otel: agents reach their hive's collector by a swarm-wide name

Agents are handed `http://hive-collector.<swarm domain>:<port>` as their
otel endpoint instead of `http://<bridge IP>:<port>`. Every hive running
its collector contributes the name to `gateway.localNames`, so its own
dnsmasq answers it with its bridge IP, the way the queue's name already
works. The rendered meta flake then carries no hive-specific address,
which a swarm-level render needs (#4857, operator ruling 100744: "b").

hive-c0re's own exporter keeps the bridge address: on the host the name
may resolve to loopback through `gateway.localHostsEntry`, where the
receiver does not listen.

The name sits under the swarm domain and starts with `hive`, a fragment
no hive name may contain, so no hive's default domain can take it.

docs/networking/network.md

@@ -116,9 +116,10 @@ schemes pick their own.
 dnsmasq is **authoritative** for the hive's own zone (`<hive-domain>`)
 plus whatever swarm-service names this host contributes via
 `gateway.localNames` — `forge.<swarm-domain>` and `chat.<swarm-domain>`
-(matrix) when this host runs those services, and `auth.<swarm-domain>`
+(matrix) when this host runs those services, `auth.<swarm-domain>`
-when it runs authelia — answering each with the bridge IP (where nginx
+when it runs authelia, and `hive-collector.<swarm-domain>` on every hive
-is reachable). Note forge and matrix are swarm-domain names, not
+with `otel.enable` — answering each with the bridge IP (where nginx
+is reachable, and where the hive's own collector listens). Note forge and matrix are swarm-domain names, not
 sub-domains of `<hive-domain>`: a swarm runs one forge and one
 homeserver, so their names belong to the swarm rather than to whichever
 hive happens to host them. dnsmasq forwards everything else to the host's

docs/scheduler/observability.md

@@ -42,8 +42,10 @@ destination in its own right. With both, telemetry goes to both. See
 **Telemetry leaves a hive exactly one way: through the collector that
 `enable` starts on the host.** Agents never talk to `endpoint` themselves —
 they export unauthenticated to a bridge address only their own containers can
-reach (`http://<bridgeIp>:<collector port>`; the otel module opens that port
+reach. They dial it as `http://hive-collector.<swarm domain>:<collector port>`,
-on the bridge itself). That collector forwards to the swarm's
+a name identical on every hive, which each hive's own dnsmasq
+answers with its bridge IP, so an agent's build carries no hive-specific
+address. The otel module opens that port on the bridge itself. That collector forwards to the swarm's
 ([`swarm/services.md`](../swarm/services.md#telemetry-collector-otel)), the
 single process holding the upstream credential and the only writer to the
 swarm's stores. No agent holds a copy, and neither does the hive.

hive-c0re/src/stats/otel_metrics.rs

@@ -2,7 +2,8 @@
 //! sampled in [`super::container_stats`] feeds: it rides those gauges out to the
 //! hive's own collector — the same first hop the agents use, named by the
 //! standard `OTEL_EXPORTER_OTLP_ENDPOINT` (see `nix/host-modules/hive-c0re`,
-//! which sets it from the same binding it hands agents). No toggle of
+//! which sets it to the collector's bridge address where agents get its
+//! swarm-wide name). No toggle of
 //! its own, and no credential: a hive's collector takes unauthenticated OTLP
 //! on the bridge, and the only hop that presents anything is the swarm tier's,
 //! which is the one that leaves the swarm.

