feat: run qdrant and ollama inside thothii

scripts/run-stack.sh

@@ -1,8 +1,8 @@
 #!/usr/bin/env bash
 # run-stack.sh — avvia lo stack Compose locale di ThothII in primo piano.
 #
-# Il core include Pi; DWH, vector DB, embedding e LLM sono endpoint esterni configurati
-# in deploy/env/local.env. Non richiede un eseguibile Pi sull'host.
+# Il core include Pi; DWH e LLM restano endpoint esterni configurati in deploy/env/local.env.
+# Qdrant e Ollama embedding sono servizi Compose privati. Non richiede un eseguibile Pi sull'host.
 #
 # Preparazione: cp deploy/env/local.env.example deploy/env/local.env e compilare i valori.
 # Uso: ./scripts/run-stack.sh [argomenti aggiuntivi per docker compose up]
@@ -16,5 +16,10 @@ LOCAL_ENV_FILE="${THT_LOCAL_ENV_FILE:-$ROOT/deploy/env/local.env}"
   exit 1
 }
 
+compose_files=(-f "$ROOT/compose.yaml" -f "$ROOT/deploy/compose.local.yaml")
+if [[ "${THOTH_ENABLE_EMBEDDING_GPU:-0}" == "1" ]]; then
+  compose_files+=(-f "$ROOT/deploy/compose.embedding-gpu.yaml")
+fi
+
 exec docker compose --env-file "$LOCAL_ENV_FILE" \
-  -f "$ROOT/compose.yaml" -f "$ROOT/deploy/compose.local.yaml" up --build "$@"
+  "${compose_files[@]}" up --build "$@"

scripts/test-default-compose.sh

@@ -11,16 +11,70 @@ rendered=$(mktemp)
 trap 'rm -f "$rendered"' EXIT HUP INT TERM
 
 docker compose --env-file deploy/env/local.env.example \
-  -f compose.yaml -f deploy/compose.local.yaml config >"$rendered"
+  -f compose.yaml -f deploy/compose.local.yaml config --format json >"$rendered"
 
-grep -q '^  core:' "$rendered"
-grep -q '^  frontend:' "$rendered"
-grep -q 'host_ip: 127.0.0.1' "$rendered"
-grep -q 'AUTH_MODE: none' "$rendered"
-grep -q 'THT_WORKSPACE_INSTALLATION_ID: local' "$rendered"
-if grep -Eqi 'omics_portal|chirone|localllm_default|/home/chirone' "$rendered"; then
-  echo "default Compose contains application-specific coupling" >&2
-  exit 1
-fi
+node - "$rendered" <<'NODE'
+const fs = require("fs");
+
+const config = JSON.parse(fs.readFileSync(process.argv[2], "utf8"));
+const services = Object.keys(config.services).sort();
+if (services.join(",") !== "core,embedding,embedding-model-init,frontend,qdrant") {
+  throw new Error(`unexpected service set: ${services.join(",")}`);
+}
+if (/omics_portal|chirone|localllm_default|\/home\/chirone/i.test(JSON.stringify(config))) {
+  throw new Error("default Compose contains application-specific coupling");
+}
+for (const volume of ["settings", "pi-state", "workspace-registry", "sessions", "qdrant-data", "embedding-models"]) {
+  if (!config.volumes || !config.volumes[volume]) throw new Error(`missing required volume: ${volume}`);
+}
+const core = config.services.core;
+const frontend = config.services.frontend;
+const qdrant = config.services.qdrant;
+const embedding = config.services.embedding;
+const modelInit = config.services["embedding-model-init"];
+if (!frontend.ports?.some((port) => port.host_ip === "127.0.0.1")) {
+  throw new Error("local frontend must publish a loopback port");
+}
+for (const service of [qdrant, embedding, modelInit]) {
+  if ((service.ports || []).length !== 0) throw new Error("private semantic services must not publish host ports");
+}
+if ((qdrant.expose || []).join(",") !== "6333") throw new Error("qdrant must expose only 6333");
+if ((embedding.expose || []).join(",") !== "11434") throw new Error("embedding must expose only 11434");
+if (qdrant.image !== "qdrant/qdrant:v1.18.2@sha256:75eab8c4ba42096724fdcfde8b4de0b5713d529dde32f285a1f86fdcb2c9e50c") {
+  throw new Error("qdrant image must be pinned by version and digest");
+}
+if (embedding.image !== "ollama/ollama:0.32.0@sha256:57f573b47f1f71ebb445789f279fe3e596a8beab182f7cf486db9205bad87c5a") {
+  throw new Error("embedding image must be pinned by version and digest");
+}
+if (modelInit.image !== "ollama/ollama:0.32.0@sha256:57f573b47f1f71ebb445789f279fe3e596a8beab182f7cf486db9205bad87c5a") {
+  throw new Error("embedding-model-init image must be pinned by version and digest");
+}
+const env = core.environment || {};
+for (const [key, value] of Object.entries({
+  AUTH_MODE: "none",
+  THT_WORKSPACE_INSTALLATION_ID: "local",
+  THT_INTERNAL_QDRANT_URL: "http://qdrant:6333",
+  THT_INTERNAL_EMBEDDING_URL: "http://embedding:11434",
+  THT_INTERNAL_EMBEDDING_MODEL: "qwen3-embedding:0.6b",
+  THT_INTERNAL_EMBEDDING_DIMENSIONS: "1024",
+})) {
+  if (env[key] !== value) throw new Error(`unexpected core ${key}: ${env[key]}`);
+}
+for (const forbidden of ["THT_VEC_REST_URL", "THT_VEC_WRITE_REST_URL", "THT_OLLAMA_URL"]) {
+  if (Object.hasOwn(env, forbidden) && env[forbidden] !== "") {
+    throw new Error(`core must not require external semantic binding ${forbidden}`);
+  }
+}
+const depends = core.depends_on || {};
+if (depends.qdrant?.condition !== "service_healthy") {
+  throw new Error("core must wait for qdrant health");
+}
+if (depends["embedding-model-init"]?.condition !== "service_completed_successfully") {
+  throw new Error("core must wait for embedding-model-init success");
+}
+if (JSON.stringify(embedding).includes('"devices"')) {
+  throw new Error("base embedding service must stay CPU-only");
+}
+NODE
 
