My installation includes the following:
webMethods version 9.10
Two IS servers not clustered, no Terracotta.
Load balancer in front of the two IS servers.
Three UM realms clustered in a quorum.
I reported an incident with SAG support related to a problem with IS to UM connection alias when share client prefix is enabled on both IS servers. Yes, I ensured the value is the same.
As a temporary work-around until a solution is provided, I am thinking about configuring IS #1 to UM #1 and IS #2 to UM #2. In other words, I would dismantle the UM quorum. What are the negative consequences of such an installation?
The main thing is the fact that you introduce single points of failure: if you lose an IS, messages that have not yet been picked up from that respective UM will wait there until the IS comes back up. Similarly, if you lose a UM, messages that had not yet been picked up will not be picked up until UM comes back (hopefully it does) and that respective IS will be unable to publish anything new.
I learned of another negative consequence with the proposed solution that omits clustering of UM. We have a trigger with processing mode = Serial. Consequently, incoming request going to two IS servers will most likely process two requests concurrently. Not desirable for serial processing.
True, but if the requests are coming in via the load balancer anyway, you already have the potential for requests to get processed out of order. I agree that this exacerbates it though.
George,
What you are describing (serial triggers are only serial within each IS) only applies to 9.8 and lower.
In IS+UM 9.9 and higher, serial triggers do now ensure serial processing across an IS cluster.