Async Agent Execution

This article describes how Joget runs agent jobs in the background, and how administrators can tune throughput and polling behaviour.

When an Agent Process Tool is configured with Run Mode = Async, Joget offloads the agent job to a background worker pool. The HTTP request or workflow thread that triggered the agent returns immediately; the job runs separately and stores its result when done.

Job lifecycle

Every async agent run passes through four states:

How it works

1. The Agent Process Tool writes the job to the database with status PENDING and returns to the caller immediately.
2. A background poller detects PENDING jobs at a configurable interval and claims them atomically; only one worker ever processes a given job.
3. The worker restores the original app and user context, runs the agent, then executes any output callbacks.
4. The job record is updated to DONE or FAILED.

JVM tuning properties

These properties control how the background executor behaves. Both are read once at startup a server restart is required after changing them.

Setting these properties

Pass them as JVM system properties at server startup:

-Dagent.async.pollInterval=10
-Dagent.async.workerThreads=8

Tomcat (setenv.sh / setenv.bat):

JAVA_OPTS="$JAVA_OPTS -Dagent.async.pollInterval=10 -Dagent.async.workerThreads=8"

Cluster deployments

The async executor is cluster-safe. Job claiming uses an atomic database update, so only one node in a multi-node Joget cluster will ever pick up and run a given job, even when all nodes share the same database. No additional coordination is required.