How do I tune the Event Manager of BAW (thread pool, scheduler, blackout periods) for timers and UCAs?

Question

Our timers fire late during peak hours and UCA-driven starts pile up. Which Event Manager settings exist, where are they configured, and how do we find the right values?

Answer 1

The Event Manager (EM) is the scheduler that executes timers, UCA invocations, asynchronous system tasks and BPD navigation work. It runs in every cluster member; its capacity is a thread pool per member and its behaviour is configured in 100Custom.xml (Process Server) and partially in Process Admin.

<properties>
  <server merge="mergeChildren">
    <event-manager merge="mergeChildren">
      <thread-pool merge="mergeChildren">
        <max-threads merge="replace">30</max-threads>          <!-- default 10; raise when tasks queue up but CPU / DB are idle -->
        <min-threads merge="replace">10</min-threads>
      </thread-pool>
      <loader-long-period merge="replace">15000</loader-long-period>   <!-- ms between scheduler polls when idle (default 15 s) -->
      <loader-short-period merge="replace">2000</loader-short-period>  <!-- ms between polls when work exists -->
      <loader-advance-window merge="replace">60000</loader-advance-window>   <!-- ms: how far ahead tasks are loaded -->
      <sync-queue-capacity merge="replace">200</sync-queue-capacity>
      <async-queue-capacity merge="replace">200</async-queue-capacity>
      <task-blackout merge="mergeChildren">                  <!-- no EM execution during backups -->
        <daily merge="replace">02:00-02:30</daily>
      </task-blackout>
    </event-manager>
  </server>
</properties>

How to find values: Process Admin > Event Manager > Monitor shows queued vs executing tasks per scheduler (or GET /ops/std/bpm/event_manager/monitor) - a growing "scheduled in the past" backlog with idle threads means the loader period / advance window is too slow; all threads busy means the pool is too small or the tasks are slow (look at the tasks: a slow UCA service or a system task calling a slow API - fix that first, more threads only move the bottleneck to the database). Every EM thread needs a JDBC connection - size the data source pool accordingly (question on JVM settings). Blackout periods stop timers during backups so that no half-written work exists. Timer accuracy is bounded by the loader period (a 15 s poll means timers fire up to 15 s late); keep it ≥ 2 s to avoid database load. On CP4BA the same XML goes into the custom 100Custom.xml secret and applies to every replica; the number of replicas multiplies the EM capacity.

References

• BAW documentation - Event Manager configuration and tuning (100Custom.xml)
• Operations REST API - event manager monitor