v20 Hang No Service high cpu

[RadyAir]

We're a partner, since v15 on windows...
We've recently upgraded our inhouse system to v20 (2cpu, 2Gb) running on AWS Lightsail.
We have also been experimenting with Transcription via ChatGPT and Recordings. Only 2 extensions have this on, not a tone of traffic.
Total recordings at this moment is 768MB.

Twice this week, twice in 2 days, our 3CX has dropped completely. no service on any devices.
in AWS it shows running, but cpu is steady in the burst zone and peaking at 80%, instead of hovering around 2% like normal.

Cannot get an SSH Console via Putty or browser window, and the only option is to restart. today I had to use Force Stop.

Is there any possible link between the transcription element and high cpu, or perhaps a vulnerability that is being hit on?
I have the support info generated from just after bringing this up again today.
attached is a screenshot of cpu usage just before it stopped.

 

[bitn2]

2 CPU and 2GB Ram are a little weak for PBX with chat gpt. That could be the problem. How many extensions are on this pbx?

 

[RadyAir]

Still not sure that the ChatGPT itself runs on the 3CX server...
We had the transcription files and recordings transferring to a Google Drive folder "immediately" after the call,
which was generating high IO and probably impacting CPU.

Have changed this to ship recordings out once a week instead and this has mitigated the issue.
I still want to check CPU during a recording/transcription event, but have not had time yet.

This is our inhouse but production phone system, and we're testing Transcription before implementing for a couple clients.

 

[RadyAir]

An hour later and our 3CX instance froze again, 1st time in a week.
This occurred soon (5 minutes) after a long 105 minute call was completed and the recording was being created.
the Recording itself ultimately failed to be saved.

Again, the only way to regain functionality was a Force Stop with AWS - likely why the recording failed to complete.
question is - was the recording itself the reason for the crash?

We were using the WhisperAI for transcription, shipping the recordings off to Sharepoint storage first. Disabled for now.
so ChatGPT should be operating on Microsoft servers while creating the transcript.

but - we also had Summary turned on. perhaps the summary is created locally while the full transcription is generated outside 3CX?
still searching.