Vicidial HA using Corosync, DRBD and Pacemaker

[rbbumanlag]

Hi Vicidial,

Specification: ViciBox 8.0.1.(iso) | Asterisk 11.25.3-vici | VERSION: 2.14-682a BUILD: 180807-1653 | Cluster Setup | Intel(R) Xeon(R) CPU E5-2680 V3 @ 2.5GHz | 16GB DDR4 RAM | 1TB (SAS) 10kRPM |No Digium/Sangoma Hardware | No Extra Software After Installation | Virtualized on XenServer | SVN Version 3016

I am on a clustered setup with separate database, slave database, asterisk, web and archive server. my database server is in a HA setup and Web Servers is being load balanced using HAProxy. What I am encountering is that when I do a failover on the database, I am having problems on the clustered vicidial.

I am encountering problems in which when the agent login, the softphone is not being called to join the conference then time synchronization error will occur. Tried restarting NTP and making sure all Servers are in sync when it comes to the time.

I tried doing "core restart now" but still I am unable to make the softphone trigger. What I normally do to make it working is to restart the asterisk server and the web server after a database failover.

I would like to know what service I need to restart for asterisk and web.

Thanks!

[mflorell]

We have seen problems like this with clients using virtualized clusters. We never recommend using virtualized servers for VICIdial.

[williamconley]

screen -list on the dialers. These are not "services", but they are often in need of a restart. It's possible for most of them to be killed and they will auto-restart at the :00 when keepalive runs (or you can run keepalive immediately after killing them ...).

Virtual Vicidial, as pointed out by Matt, is a bad idea. Before you announce "buildout complete!" do a FULL load test with more agents than you intend to actually run on your first day (using fake calls to fake call recipients on servers outside the Vicidial system as your test set). Vicidial on Virtual tends to fall apart after just a couple agents. Even on powerful equipment it tends to die when loaded up. So if your job relies on this working, test the hell out of it FULLY loaded. If your HA system is hardware-based, you're good. But still test before launch.

[rbbumanlag]

Thanks. We will take now on your inputs.