Repository navigation
kuma v45.3.0
-
The CI's apt routes around the azure mirror. Canonical's 2026-11-11 archive incident killed four CI rolls in ninety minutes — all "Temporary failure resolving 'azure.archive.ubuntu.com'", the traffic-manager CNAME chain GitHub's runner image points every apt source at, dead on the runners while the same name answered from off-Azure and the runners' other fetches (rustup, crates) sailed through. The new
apt-routeaction rewrites the mirror file to fullhttp://URLs onarchive.ubuntu.com/security.ubuntu.com— different domains, different DNS; the first draft's bare-host lines made apt parse the host as the method driver, its own one-roll lesson — and the action rides before every apt-get the workflows run. The same class of fix the smoke already applies to Fedora's flapping mirrors: the road bends around upstream weather instead of waiting in it. -
The heavy jobs carry their measured ceilings. The 45.2 audit gave the session daemons
MemoryMax=256Mand deliberately left the two heavy jobs uncapped until sized — now measured on a real machine (101G, 1.8M files, restic 0.19.1): a cold full backup peaks at 7.75G, the nightly incremental at 853M, a repo prune at 284M, the flatpak daily's appstream re-pull at 76M.kuma-backupgets the softMemoryHigh=8Gjust over its one heavy day and aMemoryMax=12Gleak line; the flatpak daily getsMemoryHigh=512M/MemoryMax=1G. A cap below the cold peak would have broken the legitimate backup outright — the sizing is the point. -
Doctor's mime check attributes and prunes. A dead mime default used to report as
mime → handlerwith no source: the two files it reads have different owners (the image bakes /etc's pin, the user wrote ~/.config's), and the reader could not tell whose fix it was. Each entry now names its file, and when every dead default is the user's own single-handler line, the warning carries a prune Action — a sed that deletes exactly those lines, its pattern escaped for the BRE's any-char dots and bare+/?(the quantifiers are the ESCAPED forms in a GNU BRE; the suggested command could not widen past the dead entry). -
The prune Action's command survives the shell that would run it. The review caught the first draft's sed printed its BRE pattern UNQUOTED — the pattern's
;is a shell command separator, so the suggested command failed on every input (sed: no input files, then the shell trying\?$#das a second command) and touched no file. Every-eexpression is single-quoted now and the target an absolute path (doctor runs under sudo; a pasted~expands to root's home, where the file does not exist). The line's fields are user-file text: they pass throughplain()at parse,sed_brein the pattern, and the single quotes at the shell — and a field carrying a single quote refuses the Action entirely, that line staying advice rather than a broken paste. -
The dev deploy reads the unit's LAST non-override ExecStart — the 45.2.1 bug's twin, caught before it bit.
03094c9fixed the empty- relay deploy by reading the image fragment's arguments out ofsystemctl cat, and the first draft took the FIRST non-override assignment — on a delta machine, the declaration's relay drop-in (10-kuma-nostr-relays.conf) sorts before the dev override and restates ExecStart, so the parse adopted the image's relays over the declaration's, silently, on exactly the machines whose declaration had moved. The parse now takes the LAST non-override assignment (systemd's own rule: a later non-empty assignment replaces an earlier one) and only a real fragment header — a path — may move which fragment a line belongs to, so the override's own comments cannot flip the state and its bare ExecStart is never read back as the argument source. Pinned against the real drop-in interleaving. -
Doctor reads the wire facts the daemon already reports — the 45.2.1 read-side lesson. The release's signature bug was a signer that ran, listened, and heard nothing: daemon active, vault open, EMPTY relay set — and
kuma doctorgraded it Ok, because the check read two of the four facts the status document carried and the two it skipped were the wire's. Now graded: the daemon'srelaysandconnected— an empty relay set warns naming the override/drop-in suspect, no live connection warns, and the daemon's set is diffed against the declaration's (the local relay added when the layer arms it) with a mismatch namingsystemctl --user cat kuma-nostrdas the interrogation. The reachability line stays declarative — the word "answers" belongs to the runtime grades, and 45.2.1 is the proof of why. And the socket client bounds every ask at ten seconds: the daemon side holds its mutex across keyring D-Bus calls, so a wedged daemon used to be able to hang the very probe that should have reported the hang; now it grades a warn. -
Drift is measured against the authority convergence follows. The bare
kumaprobe counted drift against the image's baked list file directly — an authority the convergers stopped consulting when the nano-incident ruling moved the answer to the CLI: the machine's own declaration where it names the axis, the baked list otherwise. A machine whose declaration dropped an app the release still bakes read as "drifted" by an app no sync would ever touch, promising work the actor no longer meant to do — the second incident of the summary-reads-one-source class that inventory.rs already records from the brew half. The probe now asksgoverningfor the same answer sync acts on; an authority that cannot answer stays quiet, because sync's refusal is the loud path and a guessed baseline was the bug. -
A delta re-bakes the lists it governs. The delta compiler's guard compared the declared list against the baseline DECLARATION — an artifact it was not rewriting — so every delta whose baseline already agreed with the declaration skipped the bake, and the release's list file rode forward through the delta chain unexamined: a machine whose declaration dropped a release default kept the release's list in its image for as long as deltas were the upgrade road. The re-bake now follows the runtime's authority shape (a declaration that names the axis governs it, empty included; one that omits it leaves the base's list standing), and the overrides tree is written unconditionally — the full build's own rule, and the same skip-forever shape removed from the lists.
