An "in-flight" upgrade means the product version changes while instances keep running and later continue on the upgraded engine - the standard in-place upgrade of IBM BPM / BAW (question 3136 for the BPM-to-BAW variant). The steps, with the checkpoints that matter:
- Plan: read the upgrade path and the deprecations of the target; check WebSphere fix pack, Java, database and OS levels; list custom code (jars, 100Custom.xml, portal customisations); schedule a window with the business (tasks stay open, the portal is down).
- Freeze and drain: stop new deployments; pause the Event Manager (Process Admin) shortly before the window so that no half-done system work exists; let running transactions finish; stop the cluster members, node agents, deployment manager.
- Back up: databases (offline or point-in-time consistent), profiles (manageprofiles -backupProfile), install root, Installation Manager data. This is the rollback.
- Upgrade the binaries: Installation Manager (imcl install the target BAW version + required WebSphere fix packs + interim fixes) on every node; deployment manager first.
- Upgrade the configuration and schemas: run the version-specific commands - BPMUpdateSystem (updates cell configuration and applications) and the database schema upgrade command of your source / target combination - the documented sequence; they migrate the instance tables in place, which is what makes the upgrade "in-flight" safe.
- Start and verify: start the deployment manager, nodes, members; check SystemOut.log for schema warnings; Process Admin > Installed Apps: the snapshots are still installed; open a few running instances and tasks (Process Inspector), complete one task per app; resume the Event Manager; check timers and UCAs fire.
- Clients and toolkits: install the matching Process Designer; upgrade System toolkits in the apps at the next snapshot (not required for running instances).
- Post-upgrade: re-apply customisations (portal theme, 100Custom.xml review), rebuild the plug-in config for IHS (GenPluginCfg), performance smoke test, then release the change freeze.
# typical command sequence (traditional BAW, all nodes stopped, dmgr up for the config step)
imcl install com.ibm.bpm.WFS.v24 -repositories /repo/baw -installationDirectory /opt/ibm/BAW -acceptLicense
/opt/ibm/BAW/bin/BPMUpdateSystem.sh -dmgrProfile DmgrProfile -user admin -password ... # cell config, apps, database schemas (runs upgrade steps)
startNode.sh ; startServer.sh SingleClusterMember1
Rollback = restore the backups (binaries + profiles + databases together). The instance-level risk is not the platform upgrade but app snapshot changes deployed afterwards - keep those separate from the product upgrade window.
References