checkin: report host disk free/total bytes (#41)

/recordings is a bind mount of the host checkout, so shutil.disk_usage on it
is the disk the recorder writes to and Docker fills. Explicit nulls on error
so C2 never keeps a stale reading. Server half: server-26 feat/41-node-disk.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

.gitea/workflows/build-edge-node.yml

@@ -38,15 +38,20 @@ jobs:
         run: |
           echo "REPO_NAME=$(echo ${GITHUB_REPOSITORY} | awk -F'/' '{print $2}')" >> $GITHUB_OUTPUT
           echo "VERSION=$(git describe --tags --always | sed 's/^v//')" >> $GITHUB_OUTPUT
+          # Source pin (node-26#19): the git tree hash of the build context.
+          # A node pulls src-<tree of its own checkout> — scripts/image-tags.sh
+          # computes the identical value — so its code and image always match.
+          echo "SRC_TAG=src-$(git rev-parse HEAD:drb-edge-node)" >> $GITHUB_OUTPUT
 
       - uses: docker/build-push-action@v6
         with:
           context: ./drb-edge-node
           file: ./drb-edge-node/Dockerfile
-          platforms: linux/arm64
+          platforms: linux/amd64,linux/arm64
           push: true
           tags: |
             git.vpn.cusano.net/${{ vars.DOCKER_ORG }}/${{ steps.meta.outputs.REPO_NAME }}/${{ env.CONTAINER_NAME }}:${{ steps.meta.outputs.VERSION }}
+            git.vpn.cusano.net/${{ vars.DOCKER_ORG }}/${{ steps.meta.outputs.REPO_NAME }}/${{ env.CONTAINER_NAME }}:${{ steps.meta.outputs.SRC_TAG }}
             git.vpn.cusano.net/${{ vars.DOCKER_ORG }}/${{ steps.meta.outputs.REPO_NAME }}/${{ env.CONTAINER_NAME }}:latest
           cache-from: type=registry,ref=git.vpn.cusano.net/${{ vars.DOCKER_ORG }}/${{ steps.meta.outputs.REPO_NAME }}/${{ env.CONTAINER_NAME }}:buildcache
           cache-to: type=registry,ref=git.vpn.cusano.net/${{ vars.DOCKER_ORG }}/${{ steps.meta.outputs.REPO_NAME }}/${{ env.CONTAINER_NAME }}:buildcache,mode=max

.gitea/workflows/build-icecast.yml

@@ -38,15 +38,20 @@ jobs:
         run: |
           echo "REPO_NAME=$(echo ${GITHUB_REPOSITORY} | awk -F'/' '{print $2}')" >> $GITHUB_OUTPUT
           echo "VERSION=$(git describe --tags --always | sed 's/^v//')" >> $GITHUB_OUTPUT
+          # Source pin (node-26#19): the git tree hash of the build context.
+          # A node pulls src-<tree of its own checkout> — scripts/image-tags.sh
+          # computes the identical value — so its code and image always match.
+          echo "SRC_TAG=src-$(git rev-parse HEAD:icecast)" >> $GITHUB_OUTPUT
 
       - uses: docker/build-push-action@v6
         with:
           context: ./icecast
           file: ./icecast/Dockerfile
-          platforms: linux/arm64
+          platforms: linux/amd64,linux/arm64
           push: true
           tags: |
             git.vpn.cusano.net/${{ vars.DOCKER_ORG }}/${{ steps.meta.outputs.REPO_NAME }}/${{ env.CONTAINER_NAME }}:${{ steps.meta.outputs.VERSION }}
+            git.vpn.cusano.net/${{ vars.DOCKER_ORG }}/${{ steps.meta.outputs.REPO_NAME }}/${{ env.CONTAINER_NAME }}:${{ steps.meta.outputs.SRC_TAG }}
             git.vpn.cusano.net/${{ vars.DOCKER_ORG }}/${{ steps.meta.outputs.REPO_NAME }}/${{ env.CONTAINER_NAME }}:latest
           cache-from: type=registry,ref=git.vpn.cusano.net/${{ vars.DOCKER_ORG }}/${{ steps.meta.outputs.REPO_NAME }}/${{ env.CONTAINER_NAME }}:buildcache
           cache-to: type=registry,ref=git.vpn.cusano.net/${{ vars.DOCKER_ORG }}/${{ steps.meta.outputs.REPO_NAME }}/${{ env.CONTAINER_NAME }}:buildcache,mode=max

.gitea/workflows/build-op25.yml

@@ -38,15 +38,20 @@ jobs:
         run: |
           echo "REPO_NAME=$(echo ${GITHUB_REPOSITORY} | awk -F'/' '{print $2}')" >> $GITHUB_OUTPUT
           echo "VERSION=$(git describe --tags --always | sed 's/^v//')" >> $GITHUB_OUTPUT
+          # Source pin (node-26#19): the git tree hash of the build context.
+          # A node pulls src-<tree of its own checkout> — scripts/image-tags.sh
+          # computes the identical value — so its code and image always match.
+          echo "SRC_TAG=src-$(git rev-parse HEAD:op25-container)" >> $GITHUB_OUTPUT
 
       - uses: docker/build-push-action@v6
         with:
           context: ./op25-container
           file: ./op25-container/Dockerfile
-          platforms: linux/arm64
+          platforms: linux/amd64,linux/arm64
           push: true
           tags: |
             git.vpn.cusano.net/${{ vars.DOCKER_ORG }}/${{ steps.meta.outputs.REPO_NAME }}/${{ env.CONTAINER_NAME }}:${{ steps.meta.outputs.VERSION }}
+            git.vpn.cusano.net/${{ vars.DOCKER_ORG }}/${{ steps.meta.outputs.REPO_NAME }}/${{ env.CONTAINER_NAME }}:${{ steps.meta.outputs.SRC_TAG }}
             git.vpn.cusano.net/${{ vars.DOCKER_ORG }}/${{ steps.meta.outputs.REPO_NAME }}/${{ env.CONTAINER_NAME }}:stable
           # Registry cache — avoids full recompile of GnuRadio/OP25 when only app code changes
           cache-from: type=registry,ref=git.vpn.cusano.net/${{ vars.DOCKER_ORG }}/${{ steps.meta.outputs.REPO_NAME }}/${{ env.CONTAINER_NAME }}:buildcache

.gitea/workflows/build-secondary-sdr.yml

@@ -38,15 +38,20 @@ jobs:
         run: |
           echo "REPO_NAME=$(echo ${GITHUB_REPOSITORY} | awk -F'/' '{print $2}')" >> $GITHUB_OUTPUT
           echo "VERSION=$(git describe --tags --always | sed 's/^v//')" >> $GITHUB_OUTPUT
+          # Source pin (node-26#19): the git tree hash of the build context.
+          # A node pulls src-<tree of its own checkout> — scripts/image-tags.sh
+          # computes the identical value — so its code and image always match.
+          echo "SRC_TAG=src-$(git rev-parse HEAD:secondary-sdr-container)" >> $GITHUB_OUTPUT
 
       - uses: docker/build-push-action@v6
         with:
           context: ./secondary-sdr-container
           file: ./secondary-sdr-container/Dockerfile
-          platforms: linux/arm64
+          platforms: linux/amd64,linux/arm64
           push: true
           tags: |
             git.vpn.cusano.net/${{ vars.DOCKER_ORG }}/${{ steps.meta.outputs.REPO_NAME }}/${{ env.CONTAINER_NAME }}:${{ steps.meta.outputs.VERSION }}
+            git.vpn.cusano.net/${{ vars.DOCKER_ORG }}/${{ steps.meta.outputs.REPO_NAME }}/${{ env.CONTAINER_NAME }}:${{ steps.meta.outputs.SRC_TAG }}
             git.vpn.cusano.net/${{ vars.DOCKER_ORG }}/${{ steps.meta.outputs.REPO_NAME }}/${{ env.CONTAINER_NAME }}:latest
           cache-from: type=registry,ref=git.vpn.cusano.net/${{ vars.DOCKER_ORG }}/${{ steps.meta.outputs.REPO_NAME }}/${{ env.CONTAINER_NAME }}:buildcache
           cache-to: type=registry,ref=git.vpn.cusano.net/${{ vars.DOCKER_ORG }}/${{ steps.meta.outputs.REPO_NAME }}/${{ env.CONTAINER_NAME }}:buildcache,mode=max

