Java: Conductor creates a lag when recovering after a server crash #219
Description
As discussed in this discord thread, workflows get stuck if conductor is enabled.
Restarting the server while parent-serial is running leads to the workflow execution being stuck for more than 5 minutes, possibly forever.
Restarting parent-parallel leads to around a minute delay before the workflow is replayed. Disabling conductor removes those delays and both workflows replay immediately.
@Workflow(name = "parent-serial")
public void parentSerial() throws InterruptedException {
System.out.println("parent-serial start");
for (int i = 0; i < 10; i++) {
final int i2 = i;
DBOS.runStep(() -> taskWorkflow(i2), "step " + i);
}
System.out.println("parent-serial completed");
}
@Workflow(name = "parent-parallel")
public void parentParallel() throws InterruptedException {
System.out.println("parent-parallel start");
List<WorkflowHandle<Void, InterruptedException>> handles = new ArrayList<>();
for (int i = 0; i < 10; i++) {
final int index = i;
WorkflowHandle<Void, InterruptedException> handle = DBOS.startWorkflow(
() -> this.proxy.taskWorkflow(index),
new StartWorkflowOptions().withQueue(this.queue));
handles.add(handle);
}
for (WorkflowHandle<Void, InterruptedException> handle : handles) {
try {
handle.getResult();
} catch (Exception e) {
System.out.println("Task failed " + e);
}
}
System.out.println("parent-parallel completed");
}