feat: run qdrant and ollama inside thothii

2026-08-08 18:38:42 +02:00
parent ca1511df60
commit 320d9ea74e
11 changed files with 502 additions and 28 deletions
@@ -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."