.gitea/workflows/publish-public.yml

@@ -95,7 +95,7 @@ jobs:
         with:
           context: ${{ matrix.context }}
           file: ${{ matrix.file }}
-          platforms: linux/arm64
+          platforms: linux/amd64,linux/arm64
           push: true
           tags: |
             ${{ vars.PUBLIC_REGISTRY }}/${{ vars.PUBLIC_NAMESPACE }}/drb-${{ matrix.name }}:${{ steps.meta.outputs.VERSION }}

.gitea/workflows/release-images.yml

@@ -0,0 +1,91 @@
+name: Release images
+
+# Make sure every node image a release ref needs exists in the registry
+# (node-26#19). A node on `v1` pulls <image>:src-<tree hash of that image's
+# directory at v1>. The per-image build workflows only run when their own
+# directory changes on main, and anything built before #19 has no src- tag at
+# all, so a release tag can point at trees nobody ever built. This fills the
+# gaps: for each image, if src-<tree> is already in the registry it does
+# nothing; otherwise it builds and pushes it (registry cache keeps op25 off a
+# full recompile when only its app code changed).
+#
+# Runs on every v* tag push, including a re-cut `v1`. Also runnable by hand
+# (workflow_dispatch) on any ref — run it once on the ref `v1` points at
+# before re-cutting `v1` onto a commit that contains #19, so nodes never see
+# a moment where their pinned images don't exist yet.
+#
+# Pushes ONLY src- tags. :latest / :stable keep tracking main, as before.
+
+on:
+  workflow_dispatch:
+  push:
+    tags:
+      - "v*"
+
+jobs:
+  ensure:
+    runs-on: ubuntu-latest
+    permissions:
+      contents: read
+      packages: write
+    strategy:
+      fail-fast: false
+      matrix:
+        include:
+          - name: edge-node
+            context: drb-edge-node
+          - name: icecast
+            context: icecast
+          - name: op25-client
+            context: op25-container
+          - name: secondary-sdr
+            context: secondary-sdr-container
+    steps:
+      - uses: actions/checkout@v4
+
+      - uses: docker/setup-qemu-action@v3
+
+      - uses: docker/setup-buildx-action@v3
+        with:
+          config-inline: |
+            [registry."git.vpn.cusano.net"]
+              http = false
+              insecure = false
+
+      - uses: docker/login-action@v3
+        with:
+          registry: git.vpn.cusano.net
+          username: ${{ gitea.actor }}
+          password: ${{ secrets.BUILD_TOKEN }}
+
+      - name: Resolve source tag
+        id: meta
+        run: |
+          REPO_NAME="$(echo ${GITHUB_REPOSITORY} | awk -F'/' '{print $2}')"
+          # Must match scripts/image-tags.sh and the build-*.yml workflows.
+          IMAGE="git.vpn.cusano.net/${{ vars.DOCKER_ORG }}/${REPO_NAME}/${{ matrix.name }}"
+          TAG="src-$(git rev-parse HEAD:${{ matrix.context }})"
+          echo "IMAGE=$IMAGE" >> $GITHUB_OUTPUT
+          echo "TAG=$TAG" >> $GITHUB_OUTPUT
+          # Present only counts if BOTH platforms are in it: src- tags pushed
+          # before node-26#28 are arm64-only, and an amd64 node can't pull them.
+          RAW="$(docker buildx imagetools inspect --raw "$IMAGE:$TAG" 2>/dev/null | tr -d ' \n' || true)"
+          if printf '%s' "$RAW" | grep -q '"architecture":"amd64"' \
+             && printf '%s' "$RAW" | grep -q '"architecture":"arm64"'; then
+            echo "EXISTS=true" >> $GITHUB_OUTPUT
+            echo "$IMAGE:$TAG already published for amd64 + arm64 — nothing to do"
+          else
+            echo "EXISTS=false" >> $GITHUB_OUTPUT
+            echo "$IMAGE:$TAG missing or not multi-arch — building it"
+          fi
+
+      - if: steps.meta.outputs.EXISTS != 'true'
+        uses: docker/build-push-action@v6
+        with:
+          context: ./${{ matrix.context }}
+          file: ./${{ matrix.context }}/Dockerfile
+          # amd64 + arm64, same as the build-*.yml workflows (node-26#28).
+          platforms: linux/amd64,linux/arm64
+          push: true
+          tags: ${{ steps.meta.outputs.IMAGE }}:${{ steps.meta.outputs.TAG }}
+          cache-from: type=registry,ref=${{ steps.meta.outputs.IMAGE }}:buildcache

.gitignore

@@ -13,6 +13,7 @@ venv/
 # Runtime-generated configs and recordings
 configs/
 recordings/
+.drb-update-compose.yml
 
 # Logs
 *.log

README.md

@@ -142,13 +142,17 @@ this repo at the `v1` tag, enrols with C2, pulls the prebuilt images and starts:
 
 ```bash
 curl -fsSL https://git.vpn.cusano.net/logan/node-26/raw/tag/v1/install.sh \
-  | sudo bash -s -- --token DRB-xxxx --node-id node-003 \
+  | sudo bash -s -- --node-id node-003 \
       --c2-url https://api.<domain> --mqtt-broker mqtt.<domain>
 ```
 
-Mint the `--token` at **Settings → Nodes** in the web app (the panel prints the
-whole command). Run `install.sh --help` for every flag; each also has a
-`DRB_*` env var. `--build` compiles op25 on the Pi (~1h) instead of pulling.
+It prompts (hidden input) for an enrollment token, minted at **Settings → Nodes**
+in the web app. The token never goes on the command line — argv lands in shell
+history and `ps` (node-26#18). Unattended installs use `--token-file PATH` (mode
+0600) or an exported `DRB_ENROLLMENT_TOKEN` (pass it through with
+`sudo --preserve-env=DRB_ENROLLMENT_TOKEN`; plain `sudo` drops it); the old `--token` flag still works
+but warns. Run `install.sh --help` for every flag; each also has a `DRB_*` env
+var. `--build` compiles op25 on the Pi (~1h) instead of pulling.
 
 ### Local dev / manual
 

docker-compose.yml

@@ -1,6 +1,10 @@
+# Image tags: a provisioned node pins every image to the source tree it has
+# checked out (DRB_TAG_* = src-<git tree hash>, written into .env by install.sh
+# and scripts/drb-update.sh via scripts/image-tags.sh — node-26#19). Unset, as
+# in local dev, they fall back to :latest / :stable, which CI builds from main.
 services:
   icecast:
-    image: ${IMAGE_REGISTRY:-git.vpn.cusano.net}/${DOCKER_ORG:-logan}/${DOCKER_REPO:-node-26}/icecast:latest
+    image: ${IMAGE_REGISTRY:-git.vpn.cusano.net}/${DOCKER_ORG:-logan}/${DOCKER_REPO:-node-26}/icecast:${DRB_TAG_ICECAST:-latest}
     build: ./icecast
     restart: unless-stopped
     network_mode: host
@@ -16,7 +20,7 @@ services:
   # rather than being exposed. edge-node still reaches both over localhost
   # because it shares this host network namespace.
   op25:
-    image: ${IMAGE_REGISTRY:-git.vpn.cusano.net}/${DOCKER_ORG:-logan}/${DOCKER_REPO:-node-26}/op25-client:stable
+    image: ${IMAGE_REGISTRY:-git.vpn.cusano.net}/${DOCKER_ORG:-logan}/${DOCKER_REPO:-node-26}/op25-client:${DRB_TAG_OP25:-stable}
     build: ./op25-container
     restart: unless-stopped
     privileged: true
@@ -38,7 +42,7 @@ services:
   # node-26#9. Same network/device access as op25 for the same reason: it
   # needs the raw USB device, not a virtualized one.
   secondary-sdr:
-    image: ${IMAGE_REGISTRY:-git.vpn.cusano.net}/${DOCKER_ORG:-logan}/${DOCKER_REPO:-node-26}/secondary-sdr:latest
+    image: ${IMAGE_REGISTRY:-git.vpn.cusano.net}/${DOCKER_ORG:-logan}/${DOCKER_REPO:-node-26}/secondary-sdr:${DRB_TAG_SECONDARY_SDR:-latest}
     build: ./secondary-sdr-container
     restart: unless-stopped
     privileged: true
@@ -48,7 +52,7 @@ services:
       - /dev:/dev
 
