Local Transcription server issues - Delete & restart

[Marco@]

Hello,

We found the local transcription server was disconnected, and even after numerous restarts of both the Transcription server and the 3CX instance, we had no luck to get them reconnected. Also attempted to use the Installation command from the existing transcription server record - no go.

Ended up having to delete the Transcription engine record and generate a new one - this worked.

I did notice that the installer shows version 20.6.0.39696

_____ ______ __
|___ / / ___\ \/ /
|_ \| | \ /
___) | |___ / \
|____/ \____/_/\_\

3CX OnBoard AI installation 20.6.0.39696
Debian Version: bookworm


however the status page in 3cx shows it as version 20.5.0.38991. Is this expected?




Prior server that was disco and could not be reconnected:

 

[Marco@]

Additional question; it shows MAX CONVERSIONS is now upped to 6, however even with the backlogged queue, it doesn't run past a max of 3 concurrent. Just inquiring what/how and if that will change?

 

[Kevin Attard Compagno]

There can be a mismatch between the version reported in the Admin Console, and the version of the installer itself. Having said that, in the near future (should be update 10) these will become aligned.

Concurrency 6 effectively means that 3 transcription jobs and 3 analysis (summary/sentiment) jobs can run concurrently.

 

[cmp1]

Kevin,

Would it be possible to have you guys add alerts if the transcription engine is no longer reachable? Similar to when a bridge is down, or service fails?

Additionally it would be fantastic if there is a way to add a backup, redundant/failover service. If primary transcription service is not reachable/fails, move on to secondary. (Maybe more than 1hr or x# or queue)

 

[Kevin Attard Compagno]

@cmp1 ,

Alerts is something we could consider.

There are some significant changes coming along to the AI Server, which, among other things, brings the possibility to connect multiple AI Servers to the same PBX.

Hang on in there...

 

[cmp1]

@cmp1 ,

Alerts is something we could consider.

There are some significant changes coming along to the AI Server, which, among other things, brings the possibility to connect multiple AI Servers to the same PBX.

Hang on in there...

Multiple AI servers to same PBX is nice for redundancy, multiple PBX to same AI server is needed for customers/providers with multiple offerings

 

[Kevin Attard Compagno]

@cmp1 ,

Hold on to your hat, that's coming too

 

[GeorgeNotDigital]

I host a local transcription machine from home on a dynamic IP and during some scheduled maintanance my IP changed. My yec2p-wmr.my3cx.net FQDN would not refresh. Following this guide helped me request a new FQDN with the correct IP.

Thank you, I'm glad I found this post!

Here's my steps:
# 1. Deleted and rebuilt transcription engine (new FQDN ****9-wmr.my3cx.net)
I had to reboot the PBX after deleting.

# 2. Patch le-posthook.sh — certbot runs hooks without /usr/sbin in PATH

cp -a /opt/3cxwm/setup/le-posthook.sh /root/le-posthook.sh.bak
sed -i '1a export PATH=/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin' \
/opt/3cxwm/setup/le-posthook.sh

# 3. Patch le-posthook.sh — `service --status-all` never matches a systemd-only
# unit, so the restart was skipped and the script still exited 0

sed -i "s|^if service --status-all.*then$|if systemctl cat 3CXWMConverterService >/dev/null 2>\&1; then|" \
/opt/3cxwm/setup/le-posthook.sh

# 4. Remove stale cert

certbot delete --cert-name yec2p-wmr.my3cx.net

# 5. crontab -e — removed yec2p line, added 2>&1 to cert line

0 1 * * 0 /opt/3cxwm/converterservice/aptupgrade.sh >>/var/lib/3cxpbx/WebMeeting/Logs/converterupdate.log 2>&1
@reboot /opt/3cxwm/converterservice/headers-update.sh >>/var/lib/3cxpbx/WebMeeting/Logs/converterupdate.log 2>&1
0 3 * * * /usr/bin/certbot certonly --no-autorenew --standalone --non-interactive -m "*****@gmail.com" --agree-tos --preferred-challenges http -d ****9-wmr.my3cx.net --post-hook /opt/3cxwm/setup/le-posthook.sh >> /var/lib/3cxpbx/WebMeeting/Logs/onboard-install.log 2>&1

Steps 2 and 3 are overwritten by the installer.

 

[Kevin Attard Compagno]

@GeorgeNotDigital

Once Update 10 goes Final, and you move to the latest 3CX AI Server, this issue will simply cease to exist.

We are switching to a client-model for the 3CX AI Server, and it will not need a dedicated FQDN or SSL certificates to connect to the PBX(s).

 

[adrian.simpson]

@GeorgeNotDigital

Once Update 10 goes Final, and you move to the latest 3CX AI Server, this issue will simply cease to exist.

We are switching to a client-model for the 3CX AI Server, and it will not need a dedicated FQDN or SSL certificates to connect to the PBX(s).

HI Kevin,

I have read through quite a few posts and just wanted to check the ability to connect multiple PBXs to a single AI Server isnt quite there yet?

We are on - 3CX OnBoard AI installation 20.6.0.39696 - Debian Version: trixie - and when I go to run the install for the second PBX I get an error. So just wanted to check that I am not doing anything wrong / missed the latest update on that feature?

Also am I understanding the above correctly that in future models it will just skip the FQDN/SSL part of the install?

Kind Regards
Adrian

 

[Kevin Attard Compagno]

Hi Adrian

PBX needs to be on V20 Update 10 (currently still in BETA); moreover, once the PBX moves from U9 to U10, you will need to delete the link between PBX and AI Server, uninstall the AI Server, re-create the link between PBX and AI Server and use the provided shell command, which you would then run on the AI Server machine to install the latest - it will be in the 20.7.0.xxxxx range.

...and yes, it will skip the FQDN/SSL part of the install.

 

[adrian.simpson]

Hi Adrian

PBX needs to be on V20 Update 10 (currently still in BETA); moreover, once the PBX moves from U9 to U10, you will need to delete the link between PBX and AI Server, uninstall the AI Server, re-create the link between PBX and AI Server and use the provided shell command, which you would then run on the AI Server machine to install the latest - it will be in the 20.7.0.xxxxx range.

...and yes, it will skip the FQDN/SSL part of the install.

Hi Kevin,

Thanks for the speedy reply, That is great news.

The testing on our fresh build is going great - the only limitation we are facing is we only have a 16GB RTX 2000 Ada (had one lying around) - we will be installing an additional matching card in the next couple of days in an attempt to pool the VRAM - this is all really for our own internal testing at the moment.

Looking forward to update 10

 

[Kevin Attard Compagno]

Hi Adrian

It *should* work. However, those cards are not something I've ever tested - or ever will, really - so their workability is an unknown quantity to me. A dual-card setup just compounds unknowns - having 2 cards does not "automagically" give you a larger VRAM pool as you might think it would.

Get a Mac Mini with minimum 24Gb - that's a much more appropriate "internal testing" platform for Update 10 and the AI Server.

You could get the best of both worlds, have the RTX 2000 and a Mac AI Server, and REALLY get to see multiple-machine setups running. 2 PBXs and 2 AI Servers, with each PBX connected to each of the AI Servers in full mesh mode, just because you can