search

rabbitmq / rabbitmq-server

Open source RabbitMQ: core server and tier 1 (built-in) plugins
https://www.rabbitmq.com/
Other
12.31k stars 3.92k forks source link

Classic config peer discovery still might select two different seed nodes #12681

Open dcorbacho opened 2 weeks ago

dcorbacho commented 2 weeks ago

Describe the bug

Sometimes in CI the testcase peer_discovery_classic_config_SUITE:successfully_discovery fails to cluster all 7 nodes. Six nodes join the same seed, and one other node selects itself as a seed and does not join the rest of the cluster.

The peer discovery seed selection needs to be reviewed

Reproduction steps

  1. It's a CI flake

Expected behavior

The 7 nodes join the cluster

Additional context

No response

michaelklishin commented 2 weeks ago

This was discussed relatively recently. Given that for this backend, the list of nodes is explicitly listed ahead of time, should we make the seed explicitly configurable, and use it in the tests?

We have tried different approaches over the years, such as "use the first node on a sorted list".

For example, Cassandra uses an explicit seed node configuration argument.

@dumbbell @mkuratczyk

dumbbell commented 1 week ago

@dcorbacho: Do you have a link to a failing CI job or node logs?