   edge-node:
-    image: ${IMAGE_REGISTRY:-git.vpn.cusano.net}/${DOCKER_ORG:-logan}/${DOCKER_REPO:-node-26}/edge-node:latest
+    image: ${IMAGE_REGISTRY:-git.vpn.cusano.net}/${DOCKER_ORG:-logan}/${DOCKER_REPO:-node-26}/edge-node:${DRB_TAG_EDGE_NODE:-latest}
     build: ./drb-edge-node
     restart: unless-stopped
     network_mode: host

drb-edge-node/app/internal/discord_radio.py

@@ -74,6 +74,43 @@ class _Voice:
     code: Optional[str] = None
 
 
+def _join_refusal(guild, channel) -> Optional[tuple[str, str]]:
+    """Why Discord would refuse this bot's voice join, or None.
+
+    Discord does not answer a join the bot lacks permission for — the gateway
+    just never sends the voice-server update, discord.py times out after 30s,
+    and the admin page used to read "(TimeoutError: )". Checking first turns
+    that into an instruction. Best effort: if the bot's own member or the
+    permission data is not cached, let the join go ahead.
+    """
+    me = getattr(guild, "me", None)
+    permissions_for = getattr(channel, "permissions_for", None)
+    if me is None or permissions_for is None:
+        return None
+    try:
+        perms = permissions_for(me)
+    except Exception:
+        return None
+    missing = [label for attr, label in (("view_channel", "View Channel"), ("connect", "Connect"))
+               if not getattr(perms, attr, True)]
+    if getattr(channel, "type", None) == discord.ChannelType.voice and not getattr(perms, "speak", True):
+        missing.append("Speak")
+    if missing:
+        return ("missing_permissions",
+                f"The radio bot lacks {', '.join(missing)} on that voice channel — "
+                "allow it in the channel's permission settings in Discord.")
+    limit = getattr(channel, "user_limit", 0) or 0
+    # The bot's own lingering voice state (a dropped connection) is not a
+    # slot it needs; counting it would block the rejoin of a full channel.
+    me_id = getattr(me, "id", None)
+    members = [m for m in (getattr(channel, "members", None) or [])
+               if me_id is None or getattr(m, "id", None) != me_id]
+    if limit and len(members) >= limit and not getattr(perms, "move_members", False):
+        return ("channel_full", f"That voice channel is full ({len(members)}/{limit}) — "
+                                "raise its user limit or free a slot.")
+    return None
+
+
 def _default_client_factory():
     intents = discord.Intents.none()
     intents.guilds = True
@@ -220,6 +257,10 @@ class RadioManager:
                     v.error = self._session_errors.get(d["token_id"], "The radio bot could not log in to Discord.")
                     continue
                 await self._ensure_voice(sess, v, d)
+            # 4) presence follows voice: "Listening to <system>" only while a
+            #    voice connection for that bot is actually up.
+            for sess in list(self._sessions.values()):
+                await self._sync_presence(sess, target)
 
     async def _ensure_session(self, token_id: str) -> Optional[_Session]:
         sess = self._sessions.get(token_id)
@@ -275,7 +316,12 @@ class RadioManager:
             if not vc.is_playing():
                 self._play(v)
             v.state, v.error, v.code = "connected", None, None
-            await self._set_presence(sess, d.get("system_name"))
+            return
+        refusal = _join_refusal(guild, channel)
+        if refusal:
+            if v.error != refusal[1]:   # once per change, not every watchdog pass
+                logger.error(f"Stream {v.stream_id}: {refusal[1]}")
+            v.state, v.code, v.error = "error", refusal[0], refusal[1]
             return
         v.state = "connecting"
         try:
@@ -287,11 +333,23 @@ class RadioManager:
                 if existing is not None:
                     await existing.disconnect(force=True)
                 v.vc = await channel.connect()
+        except asyncio.TimeoutError:
+            # discord.py's TimeoutError has no message, so the old text read
+            # "(TimeoutError: )" — useless on the admin page.
+            v.vc = None
+            v.state, v.code = "error", "connect_timeout"
+            # Permissions were checked above, so this is usually the voice
+            # handshake itself (e.g. a pre-2.7 discord.py refused with 4017 for
+            # lacking DAVE E2EE, then retried until the 30s timeout) or UDP egress.
+            v.error = ("Discord never completed the voice join (timed out). Permissions look fine, so "
+                       "check the node's discord.py voice log (DAVE close 4017, UDP blocked).")
+            logger.error(f"Stream {v.stream_id}: voice connect timed out")
+            return
         except Exception as e:
             v.vc = None
             v.state, v.code = "error", "connect_failed"
             v.error = f"Could not connect to the voice channel ({type(e).__name__}: {str(e)[:120]})."
-            logger.error(f"Stream {v.stream_id}: voice connect failed: {e}")
+            logger.error(f"Stream {v.stream_id}: voice connect failed: {type(e).__name__}: {e}")
             return
         # Bounded wait for the shared PulseAudio socket (op25 container creates it).
         if not await pulse.wait_until_ready():
@@ -300,15 +358,27 @@ class RadioManager:
             return
         self._play(v, pulse_checked=True)
         v.state, v.error, v.code = "connected", None, None
-        await self._set_presence(sess, d.get("system_name"))
         logger.info(f"Stream {v.stream_id} connected (bot {v.token_id})")
 
-    async def _set_presence(self, sess: _Session, name: Optional[str]) -> None:
-        if not name or sess.presence == name:
+    async def _sync_presence(self, sess: _Session, target: dict[str, dict]) -> None:
+        """Show "Listening to <system>" only while this bot is really in voice.
+
+        It used to be set once on the first successful join and never cleared,
+        so a bot whose voice later dropped (or kept failing to rejoin) still
+        looked healthy in Discord's member list.
+        """
+        name = None
+        for sid, v in sorted(self._voices.items()):
+            if (v.token_id == sess.token_id and v.state == "connected"
+                    and v.vc is not None and v.vc.is_connected()):
+                name = (target.get(sid) or {}).get("system_name") or name
+                if name:
+                    break
+        if sess.presence == name:
             return
         try:
-            await sess.client.change_presence(
-                activity=discord.Activity(type=discord.ActivityType.listening, name=name))
+            activity = discord.Activity(type=discord.ActivityType.listening, name=name) if name else None
+            await sess.client.change_presence(activity=activity)
             sess.presence = name
         except Exception as e:
             logger.warning(f"Presence update failed: {e}")

drb-edge-node/app/internal/key_pickup.py

@@ -50,7 +50,7 @@ async def fetch_api_key() -> Optional[str]:
     if r.status_code == 401:
         logger.error(
             "Key pickup: C2 rejected the saved pickup secret (401) — it was rotated by a re-enroll elsewhere. "
-            "Re-run install.sh with --token to recover."
+            "Re-run install.sh and give it a fresh enrollment token when it asks (node-26#18)."
         )
         return None
     if r.status_code == 404:

drb-edge-node/app/internal/mqtt_manager.py

@@ -1,5 +1,6 @@
 import asyncio
 import json
+import shutil
 from datetime import datetime, timezone
 from collections import deque
 from typing import Optional, Callable, Awaitable, Dict, Any
@@ -8,6 +9,22 @@ from app.config import settings
 from app.internal.logger import logger
 from app.internal import credentials, key_pickup
 
