The server name is no longer a server
How SMB.next turns the stock SMB client already installed on Windows, macOS and Linux into the front end of a distributed storage fabric.
Read the damage reportStorage systems / failure engineering / unwanted operating systems
Storage opinions,
mostly unwanted.
Notes from the awkward bit between “it booted” and “we can prove it is safe.”
STATUS OPERATIONALLY SUSPICIOUSAUTHORGRUMPY STORAGE GUYDEFAULT MODESHOW RECEIPTS
Latest field notes
Writing
![[censored]](https://grumpystorageguy.com/images/editorial-censor.png "Corporate identity confiscated by editorial policy.")
OS / 07 / IdentityMy recovery system authenticated a number Linux was free to change
A ZFS recovery import exposed why st_dev and mount IDs are observations of the current mount topology, not durable filesystem identity.
Read the damage report OS / 06 / Rite`online` was answering nine questions and getting several of them wrong
Why a Linux PID 1 service manager cannot compress intent, process ownership, readiness, health, recovery and runtime generation into one lifecycle enum.
Read the damage report OS / 05 / HealthBooted is not healthy
Why node-local boot success and a fleet health verdict must remain separate facts.
Read the damage report OS / 04 / RecoveryMy recovery command was structurally incapable of recovering
Six crash boundaries on real ZFS exposed a recovery loop that could never terminate.
Read the damage report OS / 03 / LoomThe most destructive thing my OS can do reported itself as the most innocuous
A correct solver and a correct reconciler combined to delete an operating system and print applied 0 files.
Read the damage reportThe Ceph cluster you never wanted
A ZFS pool needed to move six metres. The standard answer was Ceph. That was a much larger answer than the problem required.
Read the damage report OS / 02 / FoundryNo package in my OS is allowed to run a script
Four deliberately pathetic actuator verbs replace arbitrary maintainer code.
Read the damage report OS / 01 / SolverThere is no --force in my package manager
Why a dependency solver that can refuse a request also owes the operator a reason.
Read the damage reportMy SMB codec did 130 GB/s until I stopped benchmarking the optimiser
The number was magnificent, impossible, and almost published. Then the benchmark consumed what the parser actually produced.
Read the damage reportI got bored waiting for SMB.next, so I built it
The RFC sat in a folder until waiting became more irritating than implementing SMB. Then DataTicket happened because why not.
Read the damage reportOngoing series Building a bloody OS From constraint solver to recovery contract → SMB.next track Keep the clients.“Yes, I realise how this looks. Here is the proof that I have not completely lost my mind.”
Replace the assumptions. From boredom to DataTickets and benchmark crimes →