Merge pull request 'Replace k3s with direct soul-demo watchdog' (#57) from fix/checkout-auth-reveal into dev

Merge fix/checkout-auth-reveal into dev

.gitea/workflows/dev.yaml

@@ -172,11 +172,12 @@ jobs:
       - name: Touch HTML placeholder files
         run: touch src/index.html src/about.html src/terms.html src/enterprise-terms.html
 
-      - name: Create soul-demo-image.tar placeholder
+      - name: Create soul-demo placeholder
-        # Dockerfile.stage COPYs this file (used by k3s at runtime).
+        # Dockerfile.stage COPYs dist/soul-demo. We only need the binary to exist
-        # We only need the COPY to succeed here; real tar is built by
+        # for the Docker build to succeed; the real binary is compiled in stage CI.
-        # build-stage.sh in the deploy pipeline.
+        run: |
-        run: touch dist/soul-demo-image.tar
+          touch dist/soul-demo
+          chmod +x dist/soul-demo
 
       - name: Build Docker image (local only — no push)
         run: |

.gitea/workflows/stage.yaml

@@ -173,15 +173,15 @@ jobs:
 
       # ── Docker build + push ───────────────────────────────────────────────
 
-      - name: Build soul-demo image tar
+      - name: Build soul-demo binary
-        # Dockerfile.stage COPYs dist/soul-demo-image.tar so k3s can import
+        # Compile soul-demo directly on the host runner (ci-base has gcc).
-        # soul-demo:local at runtime. We compile soul-demo from source on the
+        # Cloud Run runs soul-demo as a direct subprocess with a watchdog loop —
-        # host runner (ci-base has gcc), build a minimal OCI image, and save it.
+        # no k3s, no OCI image needed. One binary per container; Cloud Run
+        # handles horizontal scaling.
         # Moved AFTER JS compilation to avoid Docker memory pressure killing elc.
         if: steps.changetype.outputs.asset_only != 'true'
         run: |
           set -euo pipefail
-          # Compile el_runtime.o and soul-demo on the host runner
           cc -O2 -DHAVE_CURL -c runtime/el_runtime.c -I runtime/ -o /tmp/el_runtime.o
           cc -O2 -rdynamic -DEL_SOUL_DEMO_BUILD \
              -I runtime/ \
@@ -189,26 +189,6 @@ jobs:
              dist/soul-demo.c dist/vessel_stubs.c /tmp/el_runtime.o \
              -lcurl -lpthread -ldl -lm -lssl -lcrypto
           echo "soul-demo compiled: $(ls -lh dist/soul-demo)"
-          # Package as minimal OCI image for k3s import
-          # --no-cache: prevents reuse of corrupted overlay2 layers from prior failed runs
-          docker build --no-cache -f dist/Dockerfile.soul-demo -t soul-demo:local dist/
-          docker save soul-demo:local -o dist/soul-demo-image.tar
-          echo "soul-demo-image.tar: $(du -sh dist/soul-demo-image.tar | cut -f1)"
-          docker rmi soul-demo:local 2>/dev/null || true
-
-      - name: Download k3s binary
-        # Pre-download k3s on the host runner so Dockerfile.stage can COPY it
-        # directly. Previously k3s was downloaded inside the Docker builder stage,
-        # which combined with build-essential and C compilation caused RWLayer nil
-        # corruption on the runner's overlay2 driver. Host-runner download is safe.
-        if: steps.changetype.outputs.asset_only != 'true'
-        run: |
-          set -euo pipefail
-          curl -fL --retry 3 --retry-delay 10 \
-            https://github.com/k3s-io/k3s/releases/download/v1.32.4%2Bk3s1/k3s \
-            -o dist/k3s
-          chmod +x dist/k3s
-          echo "k3s: $(ls -lh dist/k3s)"
 
       - name: Build and tag image
         if: steps.changetype.outputs.asset_only != 'true'

Dockerfile.stage

@@ -4,15 +4,13 @@
 #   - neuron-web  on port 8080 (landing page server)
 #   - soul-demo   on port 7772 (demo chat, localhost only)
 #
-# All binaries (neuron-web, soul-demo, k3s) are pre-built by CI on the host
+# All binaries (neuron-web, soul-demo) are pre-built by CI on the host runner
-# runner before this Dockerfile runs. This keeps the Docker build single-stage
+# before this Dockerfile runs. This keeps the Docker build single-stage with
-# with no compilation and no network downloads, eliminating the multi-stage
+# no compilation and no network downloads.
-# complexity that caused RWLayer corruption on the runner's overlay2 driver.
 #
 # CI pre-build steps (in stage.yaml):
 #   - neuron-web: built by `elb build` → dist/neuron-landing
 #   - soul-demo:  compiled by cc on host → dist/soul-demo
-#   - k3s:        downloaded by curl on host → dist/k3s
 
 FROM ubuntu:24.04
 
@@ -27,9 +25,7 @@ RUN apt-get update \
  && groupadd -r landing && useradd -r -g landing landing \
  && mkdir -p /srv/landing/assets /srv/landing/js /srv/landing/shares \
  && mkdir -p /srv/soul/engram-demo \
- && chown -R landing:landing /srv/landing /srv/soul \
+ && chown -R landing:landing /srv/landing /srv/soul
- && mkdir -p /var/lib/rancher/k3s /tmp/k3s \
- && chown -R landing:landing /var/lib/rancher /tmp/k3s
 
 # neuron-web binary — produced by `elb build` in CI (linux/amd64)
 COPY dist/neuron-landing /usr/local/bin/neuron-web
@@ -39,18 +35,6 @@ RUN chmod +x /usr/local/bin/neuron-web
 COPY dist/soul-demo /usr/local/bin/soul-demo
 RUN chmod +x /usr/local/bin/soul-demo
 
