Serdar Gökay created CAMEL-26112:
------------------------------------
Summary: camel-milo - handleConnectionFailure disconnects the
failed client while it holds the manager lock
Key: CAMEL-26112
URL: https://issues.apache.org/jira/browse/CAMEL-26112
Project: Camel
Issue Type: Improvement
Components: camel-milo
Affects Versions: 4.23.0
Reporter: Serdar Gökay
Follow-up of CAMEL-25912, named there as not done.
{{SubscriptionManager.handleConnectionFailure()}} disposes the failed
connection under the single manager lock, and the dispose calls
{{OpcUaClient.disconnect()}}, which sends CloseSession and waits for the
answer. During a silent drop of the link that call blocks until the request
times out, and every caller of the lock waits with it: a consumer that starts
or stops, and the subscription recovery itself. Since CAMEL-25912 the producers
and {{MiloClientConnection.getOpcUaClient()}} no longer take the lock, so only
the subscription work after a consumer-side failure still waits.
Moving the disconnect out of the lock is not a small change:
{{whenConnected()}} (under {{registerItems}} and {{unregisterItems}}),
{{recreateSubscription()}} and {{recoverSubscription()}} call
{{handleConnectionFailure()}} while they already hold the lock, so they would
have to hand the failed connection out of their own lock scope, with a check
that the connection they disconnect is still the one that failed. An
asynchronous disconnect would instead occupy the single {{MiloClient}} or
{{MiloClientReconnect}} thread on a machine with one processor. This issue is
for choosing between these options and implementing one.
--
This message was sent by Atlassian Jira
(v10.3.26#10030026)
