Skip to content

changelog

every publish of this site, newest first. times are pacific.

publishes

when publish kind what changed
2026-10-06 17:43 #45 docs + features
  • added doc: home assistant: what is installed, the yaml, integrations, add-ons, dashboards and custom cards, and the esphome devices
  • changed feature: infinitude keeps the thermostat's state on cephfs, so it survives the task moving node
  • changed doc: proxmox backup server has all of PBS's own setup, from creating the datastore to its certificate and signing in with entra, and links to building pbs1 at the top
  • changed doc: containers keeps truenas containers in general apart from pbs1, which covers the container and building it up to an installed PBS
2026-10-05 16:26 #44 docs + features
  • added doc: fan tuning: where truenas1's fans started and ended, the BMC's sensors and curves, what the box does under load, and how claude tested it
  • changed feature: truenas1's fans each follow the sensor that matches what they cool, rather than one index for the whole box, which is much quieter at idle
  • added feature: gatus also watches the fan on truenas1's asus m.2 card, which cools the fast pool, and fails below 1,200 RPM
  • added feature: gatus watches the 60 mm fan over truenas1's pcie slots through the BMC's redfish api, and fails below 1,500 RPM. the fan keeps the m.2 cards there alive
  • changed feature: truenas1's glances names each sensor by the slot its chip sits in, which survives the kernel renumbering drives between boots. the boot drives' second sensor, which runs near 80°C, has its own thresholds, so it no longer shows red
2026-10-04 19:02 #43 docs + fixes
  • fix: pve1 is back on the thunderbolt mesh. its udev rule had been off since the proxmox 9 upgrade, so en05 and en06 never came up and its ceph traffic went the long way, through the router
  • changed doc: openfabric mesh: no if-up.d scripts for frr. frr picks up the links by itself, and restarting it from one hangs boot on proxmox 9
  • changed doc: thunderbolt networking: the udev rule is needed, because networking runs before en05 and en06 exist
2026-10-04 18:12 #42 docs + features + fixes
  • changed feature: each process in glances shows its full command line with secret-looking values masked, tagged with the container it runs in
  • added feature: a loopback-only docker proxy on syn02 and pi-zwave01, so glances can name containers
  • fix: chrony on pi-zwave01 starts at boot again. the gps time override no longer orders it after gpsd, which made a loop that systemd broke by dropping chrony
2026-10-04 13:54 #41 docs + features
  • changed feature: glances lists each host's processes, read from the host's /proc, as nobody with no capabilities
  • changed feature: truenas1's app configs are snapshotted hourly for three days and daily for two weeks, to stay under TrueNAS's 10,000-snapshot warning
2026-10-03 14:22 #40 docs + features + fixes
  • added feature: a third UPS, in the smc closet, for the closet's network gear and syn02. a NUT server in a container on truenas1 reads it, for PeaNUT and prometheus
  • added feature: gatus checks each UPS card, each NUT server, and whether each server's data is fresh, judged by what prometheus has from the NUT exporter
  • changed feature: pve2 and pve3's NUT servers listen on the lan too, for the exporter
  • added feature: the UPS cards send their logs to victorialogs
  • fix: the docker hosts answer arp for their own address on one interface only, so the gateway no longer sees it flip between eth0 and the macvlan shim and raise an ip conflict
  • removed feature: nginx proxy manager is retired: its stack, its name on traefik, its gatus checks and its homepage tiles are gone
  • changed doc: nginx proxy manager moved to no longer used, without its stack file
2026-10-02 18:23 #39 docs + features + fixes
  • added doc: a UPS section: truenas1, the proxmox cluster and PeaNUT
  • added feature: NUT on truenas1 and on all three proxmox nodes, reading both UPSes over their network cards, so each shuts down what it feeds before the battery runs out
  • added feature: the proxmox cluster shuts down together in a power cut: Ceph flags set, guests stopped, a hold so no node leaves Ceph early, and the flags cleared on the next boot
  • changed feature: the proxmox HA shutdown policy is freeze
  • added feature: PeaNUT on truenas1, a dashboard for both UPSes
  • added doc: unpoller: UniFi metrics for prometheus, with the queries i use
  • added feature: UnPoller on truenas1, scraped by prometheus, with a gatus check on what prometheus has from it
