pjfanning opened a new pull request, #3652:
URL: https://github.com/apache/pekko/pull/3652
### Motivation
Two actor systems with the same uid break Artery associations: the uid
collision surfaced as an unexpected
`IllegalArgumentException` during the handshake, a `HandshakeRsp` was still
sent, and the association was left in a
confusing state. The uid itself was a plain random Long chosen when the
ActorSystem is created, so collisions were
not as unlikely as they could be.
### Modification
- Artery: `AssociationRegistry.setUID` throws a dedicated internal
`UidCollisionException`. `ArteryTransport.completeHandshake`
catches it, logs a warning, quarantines the colliding peer uid and fails
the handshake. `InboundHandshake` only sends
`HandshakeRsp` when completing the handshake succeeded.
- System uid: new internal `systemUid` on `ActorRefProvider` and
`RemoteTransport`. `ExtendedActorSystem.uid` delegates
to the provider (and throws `IllegalStateException` if read before the
provider is initialized).
`LocalActorRefProvider` and classic `Remoting` keep a random uid;
`ArteryTransport` generates it at start from the
bind host, port and several timestamps.
- Internal `pekko.test-only-uid` setting (not in reference.conf) to force a
uid in tests.
- `Slf4jLogger` resolves the uid lazily and omits the `pekkoUid` MDC value
until it is available.
- Tests: uid collision tests in `RemoteConnectionSpec` (connections TO and
FROM two systems sharing a uid), new `SystemUidSpec`.
### Result
Uid collisions no longer break existing associations; the colliding peer is
rejected cleanly. Artery uids are much
less likely to collide.
### Tests
- Not run locally - left to CI
### References
Ported from (Apache-2.0 since 2026-10-23):
- https://github.com/akka/akka-core/pull/32161
- https://github.com/akka/akka-core/pull/32168
### Notes for reviewers
- Wire format unchanged: the uid is still a Long, so 2.0.0 and newer nodes
interoperate in a rolling upgrade.
- MiMa excludes (`ReversedMissingMethodProblem`) in `actor/` and `remote/`
`2.0.x.backwards.excludes/system-uid.excludes`
for the new abstract `systemUid` on the @DoNotInherit `ActorRefProvider`
and the internal `RemoteTransport`.
- Init ordering: `system.uid` now throws if read before the remote transport
has started. Callers checked:
`AddressUidExtension`, cluster, `RemoteWatcher`, the receptionist (all
after `transport.start()`); `Slf4jLogger`
handles the early case.
- The collision is logged with `log.warning(exc.getMessage)`; the ported
tests use
`EventFilter[UidCollisionException]().intercept`, so check CI for that
expectation.
- Classic remoting behaviour is unchanged (random uid).
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
