Implement Canvas V2 subgraphs, assets and asynchronous AI execution

deploy/check-canvas-real.sh

@@ -0,0 +1,49 @@
+#!/usr/bin/env bash
+# Explicit remote-runner opt-in: invokes real models using the dedicated CLI login.
+set -euo pipefail
+umask 077
+binary=$(realpath "$1")
+cli=${FOUNDRY_REAL_CODEX_INSTALL:-/opt/endoleap-infra/codex-cli}
+login=${FOUNDRY_REAL_CODEX_HOME:-/opt/endoleap-infra/foundry/codex/home}
+: "${FOUNDRY_CANVAS_IMAGE_MODELS:?Set only backend model names verified for image output}"
+test -x "$binary"
+test -s "$login/auth.json"
+test -s "$login/models_cache.json"
+suffix=$(openssl rand -hex 8)
+database=foundry-real-db-$suffix
+worker=foundry-real-worker-$suffix
+scratch=$(mktemp -d /tmp/foundry-real.XXXXXXXX)
+cleanup(){ docker rm -f "$worker" "$database" >/dev/null 2>&1 || true; rm -rf "$scratch"; }
+trap cleanup EXIT
+install -d -m 700 "$scratch/run"
+chown 1000:1000 "$scratch/run"
+docker run -d --name "$database" --network none -e POSTGRES_HOST_AUTH_METHOD=trust postgres:17-alpine >/dev/null
+for attempt in {1..30}; do if docker exec "$database" pg_isready -h 127.0.0.1 -U postgres >/dev/null 2>&1; then break; fi; sleep 1; done
+docker exec -i "$database" psql -X -U postgres -v ON_ERROR_STOP=1 < db/001-auth.sql
+docker exec "$database" psql -X -U postgres -v ON_ERROR_STOP=1 -c 'CREATE ROLE foundry_fixture LOGIN'
+python3 deploy/migrate.py --container "$database" --database postgres --role foundry_fixture
+# Only the CLI container sees the login. Core/tests receive the private socket,
+# never the login, runtime configuration or Docker socket.
+docker run -d --name "$worker" --init --network bridge --user 1000:1000 \
+ --cap-drop ALL --security-opt no-new-privileges --read-only --pids-limit 128 --memory 1g \
+ --tmpfs /tmp:rw,nosuid,nodev,size=128m --tmpfs /work:ro,nosuid,nodev \
+ --tmpfs /home/node/.codex:rw,nosuid,nodev,size=32m,uid=1000,gid=1000 --workdir /work \
+ -e "FOUNDRY_CANVAS_IMAGE_MODELS=$FOUNDRY_CANVAS_IMAGE_MODELS" \
+ --mount "type=bind,source=$binary,target=/opt/foundry,readonly" \
+ --mount "type=bind,source=$cli,target=/opt/codex,readonly" \
+ --mount "type=bind,source=$login/auth.json,target=/home/node/.codex/auth.json,readonly" \
+ --mount "type=bind,source=$login/models_cache.json,target=/home/node/.codex/models_cache.json,readonly" \
+ --mount "type=bind,source=$scratch/run,target=/run/foundry-codex" \
+ --mount type=bind,source=/etc/ssl/certs/ca-certificates.crt,target=/etc/ssl/certs/ca-certificates.crt,readonly \
+ node:24-bookworm-slim /opt/foundry codex-worker >/dev/null
+for attempt in {1..30}; do if [[ -S "$scratch/run/worker.sock" ]]; then break; fi; sleep 1; done
+test -S "$scratch/run/worker.sock"
+CGO_ENABLED=0 go test -mod=readonly -c -o bin/platform-real-tests ./internal/platform
+docker run --rm --network "container:$database" --user 1000:1000 \
+ --read-only --cap-drop ALL --security-opt no-new-privileges \
+ --tmpfs /tmp:rw,nosuid,nodev,noexec,size=16m,mode=1777 \
+ --mount "type=bind,source=$(pwd)/bin/platform-real-tests,target=/opt/integration-tests,readonly" \
+ --mount "type=bind,source=$scratch/run,target=/run/canvas-real,readonly" \
+ -e PLATFORM_TEST_DATABASE_URL=postgres://foundry_fixture@127.0.0.1:5432/postgres?sslmode=disable \
+ -e FOUNDRY_REAL_CODEX_SOCKET=/run/canvas-real/worker.sock \
+ --entrypoint /opt/integration-tests postgres:17-alpine -test.v -test.run '^TestCanvasAIRealWorker$' -test.timeout 6m

deploy/check-schema.sh

@@ -71,6 +71,7 @@ for package in platform auth; do
   CGO_ENABLED=0 go test -mod=readonly -c -o "bin/$package-tests" "./internal/$package"
   docker run --rm --network "container:$container" --user 1000:1000 \
     --read-only --cap-drop ALL --security-opt no-new-privileges \
+    --tmpfs /tmp:rw,nosuid,nodev,noexec,size=16m,mode=1777 \
     --mount "type=bind,source=$(pwd)/bin/$package-tests,target=/opt/integration-tests,readonly" \
     -e PLATFORM_TEST_DATABASE_URL=postgres://foundry_fixture@127.0.0.1:5432/postgres?sslmode=disable \
     --entrypoint /opt/integration-tests postgres:17-alpine -test.v

deploy/codex-worker.sh

@@ -10,15 +10,19 @@ test -s "$dir/grants/projects.json"
 install -d -m 700 "$dir/run"
 chown 1000:1000 "$dir/run"
 