nix/host-modules/hive-c0re/environment.nix

@@ -83,11 +83,13 @@ in
   # don't render no-op env lines.
   let
     otel = config.services.hyperhive.otel;
-    # The first hop, bound once and consumed twice below: what agents are
+    # The first hop, this hive's own collector, under the two spellings its
-    # handed, and where hive-c0re's own exporter sends. One binding so the
+    # two kinds of caller need. Agents get the swarm-wide name, so their
-    # address a hive tells its agents about and the one it uses itself
+    # builds carry nothing hive-specific; this daemon gets the bridge
-    # cannot drift apart.
+    # address, because on the host the name may resolve to loopback
-    firstHop = "http://${config.services.hyperhive.network.bridgeIp}:${toString otel.collector.port}";
+    # (`gateway.localHostsEntry`), where the receiver does not listen.
+    agentFirstHop = "http://${otel.collector.host}:${toString otel.collector.port}";
+    hostFirstHop = "http://${config.services.hyperhive.network.bridgeIp}:${toString otel.collector.port}";
   in
   {
     # `otel.endpoint` means "where telemetry ultimately goes" and keeps
@@ -95,7 +97,7 @@ in
     # always this hive's own collector. Deriving it rather than
     # redefining `endpoint` is what lets every existing deployment keep
     # its configured value untouched.
-    HYPERHIVE_OTEL_ENDPOINT = firstHop;
+    HYPERHIVE_OTEL_ENDPOINT = agentFirstHop;
     # hive-c0re's OWN container-resource exporter (stats/otel_metrics.rs)
     # reads the STANDARD OTLP variable — the same one hive-metric and every
     # agent read — and lets the SDK resolve the URL, which appends the
@@ -108,7 +110,7 @@ in
     # replace this one:
     # it is the agent-config transport meta.rs reads, and the SDK does not
     # know that name.
-    OTEL_EXPORTER_OTLP_ENDPOINT = firstHop;
+    OTEL_EXPORTER_OTLP_ENDPOINT = hostFirstHop;
     # The first hop is the collector's OTLP/HTTP receiver, which speaks
     # protobuf regardless of what the upstream wants — `otel.protocol`
     # describes the *upstream* link, and the collector's own exporter is

nix/host-modules/otel.nix

@@ -203,6 +203,28 @@ in
       '';
     };
 
+    collector.host = lib.mkOption {
+      type = lib.types.str;
+      readOnly = true;
+      # Total on a null swarm domain, as in ./swarm-otel-service.nix: the
+      # required-domain assertion should be what an operator sees.
+      default = "hive-collector.${
+        if config.services.hyperhive.swarm.domain == null then
+          "invalid"
+        else
+          config.services.hyperhive.swarm.domain
+      }";
+      defaultText = lib.literalExpression ''"hive-collector.''${services.hyperhive.swarm.domain}"'';
+      description = ''
+        The name agents reach this hive's collector on. The same string on
+        every hive: each hive's own resolver answers it with its own bridge
+        IP, so an agent's build carries no hive-specific address.
+
+        No hive can claim this name as its domain:
+        `nix/reserved-hive-fragments.nix` forbids `hive` in any hive name.
+      '';
+    };
+
     telemetryPort = lib.mkOption {
       type = lib.types.port;
       default = 8888;
@@ -357,6 +379,9 @@ in
       # the port on the bridge interface only.
       services.hyperhive.network.exposeHostPorts = [ otel.collector.port ];
 
+      # The name agents dial, answered with the bridge IP the receiver binds.
+      services.hyperhive.gateway.localNames = [ otel.collector.host ];
+
       # …and the receiver binds that bridge address, so the interface has
       # to exist. `exposeHostPorts` is a firewall hole, not a demand.
       services.hyperhive.network.enable = lib.mkDefault true;

nix/module-eval/core-toggle.nix

@@ -199,6 +199,20 @@ let
         && !(lib.hasInfix "127.0.0.1" e.HIVE_AGENT_NATS_URL)
         && e.HIVE_AGENT_NATS_URL == e.HIVE_C0RE_NATS_URL;
     }
+    {
+      # What agents are handed lands in their builds, so it has to be the same
+      # on every hive: a name, which this hive's resolver answers with the
+      # bridge IP its collector binds. hive-c0re itself keeps the address.
+      name = "the agents' collector address is a swarm-wide name this hive resolves to its bridge";
+      ok =
+        let
+          e = allLocal.systemd.services.hive-c0re.environment;
+          bridge = allLocal.services.hyperhive.network.bridgeIp;
+        in
+        e.HYPERHIVE_OTEL_ENDPOINT == "http://hive-collector.t.local:4318"
+        && e.OTEL_EXPORTER_OTLP_ENDPOINT == "http://${bridge}:4318"
+        && lib.elem "/hive-collector.t.local/${bridge}" allLocal.services.dnsmasq.settings.address;
+    }
     {
       # The agents mint against the swarm's IdP, the same endpoint the hive's
       # own client uses — a hive-local guess would produce a token the queue