 echo "default Compose contract passed."

scripts/test-internal-semantic-compose.sh

@@ -0,0 +1,148 @@
+#!/usr/bin/env bash
+set -euo pipefail
+
+cd "$(dirname "$0")/.."
+
+tmp=$(mktemp -d)
+trap 'rm -rf "$tmp"' EXIT HUP INT TERM
+
+docker compose --env-file deploy/env/local.env.example \
+  -f compose.yaml -f deploy/compose.local.yaml -f deploy/compose.embedding-gpu.yaml \
+  config --format json >"$tmp/compose-gpu.json"
+
+node - "$tmp/compose-gpu.json" <<'NODE'
+const fs = require("fs");
+
+const config = JSON.parse(fs.readFileSync(process.argv[2], "utf8"));
+const devices = config.services.embedding?.deploy?.resources?.reservations?.devices;
+if (!Array.isArray(devices) || devices.length !== 1) {
+  throw new Error("GPU override must add one embedding device reservation");
+}
+const [device] = devices;
+if (JSON.stringify(device.capabilities) !== JSON.stringify(["gpu"])) {
+  throw new Error("GPU override must request gpu capability only");
+}
+NODE
+
+mock_bin="$tmp/mock-bin"
+mkdir -p "$mock_bin"
+
+cat >"$mock_bin/ollama" <<'EOF'
+#!/usr/bin/env bash
+set -euo pipefail
+
+state_dir=${MOCK_STATE_DIR:?}
+printf '%s\n' "$*" >>"$state_dir/ollama-calls"
+if [[ "$1" != "pull" ]]; then
+  echo "unexpected ollama command: $*" >&2
+  exit 1
+fi
+cat >"$state_dir/tags.json" <<JSON
+{"models":[{"name":"${2}"}]}
+JSON
+EOF
+chmod +x "$mock_bin/ollama"
+
+cat >"$tmp/mock-tags-server.py" <<'PY'
+import http.server
+import os
+from pathlib import Path
+
+state_dir = Path(os.environ["MOCK_STATE_DIR"])
+port_file = Path(os.environ["MOCK_PORT_FILE"])
+
+
+class Handler(http.server.BaseHTTPRequestHandler):
+    def do_GET(self):
+        if self.path != "/api/tags":
+            self.send_response(404)
+            self.end_headers()
+            return
+        count_file = state_dir / "curl-count"
+        count = int(count_file.read_text() or "0") if count_file.exists() else 0
+        count += 1
+        count_file.write_text(str(count))
+        fail_until = int((state_dir / "fail-until").read_text()) if (state_dir / "fail-until").exists() else 0
+        if count <= fail_until:
+            self.send_response(503)
+            self.end_headers()
+            self.wfile.write(b'{"models":[]}')
+            return
+        payload = (state_dir / "tags.json").read_bytes()
+        self.send_response(200)
+        self.send_header("Content-Type", "application/json")
+        self.send_header("Content-Length", str(len(payload)))
+        self.end_headers()
+        self.wfile.write(payload)
+
+    def log_message(self, format, *args):
+        return
+
+
+server = http.server.ThreadingHTTPServer(("127.0.0.1", 0), Handler)
+port_file.write_text(str(server.server_address[1]))
+server.serve_forever()
+PY
+
+start_server() {
+  local state_dir=$1
+  local port_file="$state_dir/port"
+  MOCK_STATE_DIR="$state_dir" MOCK_PORT_FILE="$port_file" \
+    python3 "$tmp/mock-tags-server.py" >/dev/null 2>&1 &
+  local server_pid=$!
+  for _ in $(seq 1 50); do
+    [[ -f "$port_file" ]] && break
+    sleep 0.1
+  done
+  [[ -f "$port_file" ]] || {
+    echo "mock tags server did not start" >&2
+    kill "$server_pid" >/dev/null 2>&1 || true
+    exit 1
+  }
+  printf '%s %s\n' "$server_pid" "$(cat "$port_file")"
+}
+
+run_cached() {
+  local state_dir="$tmp/cached"
+  mkdir -p "$state_dir"
+  cat >"$state_dir/tags.json" <<'JSON'
+{"models":[{"name":"qwen3-embedding:0.6b"}]}
+JSON
+  read -r server_pid port < <(start_server "$state_dir")
+  PATH="$mock_bin:$PATH" \
+  MOCK_STATE_DIR="$state_dir" \
+  OLLAMA_BASE_URL="http://127.0.0.1:$port" \
+  OLLAMA_MODEL="qwen3-embedding:0.6b" \
+  OLLAMA_WAIT_TIMEOUT_SEC=2 \
+  ./docker/embedding-model-init.sh
+  kill "$server_pid" >/dev/null 2>&1 || true
+  wait "$server_pid" 2>/dev/null || true
+  if [[ -e "$state_dir/ollama-calls" ]]; then
+    echo "cached bootstrap must not call ollama pull" >&2
+    exit 1
+  fi
+}
+
+run_pull() {
+  local state_dir="$tmp/pull"
+  mkdir -p "$state_dir"
+  cat >"$state_dir/tags.json" <<'JSON'
+{"models":[]}
+JSON
+  printf '2' >"$state_dir/fail-until"
+  read -r server_pid port < <(start_server "$state_dir")
+  PATH="$mock_bin:$PATH" \
+  MOCK_STATE_DIR="$state_dir" \
+  OLLAMA_BASE_URL="http://127.0.0.1:$port" \
+  OLLAMA_MODEL="qwen3-embedding:0.6b" \
+  OLLAMA_WAIT_TIMEOUT_SEC=5 \
+  ./docker/embedding-model-init.sh
+  kill "$server_pid" >/dev/null 2>&1 || true
+  wait "$server_pid" 2>/dev/null || true
+  grep -qx 'pull qwen3-embedding:0.6b' "$state_dir/ollama-calls"
+}
+
+run_cached
+run_pull
+
+echo "internal semantic Compose/script contracts passed."

