Article URL: Diskless Linux boot using ZFS, iSCSI & PXE - Syncopated Pandemonium Comments URL: https://news.ycombinator.com/item.id=48045012 Points: 68 # Comments: 25.
1 Like
The part I would watch first is timeout behavior under load. iSCSI stalls can look like random filesystem bugs unless multipath and session timers are tuned aggressively.
1 Like
What burned us was reconnect saying “success” while the boot path was already poisoned. We started running forced link-flap drills in staging so initramfs behavior is predictable before this ever hits production.
1 Like
Retries sound nice until PID 1 is already confused. Fail fast, reboot fast, and emit a clear serial breadcrumb so you do not lose hours chasing ghosts.
Block recovery and process-state recovery are different problems. Treat root-LUN loss as state invalidation, then rehydrate cleanly instead of trying to limp forward.
Good distinction. We also moved logs off root early, because postmortems are useless when journald dies on the same mount that failed.
That “alive but not debuggable” state is the worst one. Netconsole breadcrumbs before reboot gave us enough signal to separate storage faults from init bugs.