+
+def disk_report(path: str) -> Dict[str, Optional[int]]:
+    """Free/total bytes of the filesystem holding `path` (node-26#41).
+
+    /recordings is a bind mount of the host checkout, so this is the host disk
+    the recorder writes to and Docker's images/build cache live on. A full disk
+    stopped node-002 recording for 17 days while it still showed Online. Nulls
+    on error, so C2 overwrites a stale reading instead of keeping it.
+    """
+    try:
+        usage = shutil.disk_usage(path)
+    except OSError:
+        return {"disk_free_bytes": None, "disk_total_bytes": None}
+    return {"disk_free_bytes": usage.free, "disk_total_bytes": usage.total}
+
+
 CommandCallback = Callable[[Dict[str, Any]], Awaitable[None]]
 ConfigCallback = Callable[[Dict[str, Any]], Awaitable[None]]
 ApiKeyCallback = Callable[[Dict[str, Any]], Awaitable[None]]
@@ -372,6 +389,7 @@ class MQTTManager:
             count = devices.get("count") or None
         if count is not None:
             payload["sdr_count"] = count
+        payload.update(disk_report(settings.recordings_path))
         self._publish(self._t_checkin, payload, qos=1)
 
     def _publish(self, topic: str, payload: dict, qos: int = 0, retain: bool = False):

drb-edge-node/requirements.txt

@@ -3,7 +3,9 @@ uvicorn[standard]
 pydantic-settings
 paho-mqtt>=2.0.0
 httpx
-discord.py[voice]
+# >=2.7: Discord requires DAVE (E2EE) voice since 2026-03-01; 2.7 is the first
+# discord.py with it (the [voice] extra pulls `davey`). Older releases cannot join voice.
+discord.py[voice]>=2.7
 PyNaCl
 python-multipart
 pytest

drb-edge-node/tests/test_discord_radio.py

@@ -47,8 +47,18 @@ class FakeVC:
 class FakeChannel:
     def __init__(self, guild, cid, kind=discord.ChannelType.voice):
         self.guild, self.id, self.type = guild, cid, kind
+        self.perms = discord.Permissions(view_channel=True, connect=True, speak=True)
+        self.user_limit = 0
+        self.members = []
+        self.connect_error = None
+
+    def permissions_for(self, member):
+        assert member is self.guild.me
+        return self.perms
 
     async def connect(self):
+        if self.connect_error:
+            raise self.connect_error
         vc = FakeVC(self)
         self.guild.voice_client = vc
         return vc
@@ -57,6 +67,7 @@ class FakeChannel:
 class FakeGuild:
     def __init__(self, gid, channel_ids):
         self.id = gid
+        self.me = object()
         self.voice_client = None
         self.channels = {c: FakeChannel(self, c) for c in channel_ids}
 
@@ -149,6 +160,7 @@ async def test_apply_connects_and_reports(env):
     assert env.mgr.is_connected
     vc = env.guilds[1].voice_client
     assert vc.plays == 1 and vc.channel.id == 11
+    assert FakeClient.instances[0].presence.name == "County"
 
 
 async def test_one_session_serves_two_guilds(env):
@@ -258,3 +270,64 @@ async def test_no_message_content_intent_on_real_client():
         assert client.intents.voice_states and client.intents.guilds
     finally:
         await client.close()
+
+
+def _presence(client):
+    activity = getattr(client, "presence", None)
+    return activity.name if activity else None
+
+
+async def test_missing_connect_permission_is_named_not_a_timeout(env):
+    """Discord ignores a join the bot may not make; say which permission is missing."""
+    env.guilds[1].channels[11].perms = discord.Permissions(view_channel=True, speak=True)
+    await env.mgr.apply_desired(_desired(1, (1, 11, "t1")))
+    st = env.published[-1]["streams"]["1:11"]
+    assert st["state"] == "error" and st["code"] == "missing_permissions"
+    assert "Connect" in st["error"] and "View Channel" not in st["error"]
+    assert env.guilds[1].voice_client is None
+    assert _presence(FakeClient.instances[0]) is None   # not "Listening to …" while out of voice
+
+
+async def test_full_channel_is_reported(env):
+    ch = env.guilds[1].channels[11]
+    ch.user_limit, ch.members = 2, [object(), object()]
+    await env.mgr.apply_desired(_desired(1, (1, 11, "t1")))
+    st = env.published[-1]["streams"]["1:11"]
+    assert st["code"] == "channel_full" and "2/2" in st["error"]
+
+
+async def test_bots_own_lingering_voice_state_does_not_fill_the_channel(env):
+    """A dropped connection can leave the bot listed in the channel; that is
+    not a slot it needs to rejoin."""
+    guild = env.guilds[1]
+    guild.me = type("Me", (), {"id": 999})()
+    ch = guild.channels[11]
+    ch.user_limit, ch.members = 2, [object(), guild.me]
+    await env.mgr.apply_desired(_desired(1, (1, 11, "t1")))
+    assert env.published[-1]["streams"]["1:11"]["state"] == "connected"
+
+
+async def test_connect_timeout_has_a_readable_error(env):
+    env.guilds[1].channels[11].connect_error = asyncio.TimeoutError()
+    await env.mgr.apply_desired(_desired(1, (1, 11, "t1")))
+    st = env.published[-1]["streams"]["1:11"]
+    assert st["state"] == "error" and st["code"] == "connect_timeout"
+    assert "timed out" in st["error"] and "TimeoutError: )" not in st["error"]
+
+
+async def test_presence_is_cleared_when_voice_is_lost_and_rejoin_fails(env):
+    """The owner saw "Listening to <system>" on a bot that was not in voice:
+    presence was set once on join and never cleared."""
+    await env.mgr.apply_desired(_desired(1, (1, 11, "t1")))
+    client = FakeClient.instances[0]
+    assert _presence(client) == "County"
+    env.guilds[1].voice_client.connected = False
+    env.guilds[1].voice_client = None
+    env.guilds[1].channels[11].connect_error = RuntimeError("voice handshake failed")
+    await env.mgr.tick()
+    st = env.mgr.snapshot()["streams"]["1:11"]
+    assert st["state"] == "error" and st["code"] == "connect_failed"
+    assert _presence(client) is None
+    env.guilds[1].channels[11].connect_error = None
+    await env.mgr.tick()
+    assert _presence(client) == "County"

drb-edge-node/tests/test_host_updater.py

@@ -24,21 +24,37 @@ pytestmark = pytest.mark.skipif(
 )
 
 REPO = "https://git.example.net/logan/node-26.git"
-OLD, NEW = "a" * 40, "b" * 40
+OLD, NEW, TREE = "a" * 40, "b" * 40, "c" * 40
+REPO_ROOT = SCRIPT.parents[1]
 
 STUB_GIT = r'''#!/usr/bin/env bash
 echo "git $*" >> "$STUB/calls"
-while [ "$1" = -c ]; do shift 2; done
+while [ "$1" = -c ] || [ "$1" = -C ]; do shift 2; done
 case "$1" in
-  rev-parse) if [ "$2" = HEAD ]; then cat "$STUB/head"; else echo "$STUB_NEW"; fi ;;
+  rev-parse)
+    if [ "$2" = HEAD ]; then cat "$STUB/head"
+    elif [ "$2" = --verify ]; then echo "$STUB_TREE"   # tree hash of <commit>:<dir>
+    else echo "$STUB_NEW"; fi ;;
   remote) echo "$STUB_ORIGIN" ;;
   fetch) exit 0 ;;
   show-ref) [ "$4" = "refs/tags/v1" ] ;;
+  cat-file) [ -z "${STUB_NO_PIN:-}" ] ;;   # target has scripts/image-tags.sh
+  show) case "$2" in
+          *:scripts/image-tags.sh) cat "$STUB_REPO/scripts/image-tags.sh" ;;
+          *:docker-compose.yml) cat "$STUB_REPO/docker-compose.yml" ;;
+        esac ;;
   checkout) echo "${@: -1}" > "$STUB/head" ;;
   *) exit 0 ;;
 esac
 '''
-STUB_DOCKER = '#!/usr/bin/env bash\necho "docker $*" >> "$STUB/calls"\n'
+STUB_DOCKER = r'''#!/usr/bin/env bash
+echo "docker $*" >> "$STUB/calls"
+case "$*" in
+  *"config --images"*) echo "reg.example/edge-node:$STUB_PULLED_TAG" ;;
+  "image inspect"*) [ -z "${STUB_MISSING_IMAGE:-}" ] || exit 1 ;;
+esac
+exit 0
+'''
 STUB_CURL = r'''#!/usr/bin/env bash
 echo "curl $*" >> "$STUB/calls"
 for a in "$@"; do case "$a" in @*) cat "${a#@}" >> "$STUB/headers" ;; esac; done
@@ -79,6 +95,7 @@ def env(tmp_path):
     e.update({
         "PATH": f"{bindir}:{os.environ['PATH']}",
         "STUB": str(stub), "STUB_NEW": NEW, "STUB_ORIGIN": REPO,
+        "STUB_TREE": TREE, "STUB_REPO": str(REPO_ROOT), "STUB_PULLED_TAG": f"src-{TREE}",
         "DRB_UPDATER_CONF": str(conf),
         "DRB_UPDATER_STATE": str(tmp_path / "state"),
         "DRB_UPDATER_LOCK": str(tmp_path / "lock"),
@@ -94,6 +111,10 @@ def _calls(env):
     return (env["stub"] / "calls").read_text()
 