scripts/test-unified-compose.sh

@@ -25,17 +25,65 @@ const fs = require("fs");
 const [configPath, profile] = process.argv.slice(2);
 const config = JSON.parse(fs.readFileSync(configPath, "utf8"));
 const services = Object.keys(config.services).sort();
-if (services.join(",") !== "core,frontend") throw new Error("mandatory stack must be core,frontend");
+if (services.join(",") !== "core,embedding,embedding-model-init,frontend,qdrant") {
+  throw new Error("mandatory stack must include core, frontend, qdrant, embedding, and embedding-model-init");
+}
 if (/omics_portal|chirone|localllm_default|\/home\/chirone/i.test(JSON.stringify(config))) {
   throw new Error("forbidden application coupling");
 }
 if (!config.networks || !config.networks.thothii) throw new Error("base stack must define the thothii network");
-if (!Object.hasOwn(config.services.core.environment || {}, "THT_LLM_URL")) {
+for (const volume of ["qdrant-data", "embedding-models"]) {
+  if (!config.volumes || !config.volumes[volume]) throw new Error(`missing required volume: ${volume}`);
+}
+if (profile === "local") {
+  for (const volume of ["settings", "pi-state", "workspace-registry", "sessions"]) {
+    if (!config.volumes || !config.volumes[volume]) throw new Error(`missing local required volume: ${volume}`);
+  }
+}
+const coreEnv = config.services.core.environment || {};
+if (!Object.hasOwn(coreEnv, "THT_LLM_URL")) {
   throw new Error("core must expose a generic THT_LLM_URL endpoint contract");
 }
+for (const [key, value] of Object.entries({
+  THT_INTERNAL_QDRANT_URL: "http://qdrant:6333",
+  THT_INTERNAL_EMBEDDING_URL: "http://embedding:11434",
+  THT_INTERNAL_EMBEDDING_MODEL: "qwen3-embedding:0.6b",
+  THT_INTERNAL_EMBEDDING_DIMENSIONS: "1024",
+})) {
+  if (coreEnv[key] !== value) throw new Error(`unexpected core ${key}: ${coreEnv[key]}`);
+}
+for (const forbidden of ["THT_VEC_REST_URL", "THT_VEC_WRITE_REST_URL", "THT_OLLAMA_URL"]) {
+  if (Object.hasOwn(coreEnv, forbidden) && coreEnv[forbidden] !== "") {
+    throw new Error(`core must not require external semantic binding ${forbidden}`);
+  }
+}
 if (/docker\.sock|\/var\/run\/docker|docker[-_]?daemon/i.test(JSON.stringify(config.services))) {
   throw new Error("Compose must not mount the Docker socket or daemon");
 }
+if (config.services.qdrant.image !== "qdrant/qdrant:v1.18.2@sha256:75eab8c4ba42096724fdcfde8b4de0b5713d529dde32f285a1f86fdcb2c9e50c") {
+  throw new Error("qdrant image must be pinned by version and digest");
+}
+for (const serviceName of ["embedding", "embedding-model-init"]) {
+  if (config.services[serviceName].image !== "ollama/ollama:0.32.0@sha256:57f573b47f1f71ebb445789f279fe3e596a8beab182f7cf486db9205bad87c5a") {
+    throw new Error(`${serviceName} image must be pinned by version and digest`);
+  }
+}
+for (const serviceName of ["qdrant", "embedding", "embedding-model-init"]) {
+  if ((config.services[serviceName].ports || []).length !== 0) {
+    throw new Error(`${serviceName} must not publish a host port`);
+  }
+}
+if ((config.services.qdrant.expose || []).join(",") !== "6333") throw new Error("qdrant must expose only 6333");