  • added doc: victorialogs: the log store on truenas1, with UniFi's events arriving by syslog and the queries i use
  • added feature: VictoriaLogs on truenas1, UniFi's SIEM target, with a gatus check on its health
  • added feature: capped truenas1's GPU at 350 W, set again at every boot, to stay inside the UPS's power budget, see hardware
  • changed feature: raised wordpress's apache header limit to 32 KB, so a browser signed in to oauth2-proxy isn't refused with a 400
  • fix: verify-restore restores to the container's disk, not its tmpfs /tmp, where a large directory got the restore killed for memory, and counts the backup's entries, which it counted as 0
  • changed doc: wordpress: how to put the whole site back from one backup's files and dump, and why the dump has to be copied out first
  • added feature: pbs-restore, a file-level restore portal for the VM and container backups, lan only, with proxmox's own sign-in, in the access table
  • added feature: traefik can refuse requests another site started, per name, except to listed paths (cross_site_paths)
2026-09-28 13:32 #38 docs
  • changed doc: wordpress: open the network admin in a browser window that maps the root site's name to traefik, with the commands for chrome and edge, instead of a hosts file
2026-09-28 11:50 #37 docs + features
  • removed feature: the last names traefik passed through to nginx proxy manager, which served nothing once it stopped
2026-09-28 11:14 #36 docs + features
  • changed feature: made wordpress's root site lan-only: outside, every path redirects to www.
  • changed feature: set up wordpress's site on another domain on traefik again, with a certificate of its own. cloudflare checks that certificate (full, strict) and uses post-quantum key agreement to traefik
  • changed feature: set up gatus to check the certificate of a name outside my domain, by name but still on the lan
2026-09-28 10:12 #35 docs + features
  • changed feature: moved the config from git sidecars to ssh over port 443, after port 22 to github stopped answering
  • added feature: replaced nginx proxy manager with traefik
  • added doc: traefik
  • added feature: set up oauth with entra ID on traefik, through oauth2-proxy, for the names with no login of their own
  • changed doc: rewrote oauth2-proxy for how it runs now, with its app registration in entra
  • added doc: auth: how people sign in to every app, on the lan and from outside, and the MFA on each way in
  • added doc: entra ID: every app registration, and the password logins for when entra is down
  • changed doc: traefik: the diagram is two, one for requests and one for config
  • changed doc: placeholders the site hid are shown on entra ID auth, let's encrypt via cloudflare and installing debian as VM. on the entra page, the rest of the list showed as a link
  • changed doc: nginx proxy manager is stopped, replaced by traefik
  • removed doc: the old traefik template
  • changed feature: set up wordpress on traefik. the root site's front page redirects to www.
  • added doc: signing in to portainer with entra ID
  • added doc: webauthn keys on every node of the proxmox cluster
  • changed feature: set up gatus to check every traefik route, and traefik itself over a private network
  • changed feature: set adguard's minimum cache TTL to 0
  • changed feature: encrypted the PBS copy in azure with rclone crypt, see backups
  • added feature: set up a nightly backup of truenas1's fast/configs to PBS, and from there to azure, see tasks and scripts
  • changed feature: set the PBS datastore's record size to 1M, see proxmox backup server
2026-09-26 10:15 #34 docs + features
  • added feature: added a bbolt check of portainer's database to every hourly cephfs backup
  • changed doc: databases: portainer is restored from the checked copy in the cephfs backup
  • added feature: replicated truenas1's fast/configs to the rust pool after every hourly snapshot, see storage and snapshots
  • added feature: set up a nightly copy of the catalogue apps' settings into fast/configs, see tasks and scripts
2026-09-26 08:35 #33 docs + features
  • added doc: databases: every database, how each is copied, and how to restore it
  • added feature: portainer's S3 backups keep 14 nights, and always the newest 14. a cron job on truenas deletes older ones, see portainer to S3
  • removed feature: two anonymous accounts on nginx proxy manager's database, which nothing used
  • changed doc: backups: portainer's database has no scheduled consistent copy, and truenas1's app databases have no copy off their pool
