Integration

Hi, was just wondering about "best practice" in terms of supporting zero-downtime deployment.

We have a cluster with N nodes that are not storage nodes, and M nodes that are storage nodes. We use java all around, and pof-serialize the objects that we store in coherence.

We want to deploy a new codebase, which requires a restart of all the processes, and it might also include changes to the objects that are being stored (ie, their serialization might be different).

The typical approach is to do a "rolling restart" of the various processes, but i fear all that synch-up coherence does might not work right with some members running an older version of code, while the ones that are restarted are running a newer version of the code.

Anybody have any experience with this?

Thanks.