Claus Ibsen created CAMEL-25500:
-----------------------------------
Summary: camel-core - removeAllRoutes / stopAllRoutes stop routes
one by one, so a route still sending to a direct route is cut off instead of
drained
Key: CAMEL-25500
URL: https://issues.apache.org/jira/browse/CAMEL-25500
Project: Camel
Issue Type: Bug
Components: camel-core
Reporter: Claus Ibsen
Fix For: 4.23.0
AbstractCamelContext.removeAllRoutes() and stopAllRoutes() stop the routes one
at a time (stopRoute per route, in reverse startup order), each as its own
graceful shutdown. The ShutdownStrategy keeps a downstream route running until
the upstream routes have drained (DirectConsumer.deferShutdown returns true),
but only within one batch of routes. Stopped one by one, the downstream route
is stopped as soon as its own inflight count is 0, and an upstream route that
still has work to send to it (a split, a loop) blocks on the direct endpoint
until the shutdown timeout forces it, and the exchange is cut off.
This is the route reload in dev mode: RouteWatcherReloadStrategy calls
removeAllRoutes() on every reload.
Reproduce: seda:start -> split(body()).delay(300).to(direct:shipment);
direct:shipment -> mock:shipped. Send a list of 4, wait for the first to
arrive, call context.getRouteController().removeAllRoutes() with a 3 s shutdown
timeout.
* Today: takes 3011 ms (the full timeout), 1 of 4 delivered, 3 cut off with
RejectedExecutionException.
* Stopping the same routes as one batch
(getShutdownStrategy().shutdown(context, allRoutes, ...)), as a CamelContext
stop does: 1011 ms, 4 of 4 delivered.
Also, one by one each stuck route can use the full timeout, so N routes can
take N x timeout.
Fix: stop all routes as one batch through the ShutdownStrategy, as
doShutdownRouteGroup already does for a route group, then stop the route
services.
Found in the local-model benchmark (route-aggregator example), together with
CAMEL-25484.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