 
+def _pulled(calls: str) -> bool:
+    return any(line.startswith("docker compose ") and line.endswith(" pull") for line in calls.splitlines())
+
+
 def test_no_request_means_checkin_only(env):
     r = _run(env)
     assert r.returncode == 0, r.stderr
@@ -108,7 +129,7 @@ def test_c2_request_updates_once_and_reports(env):
     r = _run(env)
     assert r.returncode == 0, r.stderr + r.stdout
     calls = _calls(env)
-    assert "docker compose pull" in calls
+    assert _pulled(calls)
     assert "docker compose up -d --no-build --force-recreate --remove-orphans" in calls
     assert (env["stub"] / "head").read_text().strip() == NEW
     assert '"applied_request_at":"2026-10-04T12:00:00+00:00"' in calls
@@ -126,10 +147,21 @@ def test_mqtt_sentinel_triggers_update_and_is_consumed(env):
     assert r.returncode == 0, r.stderr
     assert not sentinel.exists()
     calls = _calls(env)
-    assert "docker compose pull" in calls
+    assert _pulled(calls)
     assert "evil" not in calls and "rm -rf" not in calls
 
 
+def test_update_prunes_build_cache_and_stale_images_after_up(env):
+    # node-26#41: the build cache was never pruned and filled node-002's disk.
+    (env["stub"] / "response").write_text(json.dumps({"update_requested_at": "2026-10-04T12:00:00+00:00"}))
+    assert _run(env).returncode == 0
+    lines = _calls(env).splitlines()
+    up = next(i for i, ln in enumerate(lines) if ln.startswith("docker compose up -d"))
+    builder = next(i for i, ln in enumerate(lines) if ln.startswith("docker builder prune -af --filter until="))
+    images = next(i for i, ln in enumerate(lines) if ln.startswith("docker image prune -af --filter until="))
+    assert up < builder and up < images  # only after the new containers are running
+
+
 def test_c2_response_is_never_used_as_a_ref_or_command(env):
     (env["stub"] / "response").write_text(json.dumps({"update_requested_at": "$(touch /tmp/pwned); main"}))
     r = _run(env)
@@ -153,7 +185,7 @@ def test_refuses_when_env_registry_differs(env):
     (env["inst"] / "configs" / "update-requested").write_text("")
     r = _run(env)
     assert r.returncode != 0
-    assert "docker compose pull" not in _calls(env)
+    assert not _pulled(_calls(env))
 
 
 def test_non_https_repo_in_conf_is_rejected(env):
@@ -177,3 +209,42 @@ def test_plain_http_c2_skips_checkin(env):
     r = _run(env)
     assert r.returncode == 0
     assert "curl" not in _calls(env)
+
+
+# ── node-26#19: images pinned to the target's source tree ─────────────────
+
+def test_pins_images_and_pulls_before_checkout(env):
+    (env["inst"] / "configs" / "update-requested").write_text("")
+    r = _run(env)
+    assert r.returncode == 0, r.stderr + r.stdout
+    lines = _calls(env).splitlines()
+    pull = next(i for i, ln in enumerate(lines) if ln.startswith("docker compose ") and ln.endswith(" pull"))
+    checkout = next(i for i, ln in enumerate(lines) if ln.startswith("git") and " checkout " in ln)
+    assert pull < checkout, "images must be pulled before the new code is checked out"
+    dotenv = (env["inst"] / ".env").read_text()
+    for var in ("DRB_TAG_EDGE_NODE", "DRB_TAG_OP25", "DRB_TAG_ICECAST", "DRB_TAG_SECONDARY_SDR"):
+        assert f"{var}=src-{TREE}\n" in dotenv
+    assert dotenv.startswith("NODE_ID=node-7\n")  # existing lines kept
+    assert not (env["inst"] / ".drb-update-compose.yml").exists()
+
+
+def test_missing_pinned_image_changes_nothing(env):
+    env["env"]["STUB_MISSING_IMAGE"] = "1"
+    (env["inst"] / "configs" / "update-requested").write_text("")
+    r = _run(env)
+    assert r.returncode != 0
+    assert (env["stub"] / "head").read_text().strip() == OLD
+    calls = _calls(env)
+    assert " checkout " not in calls
+    assert "up -d" not in calls
+    assert "DRB_TAG_" not in (env["inst"] / ".env").read_text()
+
+
+def test_target_without_pinning_still_updates(env):
+    env["env"]["STUB_NO_PIN"] = "1"
+    (env["inst"] / "configs" / "update-requested").write_text("")
+    r = _run(env)
+    assert r.returncode == 0, r.stderr + r.stdout
+    assert (env["stub"] / "head").read_text().strip() == NEW
+    assert _pulled(_calls(env))
+    assert "DRB_TAG_" not in (env["inst"] / ".env").read_text()

drb-edge-node/tests/test_mqtt_manager.py

@@ -140,3 +140,16 @@ def test_build_client_does_not_touch_offline_buffer(monkeypatch):
 
     assert list(mqtt_manager._offline_buffer) == [("nodes/test/metadata", {"call_id": "sentinel"})]
     mqtt_manager._offline_buffer.clear()
+
+
+def test_disk_report_reads_free_and_total(tmp_path):
+    from app.internal.mqtt_manager import disk_report
+    rep = disk_report(str(tmp_path))
+    assert rep["disk_total_bytes"] > 0
+    assert 0 <= rep["disk_free_bytes"] <= rep["disk_total_bytes"]
+
+
+def test_disk_report_nulls_when_path_unreadable(tmp_path):
+    # node-26#41: nulls (not omitted keys) so C2 never keeps a stale reading.
+    from app.internal.mqtt_manager import disk_report
+    assert disk_report(str(tmp_path / "missing")) == {"disk_free_bytes": None, "disk_total_bytes": None}

install.sh

@@ -1,16 +1,21 @@
 #!/usr/bin/env bash
 #
-# DRB edge node — one-shot bootstrap for a clean Raspberry Pi OS (arm64).
+# DRB edge node — one-shot bootstrap for a clean Raspberry Pi OS (arm64) or Debian (amd64).
 #
 #   curl -fsSL https://git.vpn.cusano.net/logan/node-26/raw/tag/v1/install.sh \
-#     | sudo bash -s -- --token DRB-xxxx --node-id node-003 \
+#     | sudo bash -s -- --node-id node-003 \
 #         --c2-url https://api.<domain> --mqtt-broker mqtt.<domain>
 #
+# It prompts (hidden input) for the enrollment token. The token is never
+# taken from the command line by design (node-26#18, CISO condition from
+# server-26#61): argv lands in shell history and in `ps`. Unattended installs
+# use --token-file PATH (a 0600 file) or an exported DRB_ENROLLMENT_TOKEN.
+#
 # (fetch install.sh from the same tag it installs; use raw/branch/main with
 #  --track-main for the owner's own always-latest nodes)
 #
 # Hosting, the pinned ref, and setup.sh's retirement are settled (owner
-# decisions, 2026-09-06). One standing hazard remains — D3 below.
+# decisions, 2026-09-06). D3 below is mitigated by image pinning (node-26#19).
 #
 # What it does, in order:
 #   1. Preflight (root, arch, apt, SDR present)
@@ -48,15 +53,18 @@
 #       the mint panel / node-26#4). DEFAULT_REF below is `v1`. `--track-main`
 #       stays as an opt-in for the owner's own nodes.
 #
-# STANDING HAZARD
+# MITIGATED HAZARD
 #
 #   D3  SOURCE-OVERLAY. docker-compose.yml bind-mounts ./drb-edge-node/app and
-#       ./op25-container/app OVER the image's /app. So even in prebuilt mode
-#       the running Python is the CLONED REF's code against the pulled image's
-#       dependencies. `v1` == the commit CI built the current :latest/:stable
-#       from, so today they match — but the moment `v1` and the image tags
-#       diverge this silently mixes them. Fix: drop those two mounts from the
-#       prod compose path, or always retag images from the same ref as `v1`.
+#       ./op25-container/app OVER the image's /app, so the running Python is
+#       the CLONED REF's code against the pulled image's dependencies. Those
+#       used to be :latest/:stable, built from main — silently a different
+#       commit from `v1`. Since node-26#19 every image is pinned to
+#       src-<git tree hash of its own directory at the checked-out ref>
+#       (scripts/image-tags.sh, written into .env as DRB_TAG_*), which CI tags
+#       identically. The mounts now overlay the same code the image was built
+#       from. A pinned image that was never built fails the pull loudly; the
+#       "Release images" workflow builds any a release tag is missing.
 # ---------------------------------------------------------------------------
 set -euo pipefail
 
