shashank created CAMEL-25846:
--------------------------------
Summary: camel-tahu - after its route is restarted, an Edge Node
producer never connects again and drops every message (logged, the exchange
completes)
Key: CAMEL-25846
URL: https://issues.apache.org/jira/browse/CAMEL-25846
Project: Camel
Issue Type: Bug
Components: camel-tahu
Reporter: shashank
Assignee: shashank
{{TahuEdgeProducer}} (main 20cdb309f7d0) keeps the {{TahuEdgeClient}} of each
Edge Node in a static map ({{descriptorClients}}, {{computeIfAbsent}} in
{{createClient}}). When the Edge Node producer stops, it shuts its client down
({{TahuEdgeNodeProducer.doStop}}: {{tahuEdgeClient.shutdown()}}), but the
client stays in the map.
A Tahu {{EdgeClient}} cannot run again after {{shutdown()}}: {{shutdown()}}
sets {{stayRunning}} to false, only the constructor sets it to true, and
{{run()}} returns at once when it is false (Tahu 1.0.21, read with javap). When
the route is started again, Camel creates a new producer, whose constructor
takes the shut down client from the map, and {{doStart}} submits it again
({{TahuEdgeClient.startup()}}). The client never connects to the MQTT server
again: the Edge Node stays offline (its NDEATH was the last message), and every
message sent to the endpoint is dropped. {{EdgeClient.publishSparkplugMessage}}
catches the failure and logs {{Failed to publish message on topic=...}} at
ERROR, so the exchange completes without an exception. Device producers of the
same Edge Node, also in routes that were not restarted, keep the old client
too. This lasts until the JVM restarts (the map is static, so a new
CamelContext in the same JVM takes the shut down client as well). A route
restart is enough: stopping and starting the route (JMX, control bus, a route
policy, a supervising route controller).
*Fix*: the Edge Node producer removes its client from the map when it stops
(and shuts down the client's executor), and creates a new client when it
starts. A new client gets the metric maps of the Edge Node and of its Devices
again (kept in a static map by descriptor), and Device producers take the
current client of their Edge Node when they publish. A Device producer never
creates a client itself in that case.
*Tests*: new {{TahuEdgeProducerRestartTest}} (2 tests) with {{FakeMqttBroker}},
a small in-process MQTT 3.1.1 broker that acknowledges connections,
subscriptions and publishes and records the published topics.
{{edgeNodePublishesAgainAfterTheRouteIsRestarted}}: NBIRTH and one NDATA, stop
and start the route, then a second NBIRTH and NDATA are expected. On main it
fails: {{not published within 20 seconds; connects: 1, published:
[spBv1.0/RestartGroup/NBIRTH/RestartNode,
spBv1.0/RestartGroup/NDATA/RestartNode,
spBv1.0/RestartGroup/DDEATH/RestartNode/D1,
spBv1.0/RestartGroup/NDEATH/RestartNode]}}.
{{deviceOfAnotherRoutePublishesAgainAfterTheEdgeNodeRouteIsRestarted}}: only
the Edge Node route is restarted, and the Device route's DDATA is expected
after the new DBIRTH; fails on main the same way ({{connects: 1}}). Both pass
with the fix. camel-tahu unit tests with the fix: 9 run, 0 failures (the IT
tests need a container and were not run).
*How it was found*: variant analysis of CAMEL-25838 (pattern: a client or
connection that is shut down or replaced, while state that belongs to it, here
the client in a static map and the Device metric maps, is kept or lost), then
the JUnit tests.
*Affected*: main, camel-4.23.x, camel-4.18.x and camel-4.14.x (the same static
map and shutdown; read, not run; the {{stayRunning}} behaviour was read in Tahu
1.0.21 only).
_Filed with Claude Code on behalf of allthingssecurity._
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
