Posts in category "ceph"

Part 4 found four gaps the hard way, live, under pressure. This is the same migration run against tanaka the very next day with all four fixed proactively — plus a real hardware detour before any of the software even mattered, and one genuinely new finding Part 4 never hit.

Part 3 laid out the plan for migrating to smaller SSDs. Here's what happened running it against a real node with both mirror drives failing at once — a UEFI dead end (the third one), four gaps the plan never anticipated, a mistake I made and had to recover from live, and a real hardware failure that surfaced while fixing the last gap.

When both drives in your ZFS boot mirror fail simultaneously, you can't resilver your way out. Here's the full procedure for a Proxmox fresh install that preserves your Ceph OSDs and rejoins the cluster — based on a real harlan node recovery.

Nine of my fifteen Ceph OSDs use WAL-only acceleration while six use DB. I set out to measure the performance gap and discovered the real story isn't WAL vs DB — it's the USB 3.0 hardware ceiling that dominates everything. The matched-hardware comparison shows DB is 5-15% faster on reads, not the 32% that naive cross-node testing suggested.

Running Ceph on USB drives sounds crazy until you put the WAL and DB on an SSD. Here's how separating metadata onto a Crucial MX500 transformed my 15-OSD homelab cluster from sluggish to surprisingly capable — at a fraction of all-SSD costs.

A routine ZFS scrub alert on harlan turned into a multi-hour debugging session when a hostid mismatch fix collided with a pre-existing Ceph OSD failure from a dead USB drive. Here's how overlapping storage problems can mask each other and how to untangle them.

A power outage knocked my Ceph cluster from 15 healthy OSDs down to 4. The recovery took days of debugging — heartbeat cascades, a ceph.conf misconfiguration, and a dead SAN switch hiding behind NO-CARRIER flags on every node.

A comprehensive guide to resolving Jellyfin LXC disk space issues by migrating metadata to CephFS and optimizing container size for faster HA failovers.