Asterisk crashes in vmware vsphere 7 environment - asterisk 20.5.2

[shmuel4]

I run asterisk in a vmware vsphere 7 environment, I have 6 VMs with ubuntu 22.04, I run asterisk in docker and every so often, I don’t know exactly when, maybe when there is a high volume of calls, asterisk crashes.

in Asterisk’s logs I don’t see anything, but in dmesg I see the following lines:

[Wed Jan 24 11:50:50 2024] asterisk[98615]: segfault at b8 ip 00007f2b06494cda sp 00007f2ab56a5840 error 4 in libasteriskpj.so.2[7f2b0638d000+10b000]
[Wed Jan 24 11:50:50 2024] Code: 49 8b 3c 24 48 0f 48 da 48 85 db 74 d3 48 8b 36 48 89 da e8 b8 1a f0 ff 49 8b 3c 24 eb c2 66 90 f3 0f 1e fa 55 53 48 83 ec 08 <48> 8b 5f 08 48 8b 6e 08 48 85 db 7e 41 b8 01 00 00 00 48 85 ed 7e

I have another physical server (not a VM) in the same environment that does the same actions, and there it doesn’t happen.

I don’t know what could cause this, I tried to compile asterisk with:

menuselect/menuselect --disable BUILD_NATIVE menuselect.makeopts

But it doesn’t help.

Does anyone perhaps know of special instructions for building an asterisk in such a VM environment?
What could be causing the problem?

[jcolp]

Aside from disabling BUILD_NATIVE as mentioned there’s nothing else special. The only thing that can be said is a crash occurred within the PJSIP stack. A backtrace[1] would be required to know anything further.

[1] Getting a Backtrace - Asterisk Documentation

[shmuel4]

Thank you,

I could not take Backtrace because the application runs in Docker.

Anyway, I tried to do the following:

Instead of assigning 15 processor cores to the VM - which are 15 sockets, I did all the cores in one socket.

That is, instead of 15 separate processor units that each have one core, I put 15 cores in one socket.

This seems to have solved the problem, I’m still following up on the issue.

[TJNII]

If you’re running in Docker and getting segfaults this thread shows how I enabled the required reporting makeopts in a Docker build: Non-interactively enabling menuselect debug options

Might be useful to you. The segfault I experienced was in testing a new container so I was able to shell in and process the coredump by hand. I have persistent storage in my Asterisk container for things like the voicemail spool, I assume the core could be configured to dump there for post-processing but I haven’t set that up yet.