+canvas_env=()
+if [[ -s "$dir/canvas.env" ]]; then canvas_env=(--env-file "$dir/canvas.env"); fi
+
 # The host operator supplies an isolated Codex home and starts a single worker.
 # The Docker default bridge supplies egress independently of Foundry's network.
 docker run -d --name foundry-codex-worker --restart unless-stopped --init \
   --network bridge --user 1000:1000 --cap-drop ALL --security-opt no-new-privileges \
-  --read-only --pids-limit 128 --memory 384m --tmpfs /tmp:rw,nosuid,nodev \
+  --read-only --pids-limit 128 --memory 1g --tmpfs /tmp:rw,nosuid,nodev \
   --tmpfs /work:ro,nosuid,nodev --workdir /work \
   --mount "type=bind,source=$binary,target=/opt/foundry,readonly" \
   --mount "type=bind,source=$base/codex-cli,target=/opt/codex,readonly" \
   --mount "type=bind,source=$dir/home,target=/home/node/.codex" \
   --mount "type=bind,source=$dir/run,target=/run/foundry-codex" \
   --mount "type=bind,source=/etc/ssl/certs/ca-certificates.crt,target=/etc/ssl/certs/ca-certificates.crt,readonly" \
+  "${canvas_env[@]}" \
   node:24-bookworm-slim /opt/foundry codex-worker

deploy/release.sh

@@ -70,9 +70,40 @@ fi
 codex_mounts=()
 codex_dir=$foundry/codex
 if [[ -d "$codex_dir" ]]; then
-  if ! docker inspect -f '{{.State.Running}}' foundry-codex-worker 2>/dev/null | grep -qx true; then
-    bash "$(dirname "$0")/codex-worker.sh" "$release/foundry" >/dev/null
+  # Core and its private worker must speak the same Canvas artifact protocol.
+  # Keep the previous container definition for exact startup rollback.
+  worker_backup=
+  if docker inspect foundry-codex-worker >/dev/null 2>&1; then
+    worker_backup=foundry-codex-worker-previous-$(date +%s)-$$
+    docker stop -t 5 foundry-codex-worker >/dev/null
+    docker rename foundry-codex-worker "$worker_backup"
   fi
+  worker_ready=false
+  if bash "$(dirname "$0")/codex-worker.sh" "$release/foundry" >/dev/null; then
+    for attempt in {1..20}; do
+      if docker exec foundry-codex-worker node -e '
+        const http = require("node:http");
+        const req = http.get({socketPath:"/run/foundry-codex/worker.sock",path:"/canvas/models"}, res => {
+          let body = "";
+          res.on("data", chunk => { body += chunk; if(body.length > 262144) req.destroy(); });
+          res.on("end", () => { try { const value=JSON.parse(body); process.exit(res.statusCode===200 && Array.isArray(value.models) && value.models.length>0 ? 0 : 1); } catch { process.exit(1); } });
+        });
+        req.setTimeout(2000, () => req.destroy());
+        req.on("error", () => process.exit(1));
+      ' >/dev/null 2>&1; then worker_ready=true; break; fi
+      sleep 0.25
+    done
+  fi
+  if [[ "$worker_ready" != true ]]; then
+    docker rm -f foundry-codex-worker >/dev/null 2>&1 || true
+    if [[ -n "$worker_backup" ]]; then
+      docker rename "$worker_backup" foundry-codex-worker
+      docker start foundry-codex-worker >/dev/null
+    fi
+    printf 'Canvas worker startup/catalog verification failed; previous worker restored.\n' >&2
+    exit 1
+  fi
+  if [[ -n "$worker_backup" ]]; then docker rm "$worker_backup" >/dev/null; fi
   for attempt in {1..20}; do
     [[ ! -S "$codex_dir/run/worker.sock" ]] || break
     sleep 0.25
@@ -91,7 +122,7 @@ docker run -d --name "$candidate" "${codex_mounts[@]}" "${mail_env[@]}" "${slack
   --env-file "$foundry/database.env" -e HOST -e PORT -e PUBLIC_ORIGIN \
   -e GOOGLE_CLIENT_ID -e GOOGLE_CLIENT_SECRET -e FOUNDRY_PROJECTS -e FOUNDRY_ONBOARDING \
   --user 1000:1000 --cap-drop ALL --security-opt no-new-privileges \
-  --read-only --memory 128m \
+  --read-only --memory 1g \
   --mount "type=bind,source=$ca_bundle,target=/etc/ssl/certs/ca-certificates.crt,readonly" \
   -e SSL_CERT_FILE=/etc/ssl/certs/ca-certificates.crt \
   --mount "type=bind,source=$release/foundry,target=/opt/foundry,readonly" \