@@ -78,6 +86,7 @@ MQTT_BROKER="${DRB_MQTT_BROKER:-}"
 MQTT_PORT="${DRB_MQTT_PORT:-8883}"
 MQTT_TLS="${DRB_MQTT_TLS:-true}"
 ENROLLMENT_TOKEN="${DRB_ENROLLMENT_TOKEN:-}"
+unset DRB_ENROLLMENT_TOKEN   # read once; don't leak it into every child's env
 DASHBOARD_USER="${DRB_DASHBOARD_USERNAME:-admin}"
 DASHBOARD_PASS="${DRB_DASHBOARD_PASSWORD:-}"
 REGISTRY="${DRB_IMAGE_REGISTRY:-git.vpn.cusano.net}"   # D1 (public host)
@@ -108,7 +117,10 @@ usage() {
   cat <<'USAGE'
 Usage: install.sh [options]
 
-  --token TOKEN        Enrollment token (Settings -> Nodes -> New token)
+  (enrollment token)   Prompted for on the terminal, input hidden. Mint one at
+                       Settings -> Nodes -> New token.
+  --token-file PATH    read the enrollment token from PATH (unattended installs;
+                       keep it mode 0600 and delete it afterwards)
   --node-id ID         Unique node id, e.g. node-003
   --name NAME          Display name (default: node id)
   --lat N --lon N      Decimal degrees for the map
@@ -132,13 +144,19 @@ Usage: install.sh [options]
 
 Every option also has an env var: DRB_NODE_ID, DRB_C2_URL, DRB_ENROLLMENT_TOKEN,
 DRB_MQTT_BROKER, DRB_REF, DRB_INSTALL_DIR, DRB_REGISTRY_USER/PASS, ...
-Secrets are read from the environment or prompted on the TTY, never from a pipe.
+Secrets are read from a file, the environment or a TTY prompt, never from a
+pipe. Never put the token on a command line: `--token X` and
+`sudo DRB_ENROLLMENT_TOKEN=X bash` both expose it in shell history and `ps`.
+An exported DRB_ENROLLMENT_TOKEN only reaches the installer through sudo with
+`sudo --preserve-env=DRB_ENROLLMENT_TOKEN bash` (plain sudo drops it).
+(`--token` still works for old copied commands but is deprecated, see node-26#18.)
 USAGE
 }
 
 while [ $# -gt 0 ]; do
   case "$1" in
-    --token)          ENROLLMENT_TOKEN="$2"; shift 2 ;;
+    --token)          ENROLLMENT_TOKEN="$2"; TOKEN_ON_ARGV=1; shift 2 ;;
+    --token-file)     TOKEN_FILE="$2"; shift 2 ;;
     --node-id)        NODE_ID="$2"; shift 2 ;;
     --name)           NODE_NAME="$2"; shift 2 ;;
     --lat)            NODE_LAT="$2"; shift 2 ;;
@@ -162,6 +180,18 @@ while [ $# -gt 0 ]; do
   esac
 done
 
+if [ "${TOKEN_ON_ARGV:-0}" = 1 ]; then
+  warn "--token is deprecated: the token is now in your shell history and visible in"
+  warn "\`ps\` for this whole install (node-26#18). Omit it to be prompted, or use --token-file."
+fi
+if [ -n "${TOKEN_FILE:-}" ]; then
+  [ -r "$TOKEN_FILE" ] || die "--token-file: cannot read $TOKEN_FILE"
+  _mode="$(stat -c %a "$TOKEN_FILE" 2>/dev/null || echo 600)"
+  (( 8#$_mode & 8#044 )) && warn "--token-file $TOKEN_FILE is readable by other users — chmod 600 it"
+  ENROLLMENT_TOKEN="$(tr -d '[:space:]' < "$TOKEN_FILE")"
+  [ -n "$ENROLLMENT_TOKEN" ] || die "--token-file: $TOKEN_FILE is empty"
+fi
+
 [ "$ENROLL_WAIT" = forever ] || [[ "$ENROLL_WAIT" =~ ^[0-9]+$ ]] \
   || die "--wait-approval takes a number of seconds (or omit it to wait until approved)"
 
@@ -181,6 +211,14 @@ ask_secret() {
   read -rsp "$__p: " __r </dev/tty; echo >/dev/tty
   printf -v "$__v" '%s' "$__r"
 }
+# Make sure ENROLLMENT_TOKEN is set: already given, else prompt on the TTY.
+# Returns 1 if there is still no token (non-interactive, or Enter at the prompt).
+need_token() {
+  [ -n "$ENROLLMENT_TOKEN" ] || ask_secret ENROLLMENT_TOKEN "Enrollment token (input hidden)"
+  [ -n "$ENROLLMENT_TOKEN" ]
+}
+NO_TOKEN_HELP="mint one at Settings -> Nodes and re-run the install command; it prompts for
+  the token (unattended: --token-file PATH, or an exported DRB_ENROLLMENT_TOKEN)"
 genpw() { head -c 24 /dev/urandom | base64 | tr -d '\n=' ; }
 
 # ── 1. Preflight ────────────────────────────────────────────────────────────
@@ -190,8 +228,8 @@ command -v apt-get >/dev/null || die "apt-get not found — this script targets
 
 ARCH="$(dpkg --print-architecture)"
 case "$ARCH" in
-  arm64|aarch64) ok "arch $ARCH" ;;
-  *) warn "arch is $ARCH — CI only builds linux/arm64 images. Prebuilt pull will fail; use --build." ;;
+  arm64|aarch64|amd64|x86_64) ok "arch $ARCH" ;;
+  *) warn "arch is $ARCH — CI only builds linux/amd64 + linux/arm64 images. Prebuilt pull will fail; use --build." ;;
 esac
 ok "running as root"
 