-
Doctor reads identity, not presence, for shadowing binaries. An unreceipted binary in
~/.local/binused to warn for existing; shadowing is the dev-deploy mechanism itself, so presence alone is not the problem. The check now compares bytes against the image's own binary — byte-identical is the kumaui build script's legitimate staging and reads ok; anything else warns naming what it is. Bytes are read, never executed: doctor is root, and root running a user-writable binary is the opinion-refusal rule's business. A new sweep flags cargo-installed kuma binaries in~/.cargo/bin— the one deposit no receipt knows and no gate resolves, found masquerading as the CLI for a whole reboot cycle. -
A dev deploy keeps the daemon's arguments.
kuma dev installrewrote the daemon's override with the staged binary alone — and the image's unit bakes its relay set onto the same ExecStart line, so the adopted daemon came up with an EMPTY relay set: no relay threads, no wire, no error anywhere, a bunker that listens and never hears a note. The override now carries the image unit's own arguments beside the staged binary, the parse is pinned by test, and a unit with no ExecStart fails the adoption loudly instead of inventing a bare daemon.
Static x86_64 binary, no runtime dependencies.
$ curl -LO https://github.com/Letdown2491/kumaOS/releases/download/v45.3.0/kuma-x86_64-unknown-linux-musl
$ chmod +x kuma-x86_64-unknown-linux-musl && sudo mv kuma-x86_64-unknown-linux-musl /usr/local/bin/kumaThe image build stages the layer's three binaries from beside
the running kuma on every compose, nostr or not, so they
install to the same place:
$ curl -LO https://github.com/Letdown2491/kumaOS/releases/download/v45.3.0/kuma-nostrd-x86_64-unknown-linux-musl
$ curl -LO https://github.com/Letdown2491/kumaOS/releases/download/v45.3.0/kuma-nostr-x86_64-unknown-linux-musl
$ curl -LO https://github.com/Letdown2491/kumaOS/releases/download/v45.3.0/nip46-relay-x86_64-unknown-linux-musl
$ chmod +x kuma-nostrd-x86_64-unknown-linux-musl kuma-nostr-x86_64-unknown-linux-musl nip46-relay-x86_64-unknown-linux-musl
$ sudo mv kuma-nostrd-x86_64-unknown-linux-musl /usr/local/bin/kuma-nostrd
$ sudo mv kuma-nostr-x86_64-unknown-linux-musl /usr/local/bin/kuma-nostr
$ sudo mv nip46-relay-x86_64-unknown-linux-musl /usr/local/bin/nip46-relayVerify it came from this workflow:
$ cosign verify-blob \
--bundle kuma-x86_64-unknown-linux-musl.bundle \
--certificate-identity-regexp '^https://github.com/Letdown2491/kumaOS/.+@refs/tags/' \
--certificate-oidc-issuer https://token.actions.githubusercontent.com \
kuma-x86_64-unknown-linux-muslThe image this release publishes verifies with the committed
public key — the same key every kuma machine's own policy
enforces on update:
$ curl -sSLO https://github.com/Letdown2491/kumaOS/raw/v45.3.0/cosign.pub
$ cosign verify --key cosign.pub ghcr.io/letdown2491/kuma:niri-45.3.0The moving tag is the convenience and the digest is the proof;
the verify response prints the digest it checked, and the same
command against that digest pins the bytes:
$ cosign verify --key cosign.pub ghcr.io/letdown2491/kuma@sha256:<digest the verify above printed>Built from 7550ab5aaec768eda4b6dc5600b80c2b5769c426.
The installer media on this page verifies the same way:
$ cosign verify-blob \
--bundle kuma-x86_64.iso.bundle \
--certificate-identity-regexp '^https://github.com/Letdown2491/kumaOS/.+@refs/tags/' \
--certificate-oidc-issuer https://token.actions.githubusercontent.com \
kuma-x86_64.iso