swarm-bao-matrix-oidc still reads the secret from bao under matrix-ctl's
leaf, but now writes it to /run/hive-matrix-oidc on the host (tmpfs), which
the container bind-mounts read-only for tuwunel's LoadCredential. Before, it
wrote it to
/var/lib/nixos-containers/hive-matrix/var/lib/tuwunel-oidc/<id>.secret on
persistent disk, where a superseded secret stayed after rotation and went
into backups of the container root. Host tmpfiles now removes that old file.
clientSecretFile must now lie under /run/hive-matrix-oidc/, the one directory
the container shares; any other value fails at evaluation.
Rotation is unchanged: the ten-minute timer rewrites the file when the value
differs and refresh_consumer restarts tuwunel. After a reboot the secret is
fetched again; until the store answers, tuwunel does not start.
Part of #4998 (slice 4: tuwunel).
Thirty-three module-eval fixtures were a second (or later) full
`nixosSystem` evaluation of a config another fixture already evaluated:
16 shapes defined identically in two to seven cluster files, plus
duplicates inside agent-queue-bao, core-toggle and
swarm-services-switch. Each cluster imports ./lib.nix itself, so nothing
was shared between them.
nix/module-eval/shared.nix now defines each of those 16 shapes once.
nix/checks.nix instantiates it a single time and passes it as `shared`
to the 23 clusters that read one, and each cluster rebinds the fixture
under its existing local name. Assertions are unchanged. Case text
changes only in core-toggle, where two cases built the forge-host
fixture inline and now read the file's `forgeHere`, and in two inline
`let` bindings (bao-otel-collector, grafana) that now take the shared
fixture.
Six folds are value-equal rather than textually equal: they drop an
explicit `deploy.hive-controller.enable = false` or
`deploy.swarm-otel.enable = false` that equals what the option already
resolves to for that config. Nothing under nix/host-modules sets either
option apart from the `mkDefault allSwarmServices` on swarm-otel, and
the only option-priority read there is `domain.highestPrio`.
Part of #5171.
The shared `hive` fixture turned `deploy.hive-controller.enable` on for
every eval test. It is now off by default; `hiveWithController` turns it
on, and the fixtures whose claims read the hive tier (hive-c0re, the
queue-agent credential, the hive leaf, `deploy.hive-controller.*`, or an
absence that is only meaningful with a hive present) use it.
glue-nats-auth-bao-identity.nix still read the renamed
`services.hyperhive.enable`, which printed the "Obsolete option" trace
six times per CI run. The term is dropped: the pairing is gated on the
queue and the minted PKI, as its sibling glue-nats-bao-identity.nix is.
An unpublished or empty secret in swarm-bao-nats-auth-oidc and
swarm-bao-matrix-oidc now exits 1 instead of 0, the shape
swarm-bao-forwarder-oidc already uses. storeRetry then re-fetches every 30s
until swarm-secret-publish has written the secret, and refresh_consumer
starts the responder / tuwunel once the file lands. Before, a fresh swarm's
first fetch found nothing, exited 0, and neither consumer started until the
unit was restarted by hand.
The three client registrations move into glue-{controller,nats,matrix}-oidc-client.nix,
gated only on deploy.authelia.enable and reading their ids from swarm.* options,
so the host running authelia registers every one of them whichever host runs
the service. The matrix callback is a new read-only swarm.matrix.sso.redirectUri,
read by both the homeserver's config and its client entry.
swarm-secret-publish now also publishes the queue's and the homeserver's minted
secrets to swarm/services/<id>/oidc/client. The queue host reads its responder's
secret there with swarm-bao-nats-auth-oidc under the responder's own leaf
(swarm-nats-auth role gains that one read path), and the homeserver's host with
swarm-bao-matrix-oidc under matrix-ctl's leaf (swarm-matrix-ctl role gains its
one read path). The host-file copies are gone: swarm-nats-auth-secrets now
delivers only the callout seeds and the store identity, and
hive-matrix-oidc-secret is removed. A fetch unit whose leaf is unset still
renders and fails naming the option pair.
Closes#4943
Every hive's homeserver loaded a per-hive `hyperhive` appservice
registration, and the swarm minted and delivered an as_token for it
through bao. No client presented that token: swarm-controller creates
every `@hive-<hive>` sender and every agent account with the swarm
appservice token (hive_sender.rs, agent_token.rs), and no Rust code read
the per-hive one.
Removed, with everything that existed only to serve it:
- hive-matrix.nix: the `hyperhive` registration, its activation-time
render, the host bind mount of /var/lib/hyperhive/matrix-appservice,
its LoadCredential entry, and the internal options
`deploy.matrix.appserviceTokenFile` / `appserviceRegistrationScript`.
tuwunel now loads only the swarm registration.
- glue-matrix-bao-token.nix and its `swarm-bao-matrix-token` unit.
- glue-bao-tls.nix: the matrix-token leaf and its defaults.
- swarm-bao.nix: the `swarm-matrix-token-<hive>` roles and policies
(`swarm-bao-matrix-token-policy`), and the options
`deploy.bao.matrixTokenClient{Cert,Key}File` and
`matrixTokenCommonNamePrefix`, now mkRemovedOptionModule.
- swarm-secret-publisher.nix: the per-hive token mint and its state dir.
- swarm.nix: the reserved `<matrixTokenCommonNamePrefix>-<hive>` CN.
- swarm-secret-client: `matrix::appservice_token_path`.
The swarm registration (swarm-matrix-ctl, swarm-controller) is untouched.
Comments and docs that pointed at the removed glue file now point at
glue-queue-agent-credential.nix, which has the same reader shape.
Refs #4428
hive-c0re logs in with the hive's matrix sender token and creates the
Space and rooms over swarm.matrix.apiUrl. On a hive that does not run
the homeserver that crosses the network, so the default is now
https://<gatewayHost>. hive-c0re already verifies against the hive trust
bundle (SSL_CERT_FILE: system CAs, swarm root and the store's services
root), which covers the gateway's leaf for that name.
Also drops the remaining loopback-default wording in hive-c0re's matrix
module and the openFirewall description, and pins the https default on a
hive host without the homeserver in module-eval-matrix-core.
The single module-eval derivation forced ~62 full nixosSystem
fixtures live at once to compute its cases list: 10.6GB peak RSS /
5m25s to evaluate, by far the dominant cost in nix flake check.
Splits it into 21 independent checks.module-eval-* derivations
(1-7 fixtures each) sharing builders/helpers via module-eval/lib.nix,
so no single derivation needs more than a handful of fixtures live
at once. A few cases spanning two clusters carry a small duplicated
fixture rather than threading shared state through lib.nix.
