Lookups originating from hash may take several seconds or fail completely. tcpdump shows star is responding to the request immediately, but hash is not processing the incoming reply. star then retries once per second for up to 4 seconds before giving up.

I've annotated the tcpdump output to remove the need for a full dundi debug post!

ENCRYPT + DPDISCOVER request to star:
10:22:58.768384 IP 10.0.0.235.4520 > 10.0.0.239.4520: UDP, length 88

ACK to hash
10:22:58.768773 IP 10.0.0.239.4520 > 10.0.0.235.4520: UDP, length 8

ENCRYPT to hash:
10:22:58.769921 IP 10.0.0.239.4520 > 10.0.0.235.4520: UDP, length 90

ENCRYPT to hash:
10:22:59.769966 IP 10.0.0.239.4520 > 10.0.0.235.4520: UDP, length 90

ENCRYPT to hash:
10:23:00.770025 IP 10.0.0.239.4520 > 10.0.0.235.4520: UDP, length 90

(then finally after 3 seconds) ACK to star:
10:23:01.651028 IP 10.0.0.235.4520 > 10.0.0.239.4520: UDP, length 8
10:23:01.651446 IP 10.0.0.235.4520 > 10.0.0.239.4520: UDP, length 8
10:23:01.651518 IP 10.0.0.235.4520 > 10.0.0.239.4520: UDP, length 8
10:23:01.651593 IP 10.0.0.235.4520 > 10.0.0.239.4520: UDP, length 8
10:23:01.651716 IP 10.0.0.239.4520 > 10.0.0.235.4520: UDP, length 8
10:23:01.651783 IP 10.0.0.239.4520 > 10.0.0.235.4520: UDP, length 8
10:23:01.651808 IP 10.0.0.235.4520 > 10.0.0.239.4520: UDP, length 8
10:23:01.651841 IP 10.0.0.239.4520 > 10.0.0.235.4520: UDP, length 8
10:23:01.651868 IP 10.0.0.235.4520 > 10.0.0.239.4520: UDP, length 8
10:23:01.652041 IP 10.0.0.239.4520 > 10.0.0.235.4520: UDP, length 8
10:23:01.652074 IP 10.0.0.239.4520 > 10.0.0.235.4520: UDP, length 8

The rest are an assortment of ACKs, NULLs, INVALIDs and ENCREJ presumably as hash is processing the backlog (possibly out of order ?)

The problem is, /why/ is hash waiting so long? Are other asterisk threads too busy with other work? How can I determine this? I'm using 1.2.14 on both machines.

On hash, I notice other small indications like typing 'show chan' and pressing TAB... the commandline may wait for several seconds before completing 'show channel'.

Neither machine is doing anything exotic like MeetMe, queueing/agents or IVR. hash is new - an HP DL145 G2 (Opteron 246 with HyperTransport) and is dedicated to Asterisk. There is a single Sangoma A104 installed.

Cheers,
Gavin.