2026-09-26 07:28 #32 docs
  • changed doc: changelog: publishes first, each change labelled, and the newest publish's time filled in when it goes live
2026-09-25 23:27 #31 docs
  • added doc: changelog: every publish, and the pace by week
2026-09-25 23:06 #30 docs + features
  • added feature: hourly consistent dumps of the wordpress and nginx proxy manager databases, for the cephfs backup
  • added feature: gatus copies its database every hour, through the fork
  • changed feature: wordpress keeps two days of binary logs, down from thirty
  • changed doc: cephfs backups: what runs for each database, and how to restore each copy
  • changed doc: stack conventions: the adguards are the one swarm service with a healthcheck, and why
2026-09-25 18:50 #29 docs + features + fixes
  • fix: re-plugging a docker VM's network card removed adguard's macvlan address. a healthcheck on that address now gets the container replaced
  • fix: the macvlan shim is built by eth0's stanza, so it comes back with the card
  • changed feature: the two adguards keep off each other's node
2026-09-25 17:28 #28 docs + features
  • added feature: cephfs backups run in a container on the proxmox cluster that any node can run: hourly snapshots, the backup to PBS, and a restore check
  • added doc: building and rebuilding pbs1, from the truenas container api to its certificate
  • changed feature: pbs1 has one static address
  • changed doc: proxmox backup server links to the build
2026-09-25 13:07 #27 docs + features
2026-09-25 08:02 #26 docs + features + fixes
2026-09-22 23:33 #25 docs + features
  • added doc: IPv6 on the docker hosts: the addressing plan, the interface and daemon config, and how a stack turns it on
  • added feature: the docker hosts route IPv6 to containers, with no NAT66
  • added feature: a macvlan shim on every docker host, so a node reaches the adguard running on it
  • added feature: a cold copy of portainer before each upgrade, on portainer to S3
  • changed doc: portainer: each environment is added by its agent's address on port 9001
2026-09-22 12:36 #24 docs
  • changed doc: stacks in git: why portainer's force redeployment stays off, and a current screenshot of the deploy ruleset
2026-09-22 09:55 #23 docs + features
2026-09-20 08:54 #22 docs + features
2026-09-19 16:36 #21 docs
2026-09-19 16:04 #20 docs
2026-09-16 12:37 #19 docs
  • changed doc: numbered steps that restarted at 1 now count through, on eight proxmox pages
2026-09-16 11:53 #18 docs + features
  • added doc: omni-tools and bentopdf
  • added feature: omni-tools and bentopdf on the swarm
  • changed doc: stacks in git: which git reference to pick when adding a stack
2026-09-15 21:34 #12 docs
  • changed doc: the site's theme, mkdocs-material 9.5.44 to 9.7.7
2026-09-15 21:25 #17 docs
  • changed doc: stacks in git: what a standalone docker host shares with the swarm
2026-09-15 21:08 #16 docs
  • changed doc: thread: clipboard and state commands that run on each platform
2026-09-15 19:15 #15 docs
  • changed doc: thread: which ESP-IDF tree windows uses, and naming each board after its MAC as a normal step
2026-09-15 18:24 #14 docs
  • changed doc: thread starts from an unflashed ESP board: parts, ESP-IDF on each OS, flashing, and joining from home assistant
  • added doc: an optional firmware patch that names each ESP board after its MAC
  • changed doc: raspberry pi: the stacks portainer deploys there from git
  • changed doc: 34 lists that rendered as paragraphs now render as lists, and numbered steps no longer restart at 1 in 19 places
2026-09-15 16:58 #13 docs + features
2026-08-26 11:05 a937c79 docs
  • changed doc: the swarm install steps render as a numbered list
2026-08-24 12:03 #11 docs + features
2026-08-23 19:05 7e36498 docs
  • added doc: comments on the ten busiest pages, through github discussions
2026-08-23 18:53 765b1e5 docs
  • added doc: a list of the 16 gists not yet moved to the site
2026-08-23 18:43 100397a docs
  • added doc: the site: my proxmox and docker swarm gists, brought together in mkdocs

pace

  • 39 publishes in 45 days, from 2026-08-23 18:43 to 2026-10-06 17:43. the busiest day was 2026-09-15, with six publishes
  • 323 pull requests merged in the private repo since the first publish: 116 to the docs, 207 to stacks, tools and CI. the busiest day was 2026-09-24, with 50
week starting publishes docs changes stack, tool and CI changes
2026-10-05 2 3 3
2026-09-28 9 20 39
2026-09-21 12 60 80
2026-09-14 11 27 51
2026-09-07 0 0 0
2026-08-31 0 0 0
2026-08-24 2 4 24
2026-08-17 3 2 10