@@ -334,6 +372,20 @@ EOF
   ok ".env written for '$NODE_ID'"
 fi
 
+# Pin every image to the source just checked out (node-26#19, D3). Written
+# into .env — fresh or preserved — so later manual `docker compose` runs on
+# this node use the same pins.
+# A --ref older than #19 has no helper; its compose file hardcodes its tags.
+if [ -f "$INSTALL_DIR/scripts/image-tags.sh" ]; then
+  # shellcheck source=scripts/image-tags.sh
+  . "$INSTALL_DIR/scripts/image-tags.sh"
+  drb_pin_image_tags "$INSTALL_DIR/.env" "$INSTALL_DIR" "$RESOLVED_SHA" || die "could not write image pins to .env"
+  ok "images pinned to the source at ${RESOLVED_SHA:0:12} (DRB_TAG_* in .env)"
+else
+  warn "$REF predates image pinning (node-26#19) — using its compose file's default image tags"
+  drb_images_present() { return 0; }
+fi
+
 # ── 5. Enrollment ───────────────────────────────────────────────────────────
 # Client half of Server/drb-c2-core/app/routers/enrollment.py. It does NOT
 # exist in the edge-node app today (mqtt_manager.py:73-85 says so explicitly),
@@ -416,8 +468,8 @@ wait_for_key() {  # wait_for_key PICKUP_SECRET
   while :; do
     case "$CRED_HTTP" in
       200) [ -n "$(cred_field '.api_key')" ] && { trap - INT TERM; return 0; } ;;
-      401) die "C2 rejected the saved pickup secret (401) while waiting — re-run with --token DRB-… to re-enroll." ;;
-      404) die "C2 no longer knows node '$NODE_ID' (404) — it was deleted. Re-run with --token DRB-… to enroll again." ;;
+      401) die "C2 rejected the saved pickup secret (401) while waiting — re-run the installer to re-enroll (it asks for a fresh enrollment token)." ;;
+      404) die "C2 no longer knows node '$NODE_ID' (404) — it was deleted. Re-run the installer to enroll again (it asks for an enrollment token)." ;;
       *)   : ;;  # 000/5xx: transient, keep waiting
     esac
     if [ "$_shown" = 0 ]; then
@@ -435,8 +487,7 @@ wait_for_key() {  # wait_for_key PICKUP_SECRET
 
 do_fresh_enroll() {
   say "Enrolling '$NODE_ID' with $C2_URL"
-  [ -n "$ENROLLMENT_TOKEN" ] || ask_secret ENROLLMENT_TOKEN "Enrollment token"
-  [ -n "$ENROLLMENT_TOKEN" ] || die "no enrollment token — mint one at Settings -> Nodes, then re-run with --token"
+  need_token || die "no enrollment token — $NO_TOKEN_HELP"
 
   local BODY RESP PICKUP STATUS KEY
   BODY="$(jq -nc --arg id "$NODE_ID" --arg n "${NODE_NAME:-$NODE_ID}" \
@@ -459,7 +510,7 @@ do_fresh_enroll() {
   or have an admin Reissue key (Settings -> Nodes) and write the key into
   $CREDS by hand (see node-26#6)." ;;
       *401*) die "enroll refused (401): enrollment token missing/invalid/revoked —
-  mint a fresh one at Settings -> Nodes and re-run with --token." ;;
+  $NO_TOKEN_HELP." ;;
       *429*) die "enroll refused (429): rate limited. Wait ~1 minute, then re-run." ;;
       *)     die "enroll failed: $RESP" ;;
     esac
@@ -522,24 +573,26 @@ elif [ -s "$PICKUP_FILE" ]; then
       fi
       ;;
     401)
-      if [ -n "$ENROLLMENT_TOKEN" ]; then
-        warn "saved pickup secret rejected (401) — likely rotated by a re-enroll elsewhere. Re-enrolling with --token."
+      warn "saved pickup secret rejected (401) — likely rotated by a re-enroll elsewhere."
+      if need_token; then
+        warn "re-enrolling with the enrollment token."
         rm -f "$PICKUP_FILE"
         do_fresh_enroll
       else
-        die "saved pickup secret is stale (401) and no --token was given. Re-run with
-  --token DRB-… to re-enroll (only works while the node is still pending), or
-  have an admin Reissue key for an approved node and write $CREDS by hand."
+        die "saved pickup secret is stale (401) and no enrollment token was given.
+  Re-enrolling only works while the node is still pending: $NO_TOKEN_HELP.
+  For an approved node, have an admin Reissue key and write $CREDS by hand."
       fi
       ;;
     404)
-      if [ -n "$ENROLLMENT_TOKEN" ]; then
-        warn "C2 does not know node '$NODE_ID' (404) — deleted server-side or never fully enrolled. Re-enrolling with --token."
+      warn "C2 does not know node '$NODE_ID' (404) — deleted server-side or never fully enrolled."
+      if need_token; then
+        warn "re-enrolling with the enrollment token."
         rm -f "$PICKUP_FILE"
         do_fresh_enroll
       else
-        die "C2 does not know node '$NODE_ID' (404) and no --token was given.
-  Re-run with --token DRB-… to enroll it again."
+        die "C2 does not know node '$NODE_ID' (404) and no enrollment token was given:
+  $NO_TOKEN_HELP."
       fi
       ;;
     000)
@@ -570,6 +623,11 @@ if [ "$DO_START" = 1 ]; then
   requirement or a transient network error: re-run with DRB_REGISTRY_USER /
   DRB_REGISTRY_PASS set, or with --build to compile on the Pi (~1h for op25)."
     fi
+    # compose pull only WARNS for a missing image that also has a build:
+    # section, so check the pinned images really arrived.
+    drb_images_present || die "the images pinned to $REF (${RESOLVED_SHA:0:12}) are not in
+  $REGISTRY yet. CI may still be building them — wait and re-run; the owner can
+  run the 'Release images' workflow on that ref; or re-run with --build."
     docker compose up --no-build -d
   fi
   # ALWAYS recreate edge-node (node-26#14): `up -d` leaves an already-running
@@ -618,6 +676,6 @@ if [ -z "$(jq -r '.api_key // empty' "$CREDS" 2>/dev/null)" ]; then
   if [ -s "$PICKUP_FILE" ]; then
     printf "${Y}This node has no api_key yet. Once an admin approves it at %s/settings/nodes\nthe node fetches the key and comes online by itself (no re-run needed).${N}\n\n" "$APP_URL"
   else
-    printf "${Y}This node has no api_key and no saved pickup secret, so a plain re-run cannot\nfix it. Re-run with --token DRB-… to enroll; or, if the node is already\napproved, have an admin Reissue key and write it into\n%s by hand.${N}\n\n" "$CREDS"
+    printf "${Y}This node has no api_key and no saved pickup secret. Re-run the installer\nand give it an enrollment token when it asks; or, if the node is already\napproved, have an admin Reissue key and write it into\n%s by hand.${N}\n\n" "$CREDS"
   fi
 fi

scripts/drb-update.sh

@@ -17,7 +17,9 @@
 #   * Node credentials go to curl via a 0600 header file, never argv.
 #   * C2 must be https.
 # What this does NOT defend against: a compromised git host or registry —
-# refs and images are not signed (see the issue filed with node-26#15).
+# refs and images are not signed (node-26#17).
+#   * Images are pinned to the checked-out source (node-26#19): each one is
+#     pulled as src-<git tree hash of its build directory>, never :latest.
 #
 # The whole body is inside main(), called on the last line: bash parses the
 # function completely before running it, so `git checkout` replacing this
@@ -49,6 +51,12 @@ main() {
   flock -n 9 || { log "another update is already running"; return 0; }
 