-# k3s binary — downloaded from GitHub releases by CI
-COPY dist/k3s /usr/local/bin/k3s
-RUN chmod +x /usr/local/bin/k3s
-
-# soul-demo OCI image tar — k3s imports this at startup (no registry needed)
-RUN mkdir -p /var/lib/rancher/k3s/agent/images
-COPY dist/soul-demo-image.tar /var/lib/rancher/k3s/agent/images/soul-demo.tar
-
-# k3s manifests — auto-applied when k3s starts
-RUN mkdir -p /var/lib/rancher/k3s/server/manifests
-COPY dist/k3s-soul-demo.yaml /var/lib/rancher/k3s/server/manifests/soul-demo.yaml
-
 # Engram snapshot — baked in so soul has memory from cold start
 COPY dist/engram-snapshot.json /srv/soul/engram-demo/snapshot.json
 
@@ -73,11 +57,7 @@ ENV LANDING_ROOT=/srv/landing
 ENV PORT=8080
 ENV NEURON_HOME=/srv/soul/engram-demo
 ENV NEURON_PORT=7772
-ENV K3S_DATA_DIR=/var/lib/rancher/k3s
-ENV KUBECONFIG=/var/lib/rancher/k3s/server/cred/admin.kubeconfig
 
-# k3s requires root to create network namespaces and mount cgroups.
-# Cloud Run gen2 sandbox is the security boundary here.
 EXPOSE 8080
 
 CMD ["/usr/local/bin/entrypoint.sh"]

dist/entrypoint.sh

@@ -1,41 +1,26 @@
 #!/bin/sh
 set -e
 
-# SKIP_K3S=1 — bypass k3s/soul-demo startup and go straight to neuron-web.
-# Used by the dev CI smoke test where the container runtime doesn't support
-# the kernel capabilities k3s requires (overlayfs / privileged mode).
 if [ "${SKIP_K3S:-0}" = "1" ]; then
-  echo "[entrypoint] SKIP_K3S=1: starting neuron-web directly (no k3s/soul-demo)."
+  echo "[entrypoint] SKIP_K3S=1: starting neuron-web directly (no soul-demo)."
   exec /usr/local/bin/neuron-web
 fi
 
-echo "[entrypoint] Starting k3s server (embedded soul-demo orchestrator)..."
+# Soul-demo watchdog: start soul-demo and restart it automatically on crash.
+# Cloud Run gen2 doesn't reliably provide eth0 with a unicast IP, so k3s flannel
+# fails at startup. Running soul-demo directly is simpler, lighter, and fully
+# self-healing. Cloud Run handles horizontal scaling — no HPA needed.
+echo "[entrypoint] Starting soul-demo watchdog on :${NEURON_PORT:-7772}..."
+(
+  while true; do
+    echo "[soul-watchdog] starting soul-demo (NEURON_HOME=${NEURON_HOME})"
+    /usr/local/bin/soul-demo 2>&1 || true
+    echo "[soul-watchdog] soul-demo exited, restarting in 3s..."
+    sleep 3
+  done
+) &
 
-# k3s server — single-node mode, disable unused components
+# Start neuron-web immediately — do NOT block.
-# --disable traefik,servicelb: we don't need an ingress or LB
+# Cloud Run startup probe requires port 8080 to answer within the timeout.
-# --disable metrics-server: saves ~50MB RAM
+echo "[entrypoint] Starting neuron-web on port ${PORT:-8080}..."
-# --write-kubeconfig-mode=644: allow non-root reads
-# --data-dir: use the pre-chowned dir
-# --flannel-iface=eth0: explicitly set the network interface.
-#   Cloud Run gen2 provides eth0 but k3s default IP detection walks the routing
-#   table looking for a default route, which fails in Cloud Run's network sandbox.
-#   Pinning to eth0 bypasses that detection and lets k3s bind correctly.
-k3s server \
-  --disable traefik \
-  --disable servicelb \
-  --disable metrics-server \
-  --write-kubeconfig-mode=644 \
-  --data-dir /var/lib/rancher/k3s \
-  --node-name soul-node \
-  --flannel-iface=eth0 &
-
-K3S_PID=$!
-
-# Start neuron-web immediately — do NOT block on k3s becoming ready.
-# Cloud Run's startup probe requires port 8080 to be listening within the
-# startup timeout. k3s may take 30-60s to initialise; blocking here causes
-# probe failures and container termination before neuron-web ever starts.
-# soul-demo becomes available asynchronously once k3s is ready. neuron-web
-# handles soul-demo being temporarily unavailable gracefully.
-echo "[entrypoint] Starting neuron-web on port ${PORT:-8080} (k3s initialising in background)..."
 exec /usr/local/bin/neuron-web

manifest.el

@@ -16,4 +16,12 @@ build {
     c_source "dist/page_css.c"
     c_source "dist/page_ga.c"
     c_source "dist/page_schema.c"
+    // NOTE: neuron-web requires el_runtime.c to be compiled with -DHAVE_CURL
+    // so that http_get/http_post forward to libcurl instead of returning
+    // {"error":"not built with HAVE_CURL"}. The elb binary in ci-base:dev
+    // hardcodes -DHAVE_CURL in its cc invocation, but older elb versions may
+    // not. manifest.el does not support c_flags or link_flags directives —
+    // if upgrading elb breaks HTTP, ensure ci-base:dev ships an elb built
+    // with HAVE_CURL enabled in its hardcoded cc command, or pre-compile
+    // el_runtime.o with -DHAVE_CURL on the host and pass it as a c_source.
 }