Claus Ibsen created CAMEL-26313:
-----------------------------------
Summary: camel-core - ThrottlingExceptionRoutePolicy: raising
halfOpenAfter while the circuit is open leaves the route suspended for good
Key: CAMEL-26313
URL: https://issues.apache.org/jira/browse/CAMEL-26313
Project: Camel
Issue Type: Bug
Components: camel-core
Reporter: Claus Ibsen
Fix For: 4.24.0
While the circuit is open, the route's consumer is suspended, and only the
half-open timer calls {{calculateState(route)}} again. The timer is scheduled
with the value of {{halfOpenAfter}} at the time the circuit opened
({{addHalfOpenTimer}}, {{timer.schedule(new HalfOpenTask(route, timer),
halfOpenAfter)}}).
{{halfOpenAfter}} can be changed at runtime through {{setHalfOpenAfter}}, which
JMX exposes ({{ManagedThrottlingExceptionRoutePolicy.setHalfOpenAfter}}). If it
is raised while the circuit is open, the timer fires after the old value and
{{calculateState}} goes into the STATE_OPEN branch:
{code:java}
long elapsedTimeSinceOpened = TimeSource.currentTimeMillis() - openedAt;
if (halfOpenAfter <= elapsedTimeSinceOpened) {
...
} else {
LOG.debug("Keeping circuit open (time not elapsed)...");
}
{code}
The "time not elapsed" branch schedules no new timer. The consumer is
suspended, so no exchange arrives to call {{calculateState}} again, and the
route stays suspended until it is restarted.
h4. Example
# {{halfOpenAfter=10000}}, the circuit opens at t=0, and the timer is due at
t=10s.
# At t=5s, an operator sets {{HalfOpenAfter}} to 60000 over JMX.
# At t=10s, the timer fires: elapsed 10000 < 60000, nothing is scheduled, and
the circuit stays open for good.
h4. Proposed change
In the "time not elapsed" branch, schedule the timer again for the time that is
left ({{halfOpenAfter - elapsedTimeSinceOpened}}), or call
{{addHalfOpenTimer(route)}} as the keepOpen branch does. A test can open the
circuit, raise {{halfOpenAfter}}, and assert that the route half-opens after
the new delay.
Related: CAMEL-26104 (half-open timer through the ExecutorServiceManager) and
CAMEL-26291 (failure window counting) touch the same class, but neither fixes
this.
--
This message was sent by Atlassian Jira
(v10.3.26#10030026)