   cd "$INSTALL_DIR"
+  # This runs as root, but a pre-install.sh checkout (e.g. /home/pi/node-26)
+  # is owned by the login user, and git >= 2.35.2 then refuses every command
+  # with "dubious ownership" — which surfaced as origin '' != REPO_URL. Trust
+  # exactly the configured INSTALL_DIR, for this process only (env config is
+  # command-line scope, which safe.directory honours; no global config change).
+  export GIT_CONFIG_COUNT=1 GIT_CONFIG_KEY_0=safe.directory GIT_CONFIG_VALUE_0="$INSTALL_DIR"
   envget() { grep -E "^$1=" .env 2>/dev/null | head -1 | cut -d= -f2- | tr -d '"' || true; }
   local NODE_ID C2_URL
   NODE_ID="$(envget NODE_ID)"; C2_URL="$(envget C2_URL)"; C2_URL="${C2_URL%/}"
@@ -146,6 +154,47 @@ do_update() {
   fi
   local old new
   old="$(git rev-parse HEAD)"; new="$(git rev-parse "$target^{commit}")"
+
+  local env_registry
+  env_registry="$(grep -E '^IMAGE_REGISTRY=' .env | head -1 | cut -d= -f2- | tr -d '"' || true)"
+  [ "${env_registry:-git.vpn.cusano.net}" = "$REGISTRY" ] \
+    || { err ".env IMAGE_REGISTRY '$env_registry' != configured '$REGISTRY' — refusing"; return 1; }
+
+  # Image pins for the TARGET commit (node-26#19): each image is tagged with
+  # the git tree hash of its own source directory, so the images always match
+  # the code being checked out. The helper is read from the target commit so
+  # its service list matches the target's compose file.
+  # Read into a variable, not `source <(...)`: a failed `git show` must leave
+  # pin=0 rather than "succeed" with no functions defined.
+  local pin=0 helper
+  if git cat-file -e "$new:scripts/image-tags.sh" 2>/dev/null; then
+    helper="$(git show "$new:scripts/image-tags.sh")" && [ -n "$helper" ] && eval "$helper" \
+      && declare -F drb_image_tags drb_images_present drb_pin_image_tags >/dev/null && pin=1
+  fi
+  [ "$pin" = 1 ] || log "target ${new:0:12} predates image pinning — using its compose file's default tags"
+
+  if [ "$BUILD" != 1 ]; then
+    # Pull BEFORE checkout. The app code is bind-mounted from the checkout
+    # (install.sh D3) and edge-node runs uvicorn --reload, so checking out first
+    # would run the new code on the old images for as long as the pull takes —
+    # or for good, if the pinned images don't exist. Pulling against the
+    # target's own compose file means a failed pull changes nothing. It is
+    # written next to the real one so its relative paths (.env, env_file,
+    # build contexts) resolve exactly the same way.
+    local tfile="$INSTALL_DIR/.drb-update-compose.yml" pulled=0
+    git show "$new:docker-compose.yml" > "$tfile" || { rm -f "$tfile"; return 1; }
+    local -a tc=(-f "$tfile" --project-directory "$INSTALL_DIR")
+    (
+      if [ "$pin" = 1 ]; then
+        while IFS= read -r l; do export "$l"; done < <(drb_image_tags "$INSTALL_DIR" "$new")
+      fi
+      docker compose "${tc[@]}" pull || exit 1
+      if [ "$pin" = 1 ]; then drb_images_present "${tc[@]}" || exit 1; fi
+    ) && pulled=1
+    rm -f "$tfile"
+    [ "$pulled" = 1 ] || { err "images for ${new:0:12} are not available — nothing changed (CI may still be building them; the 'Release images' workflow publishes any that are missing)"; return 1; }
+  fi
+
   if [ "$old" != "$new" ]; then
     git -c advice.detachedHead=false checkout --force --detach "$new" || return 1
     log "checked out $REF: ${old:0:12} -> ${new:0:12}"
@@ -153,20 +202,37 @@ do_update() {
     bash scripts/install-updater.sh --refresh || log "unit refresh failed (continuing with the old units)"
   fi
 
-  local env_registry
-  env_registry="$(grep -E '^IMAGE_REGISTRY=' .env | head -1 | cut -d= -f2- | tr -d '"' || true)"
-  [ "${env_registry:-git.vpn.cusano.net}" = "$REGISTRY" ] \
-    || { err ".env IMAGE_REGISTRY '$env_registry' != configured '$REGISTRY' — refusing"; return 1; }
+  if [ "$pin" = 1 ]; then
+    drb_pin_image_tags "$INSTALL_DIR/.env" "$INSTALL_DIR" "$new" || { err "could not write image pins to .env"; return 1; }
+  fi
 
   if [ "$BUILD" = 1 ]; then
-    docker compose build || return 1
-  else
-    docker compose pull || return 1
+    docker compose build || return 1   # tags the local builds with the same pins
   fi
   # --force-recreate: app code is bind-mounted from the checkout (install.sh
   # D3), so a code-only change also needs every container restarted.
   docker compose up -d --no-build --force-recreate --remove-orphans || return 1
+
+  # Drop pinned images this node no longer runs. `image prune` only removes
+  # dangling images, and every previous src- tag is still tagged. `image rm`
+  # without -f refuses anything a container still uses.
+  if [ "$pin" = 1 ]; then
+    local keep ref
+    keep="$(drb_image_tags "$INSTALL_DIR" "$new" | cut -d= -f2)"
+    docker images --format '{{.Repository}}:{{.Tag}}' | grep -E ':src-[0-9a-f]{40}$' \
+      | while IFS= read -r ref; do
+          grep -qxF "${ref##*:}" <<< "$keep" || docker image rm "$ref" >/dev/null 2>&1 || true
+        done || true
+  fi
   docker image prune -f >/dev/null 2>&1 || true
+
+  # node-26#41: neither of the above ever freed BuildKit cache or untagged-
+  # but-unused images, so a BUILD=1 node grew ~40 GB of build cache plus
+  # 16 GB of stale images until the root disk hit 100% and recording silently
+  # stopped (node-002, 17 days). Both prunes skip anything a container uses;
+  # the age filters keep the cache warm for the next build.
+  docker builder prune -af --filter until=72h >/dev/null 2>&1 || true
+  docker image prune -af --filter until=168h >/dev/null 2>&1 || true
 }
 
 main "$@"

scripts/image-tags.sh

@@ -0,0 +1,65 @@
+# shellcheck shell=bash
+#
+# Pin each node image to the exact source it was built from (node-26#19).
+# Sourced by install.sh and scripts/drb-update.sh; not run directly.
+#
+# The tag is the git TREE hash of the image's build context, `src-<tree>`.
+# CI computes the same value (`git rev-parse HEAD:<dir>` in the build
+# workflows), so a node asks for the image built from the very directory it
+# has checked out. That is exact even though each image is only rebuilt when
+# its own directory changes: an unchanged directory keeps the same tree hash,
+# and so the same image, across commits. :latest / :stable still exist for
+# local dev; a provisioned node never uses them.
+#
+# docker-compose.yml reads these as ${DRB_TAG_*:-latest}. The values live in
+# .env so every manual `docker compose ...` on the node sees the same pins.
+
+# VAR:build-context-dir, one per compose service with an image.
+DRB_IMAGE_SOURCES="DRB_TAG_ICECAST:icecast DRB_TAG_OP25:op25-container \
+DRB_TAG_SECONDARY_SDR:secondary-sdr-container DRB_TAG_EDGE_NODE:drb-edge-node"
+
+# drb_image_tags REPO_DIR COMMIT -> "VAR=src-<tree>" lines. A directory that
+# does not exist at COMMIT is skipped (its compose file won't reference it).
+drb_image_tags() {
+  local _p _var _dir _tree
+  for _p in $DRB_IMAGE_SOURCES; do
+    _var="${_p%%:*}"; _dir="${_p#*:}"
+    _tree="$(git -C "$1" rev-parse --verify -q "$2:$_dir" 2>/dev/null)" || continue
+    printf '%s=src-%s\n' "$_var" "$_tree"
+  done
+}
+
+# drb_env_set ENV_FILE KEY VALUE — replace KEY's line in place, or append it.
+# Rewrites through the same inode so .env keeps its 0600 mode and owner.
+drb_env_set() {
+  local _tmp; _tmp="$(mktemp)"
+  awk -v k="$2" -v v="$3" '
+    index($0, k "=") == 1 { if (!done) print k "=" v; done = 1; next }
+    { print }
+    END { if (!done) print k "=" v }' "$1" > "$_tmp" && cat "$_tmp" > "$1"
+  local _rc=$?; rm -f "$_tmp"; return $_rc
+}
+
+# drb_pin_image_tags ENV_FILE REPO_DIR COMMIT — write the pins into .env and
+# export them, so compose sees them whether or not it re-reads .env.
+drb_pin_image_tags() {
+  local _line
+  while IFS= read -r _line; do
+    drb_env_set "$1" "${_line%%=*}" "${_line#*=}" || return 1
+    export "$_line"
+  done < <(drb_image_tags "$2" "$3")
+}
+
+# drb_images_present COMPOSE_ARGS... — 0 if every image the compose project
+# names is in the local image store. `compose pull` only WARNS when a service
+# that also has a build: section fails to pull, so its exit code alone does
+# not prove the pinned images exist.
+drb_images_present() {
+  local _imgs _img _missing=0
+  _imgs="$(docker compose "$@" config --images)" || return 1
+  while IFS= read -r _img; do
+    [ -n "$_img" ] || continue
+    docker image inspect "$_img" >/dev/null 2>&1 || { printf 'missing image: %s\n' "$_img" >&2; _missing=1; }
+  done <<< "$_imgs"
+  return $_missing
+}