vinzenz/hyperhive
vinzenz/hyperhive
Watch
0
0
Fork
You've already forked hyperhive
 0
Code Activity

glue-controller-env: the controller always exports otlp to the swarm's collector

Browse source
This commit is contained in:
atlas 2026-10-17 14:50:49 +02:00 • committed by mara
commit 2507f4f713
3 changed files with 45 additions and 17 deletions
Download patch file Download diff file Expand all files Collapse all files

29
nix/host-modules/glue-controller-env.nix
View file

@ -1,27 +1,31 @@
# Glue: tell the swarm controller about the authelia and the collector that run
# on this host.
# Glue: tell the swarm controller where the swarm's collector is and which
# authelia runs on this host.
#
# ONE PAIRING PER FILE — controller ← the services whose coordinates it reads
# from the environment. Deleting this leaves a controller that exports no
# metrics, and a `swarmctl` that fails `user add` saying the authelia binary and
# users file are unset.
#
# ⚠️ Two gates, one per service, each ANDed with the controller being here:
# ⚠️ Everything here is gated on the controller being here:
# `systemd.services.swarm-controller.environment` on a host without the
# controller would define a unit with no ExecStart.
#
# authelia → `swarmctl`'s wrapper, not the daemon's unit. Gated on authelia
# authelia → `swarmctl`'s wrapper, not the daemon's unit. Also gated on authelia
# running HERE: the controller can be enabled on a host that is not the swarm's
# SSO host, and there `swarmctl user add` must fail saying the value is unset —
# a guessed path would resolve cleanly and write a file nothing reads. The
# CONFIGURED authelia, not whatever is on PATH: the argon2 parameters baked into
# a hash have to match the verifier's.
#
# swarm-otel → the daemon's unit. `vcs_metrics`/`hive_jobq_metrics` read the
# standard `OTEL_EXPORTER_OTLP_ENDPOINT` and are silent no-ops without it. The
# endpoint is the collector's swarm-service name through the gateway, never a
# collector → the daemon's unit. `vcs_metrics`/`hive_jobq_metrics` read the
# standard `OTEL_EXPORTER_OTLP_ENDPOINT` and are silent no-ops without it. No
# gate beyond the controller: the controller always exports to the swarm's
# collector, which the swarm always has (./swarm-bao.nix,
# ./glue-swarm-otel-oidc-client.nix), and `deploy.swarm-otel.enable` only says it
# runs HERE. The endpoint is its swarm-service name through the gateway, never a
# loopback literal, with no trailing slash: the http exporter appends
# `/v1/metrics` and the gateway route strips `/${producerName}`.
# `/v1/metrics` and the gateway route strips `/${producerName}`. The port suffix
# is elided on 443, as in ./swarm-bao.nix.
# `SWARM_CONTROLLER_OTEL_AUDIENCE` is the audience ./glue-controller-oidc-client.nix
# registers and ./swarm-otel.nix's `oidc/${producerName}` authenticator checks.
{
@ -34,6 +38,9 @@ let
deployCfg = hyperhiveCfg.deploy;
otelSwarmCfg = hyperhiveCfg.swarm.otel;
queueClientId = hyperhiveCfg.swarm.controller.queueClientId;
gatewayPortSuffix = lib.optionalString (
hyperhiveCfg.gateway.httpsPort != 443
) ":${toString hyperhiveCfg.gateway.httpsPort}";
in
{
config = lib.mkIf deployCfg.swarm-controller.enable (
@ -44,12 +51,12 @@ in
SWARMCTL_AUTHELIA_USERS_FILE = deployCfg.authelia.hostUsersFile;
};
})
(lib.mkIf deployCfg.swarm-otel.enable {
{
systemd.services.swarm-controller.environment = {
OTEL_EXPORTER_OTLP_ENDPOINT = "https://${otelSwarmCfg.domain}/${otelSwarmCfg.producerName}";
OTEL_EXPORTER_OTLP_ENDPOINT = "https://${otelSwarmCfg.domain}${gatewayPortSuffix}/${otelSwarmCfg.producerName}";
SWARM_CONTROLLER_OTEL_AUDIENCE = queueClientId;
};
})
}
]
);
}