Hello <!channel>, we're live now with @salmon-doctor-9726 discussing the newest Harvester release. Join us ⬇️ • youtube.com/live/KfJYOGBpmzM?si=… • linkedin.com/events/globalonline…/theater

Thanks for the link!

Joining nodes can start in a cordoned state is huge.

If there are questions for David, please feel free to ask them here. I can relay them to him

Mostly just tell him thanks for being awesome.

Will do 😂

Great to see work on upgrade resilience and debugging - it's the one major pain point we have!

For third party storage backups, does that include RBD snapshots > S3 ?

Only once I want to do a Harvester upgrade without issues. Please focus more on the lifecycle, not on new features!

For third party storage backups, does that include RBD snapshots > S3 ?

Just noticed this. Corrected the answer: no, RBD, which is specific to Ceph; it's a general backup based on Restic and Kopia first, a different backup engine. Will do more. But, S3 yes.

looking forward to 1.9.1 when it's ready!

How will the migration path from Longhorn v1 to v2 look like? I saw somewhere that there are no backing images anymore in v2, but everything relies on CDI. We expect a way to life migrate the volumes and backing images without VM downtime.

@bumpy-portugal-40754 fwiw, I upgraded 7 clusters last month. 6/7 had no issues.

@bland-article-62755 See what I mean? We do upgrades again since last week, 2 new issues (with v1.7.3). We had to wait until 1.7.3 because of the "unexpected" network device name changes. Sigh.

Anecdotal obviously, but I believe we've yet to have an clean upgrade run on any of our clusters.

Yeah, if you can find a root cause I'd encourage y'all to open issues. For us, upgrades have gotten more stable over time, but a good portion of that was finding out about the upgrade-helpers script before kicking off an upgrade. Also anecdotal, but at $PreviousJob upgrades for things (OpenShift, NetScaler, Citrix, etc) had a way worse failure rate, so my bar for acceptable might vary differently because of how terrible the other product upgrades would always go. Cause 6/7, while frustrating for the 1, still has me feeling pretty good after the dust has settled.

Things improved over time, indeed. The bad thing is that fixes only make it into 2 versions later because of the long release cycle.

I feel like a 6mo release cycle is great, but that's probably because I work in Higher Education. Different businesses have different needs and you can't make everyone happy. ¯\_(ツ)_/¯

Harvester release cycle is faster, it's connected to the k8s release cycle, so 3 releases per year. That's ok, but there are too few minor releases and no patch releases at all.