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>

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/tests/test_host_updater.py

@@ -151,6 +151,17 @@ def test_mqtt_sentinel_triggers_update_and_is_consumed(env):
     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)

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}

scripts/drb-update.sh

@@ -51,6 +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%/}"
@@ -219,6 +225,14 @@ do_update() {
         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 "$@"