Marcus Ramberg

Note that this node (vernemq-0 actually warned it couldn't join the cluster): ``` vernemq-0 vernemq 2024-08-13T13:01:18.406757+00:00 [info] vmq_swc_peer_service:attempt_join/1:61: Cannot join a cluster, as local node '[email protected]' is non-empty. vernemq-0 vernemq...

https://gist.github.com/marcusramberg/8959dd3f1757e492a42797c5cd909f3e here's the startup log for the non-empty node. The sync values that repeat seem to go on until we scale down to 0 and back up again. We are...

I've been experimenting quite a bit, including trying to modify the startup script to always leave the cluster on pod replacement, to never leave the cluster, and to always flush...

https://gist.github.com/marcusramberg/db37a342e172ea25c5a34d3fec1e16f2 here's the full helm chart configuration values we use, with some internal details marked as . I can't really share the traffic generator I've been using as it relies...

Sorry, fixed the gist link now. I've now managed to reproduce by disabling auth/webhooks and running this cronjob with mqtt-stresser: https://gist.github.com/marcusramberg/52b0978a76e1a716659c5653b2401993

The stresser runs as a cronjob inside the Kubernetes cluster, so you can get the YAML from the second gist I posted and run kubectl apply -f to get it...

Hey, they are expected to time out after some minutes and get retried, given we don't really care about the actual results of the stress test but are just trying...

Here's some metrics for how it looks in production. Seems like after a restart, the cluster very quickly starts accumulating tombstones? 

> ... is that `2.0.1` as well? > > > > --- > > 👉 Thank you for supporting VerneMQ: https://github.com/sponsors/vernemq > > 👉 Using the binary VerneMQ packages commercially...

> Can you check what the creation date of your `/data` dirs or subdirs is? does it correspond to pod boot time? Yepp, I can confirm that they are created...