• plateee@piefed.social
    link
    fedilink
    English
    arrow-up
    3
    ·
    3 days ago

    A Proxmox cluster with Ceph pretty much won’t work without 10Gbps.

    It’ll slowly fail and all your data will get lost. Ask me how I know.

    (Also, you need at least 4GB memory dedicated to Ceph on each node… Another compounding lesson learned)

    • kamen@lemmy.world
      link
      fedilink
      English
      arrow-up
      4
      ·
      2 days ago

      If software depends on specific infrastructure to work, then it should have it as a hard requirement. Haven’t tried Ceph myself, but this sounds like there’s something very wrong with it conceptually or the way it’s set up.

      • plateee@piefed.social
        link
        fedilink
        English
        arrow-up
        3
        ·
        2 days ago

        It’s more that Proxmox has a weird niche laying between both enterprise/small-medium business and also homelab infrastructure.

        Ceph is designed towards the former where 10Gbe fibre and hosts with a boatload of RAM is fairly commonplace.

        My problems were a result of not paying close enough attention to those requirements and just hoping it would work on small micro PCs with 12GB of ram.

        • kamen@lemmy.world
          link
          fedilink
          arrow-up
          3
          ·
          2 days ago

          Yeah, but even if that’s the case, I would expect it 1) to refuse to run at all if something’s not fit or 2) to just be bottlenecked by the network instead of making data disappear altogether.

          • plateee@piefed.social
            link
            fedilink
            English
            arrow-up
            2
            ·
            3 hours ago

            Oh, to be fair the data disappeared because I was stupid for a different reason and proxmox upgrades were done out of order.

            I’m basically 100% to blame for the data loss, but it was impacting performance from an IO and memory perspective before I fucked it all up.