@askexenow/exe-os 0.9.318 → 0.9.322

package/deploy/compose/docker-compose.yml

@@ -507,7 +507,7 @@ services:
       # offline boots fail). WIKI_EXE_OS_NPM_VERSION defaults to the stack's
       # pinned exe-os npm version and is threaded into the spawn args.
       EXE_OS_MCP_COMMAND: ${WIKI_EXE_OS_MCP_COMMAND:-npx}
-      EXE_OS_MCP_ARGS: ${WIKI_EXE_OS_MCP_ARGS:--y @askexenow/exe-os@${WIKI_EXE_OS_NPM_VERSION:-0.9.312} mcp}
+      EXE_OS_MCP_ARGS: ${WIKI_EXE_OS_MCP_ARGS:--y @askexenow/exe-os@${WIKI_EXE_OS_NPM_VERSION:-0.9.318} mcp}
       BRANDING_CONFIG_PATH: /app/branding.json
     ports:
       - "127.0.0.1:${WIKI_HOST_PORT:-3001}:3001"
@@ -1330,7 +1330,9 @@ services:
   # http://exe-otel-collector:4318 (HTTP) on the backend network.
 
   exe-otel-collector:
-    image: ${OTEL_COLLECTOR_IMAGE:-otel/opentelemetry-collector-contrib:0.114.0}
+    # bug 32d6e02f: digest-pinned default (was tag-only) — matches the
+    # stack manifest pin for otel/opentelemetry-collector-contrib:0.114.0.
+    image: ${OTEL_COLLECTOR_IMAGE:-otel/opentelemetry-collector-contrib:0.114.0@sha256:37fa87091cfaaec7234a27e4e395a40c31c2bfaea97a349a4afef6d9e9681197}
     container_name: exe-otel-collector
     restart: unless-stopped
     command: ["--config=/etc/otelcol-contrib/config.yaml"]
@@ -1343,23 +1345,33 @@ services:
       - "127.0.0.1:${OTEL_HEALTH_PORT:-13133}:13133" # Health check
     networks:
       - backend
-    # Bug 2e8ef511: otel-contrib is a distroless image — only the /otelcol-contrib
-    # binary is present (no shell/wget/curl/nc, and the binary has no probe
-    # subcommand), so a Docker `healthcheck.test` CMD that runs INSIDE the
-    # container cannot be implemented for this upstream, pull-only image. The old
-    # `test: ["NONE"]` actively DISABLED health, so a broken collector still
-    # reported "running" — a silent observability gap.
+    # Bug 2e8ef511 / 32d6e02f: otel-contrib is a distroless image — only the
+    # /otelcol-contrib binary is present (no shell/wget/curl/nc), so the usual
+    # `curl -f .../health` in-container probe is impossible. The binary DOES,
+    # however, ship the offline `validate` subcommand (since collector v0.99,
+    # present in 0.114.0), which exec-runs with no shell and exits 0 only when
+    # the binary is runnable AND the mounted config parses. We use that as the
+    # in-container Docker healthcheck below: it catches a missing binary, a
+    # corrupt image layer, or a bad config mount (the common failure modes for a
+    # pull-only image) without needing any HTTP client.
     #
-    # The collector's health_check extension serves HTTP 200 on :13133, and that
-    # port is published to 127.0.0.1 above. Health is therefore verified
-    # EXTERNALLY: status.sh now probes http://127.0.0.1:${OTEL_HEALTH_PORT}/
-    # alongside every other service endpoint, so a down/unhealthy collector is
-    # surfaced by the stack health tooling instead of hiding behind "running".
+    # `validate` is a config/binary sanity check, NOT a full runtime-liveness
+    # probe (a hung collector with a valid config could still pass it). TRUE
+    # runtime liveness — that receivers are actually listening — is verified
+    # EXTERNALLY: the collector's health_check extension serves HTTP 200 on
+    # :13133 (published to 127.0.0.1 above) and status.sh probes
+    # http://127.0.0.1:${OTEL_HEALTH_PORT}/ alongside every other service.
     # Keep that probe in lockstep with the OTEL_HEALTH_PORT mapping above.
     #
-    # NOTE: we intentionally do NOT set `test: ["NONE"]`. Leaving healthcheck
-    # unset inherits the upstream image's own HEALTHCHECK if one is ever added,
-    # whereas ["NONE"] would permanently suppress it.
+    # Nothing `depends_on` this service with `condition: service_healthy`, and
+    # `restart: unless-stopped` does not restart on `unhealthy`, so the
+    # healthcheck only surfaces status — it can never block or crash-loop the stack.
+    healthcheck:
+      test: ["CMD", "/otelcol-contrib", "validate", "--config=/etc/otelcol-contrib/config.yaml"]
+      interval: 30s
+      timeout: 10s
+      retries: 3
+      start_period: 10s
     deploy:
       resources:
         limits:

package/deploy/stack-manifests/signing-key.pub

@@ -0,0 +1,3 @@
+-----BEGIN PUBLIC KEY-----
+MCowBQYDK2VwAyEA9jWntop9h4lBttod7yfeFcXQXKqqaGOOjvaKswNJdqs=
+-----END PUBLIC KEY-----