Four node asterisk cluster

Question

I have a question regarding Asterisk servers. I have a client that have two locations connected with fiber optic. On both locations they have PRI ISDN-s. On one location there are 4 PRI ISDN, and on other there is 1 PRI ISDN. Client bought four HP servers, two per location and they also bought PRI ISDN Failover boxes. The question is, how to make stretched Asterisk cluster (two one one location, and two on other). Something like this:

Location one Location two +----------+ +----------+ |Asterisk 1|\ /|Asterisk 1| +----------+ \ / +----------+ PRI|Cluster >-------------Cluster IP-------------< PRI|Cluster +----------+ / \ +----------+ |Asterisk 2| / \|Asterisk 2| +----------+ +----------+

So basic, they want to have telephony no matter if any of sites fail the other "little" cluster should take over SIP connectivity. I can easily setup two node cluster on both locations (Corosync, Pacemaker or DRBD), but how to connect those two locations as one big cluster.

Answer 1

Some related information can be found in this ServerFault question. Making a cluster out of 2 clusters is unusual (aka a bad design), and making a 4-peer cluster with two peers per location is wasteful and useless (no value above one peer per location). In fact, with a 2-peer per location cluster you risk dual-active contention as the active peer at each location fights with the other. What a mess.

It sounds like your client is looking for a high degree of fault tolerance and has been ok investing in hardware; however, they are investing in the wrong areas. Taking the cheap approach approach with software: DRBD (which means corruption on one peer immediately corrupts the other peer), no sensing of environmental factors indicating a peer is failed, no deep hardware monitoring of the cluster, no upstream ITSP/SIP awareness, no intelligent synchronization, etc. means a very poorly functioning cluster. And what do you mean by ISDN PRI failover boxes? Manual/automatic PRI A/B switches such as the beroNet failover switch? Protocol converters?

I would suggest your client redesign their solution such that they have ISDN-to-SIP gateways (eg: beroNet gateways or digium gateways) at each location, and keep only a single Asterisk server at each site, then link them with HAAst. HAAst will take responsibility for rerouting traffic, moving IP's, synchronization files/databases on the clusters, etc., AND, it can update the IP's in the gateways at the same time. Even better, have standby/dark PRI links which are only billed when activated (many carriers/ITSP's will offer this service) at the secondary location.

If you use HAAst then it can also automatically modify the synchronized dialplan on failover so only essential services come up (I assume that is why your client has only a single PRI at the secondary location).

If you really want to keep the design above, I suspect HAAst can meet your original needs...but if your client is open to doing this the right way start over.