+if ((config.services.embedding.expose || []).join(",") !== "11434") throw new Error("embedding must expose only 11434");
+if (config.services.core.depends_on?.qdrant?.condition !== "service_healthy") {
+  throw new Error("core must wait for qdrant health");
+}
+if (config.services.core.depends_on?.["embedding-model-init"]?.condition !== "service_completed_successfully") {
+  throw new Error("core must wait for embedding-model-init success");
+}
+if (JSON.stringify(config.services.embedding).includes('"devices"')) {
+  throw new Error("base embedding service must stay CPU-only");
+}
 const piAuthMounts = (config.services.core.volumes || []).filter(
   (mount) => mount.target === "/home/thoth/.pi/agent/auth.json",
 );
@@ -108,17 +156,40 @@ const fs = require("fs");
 
 const config = JSON.parse(fs.readFileSync(process.argv[2], "utf8"));
 const services = Object.keys(config.services).sort();
-if (services.join(",") !== "core,frontend") throw new Error("mandatory stack must be core,frontend");
+if (services.join(",") !== "core,embedding,embedding-model-init,frontend,qdrant") {
+  throw new Error("mandatory stack must include core, frontend, qdrant, embedding, and embedding-model-init");
+}
 if (/omics_portal|chirone|localllm_default|\/home\/chirone/i.test(JSON.stringify(config))) {
   throw new Error("forbidden application coupling");
 }
 if (!config.networks || !config.networks.thothii) throw new Error("base stack must define the thothii network");
-for (const volume of ["settings", "pi-state", "workspace-registry", "sessions"]) {
+for (const volume of ["settings", "pi-state", "workspace-registry", "sessions", "qdrant-data", "embedding-models"]) {
   if (!config.volumes || !config.volumes[volume]) throw new Error(`missing required volume: ${volume}`);
 }
 if (!Object.hasOwn(config.services.core.environment || {}, "THT_LLM_URL")) {
   throw new Error("core must expose a generic THT_LLM_URL endpoint contract");
 }
+for (const [key, value] of Object.entries({
+  THT_INTERNAL_QDRANT_URL: "http://qdrant:6333",
+  THT_INTERNAL_EMBEDDING_URL: "http://embedding:11434",
+  THT_INTERNAL_EMBEDDING_MODEL: "qwen3-embedding:0.6b",
+  THT_INTERNAL_EMBEDDING_DIMENSIONS: "1024",
+})) {
+  if (config.services.core.environment?.[key] !== value) {
+    throw new Error(`unexpected core ${key}: ${config.services.core.environment?.[key]}`);
+  }
+}
+for (const serviceName of ["qdrant", "embedding", "embedding-model-init"]) {
+  if ((config.services[serviceName].ports || []).length !== 0) {
+    throw new Error(`${serviceName} must not publish a host port`);
+  }
+}
+if (config.services.core.depends_on?.qdrant?.condition !== "service_healthy") {
+  throw new Error("core must wait for qdrant health");
+}
+if (config.services.core.depends_on?.["embedding-model-init"]?.condition !== "service_completed_successfully") {
+  throw new Error("core must wait for embedding-model-init success");
+}
 if (/docker\.sock|\/var\/run\/docker|docker[-_]?daemon/i.test(JSON.stringify(config.services))) {
   throw new Error("Compose must not mount the Docker socket or daemon");
 }