@malloy-publisher/server 0.0.198-dev → 0.0.198-dev1

package/src/compile/protocol.ts

@@ -0,0 +1,251 @@
+/**
+ * Wire protocol between the main thread (CompileWorkerPool) and the
+ * compile worker threads. Messages flow in both directions over the
+ * worker_threads MessagePort:
+ *
+ *   main ──▶ worker:  CompileJobRequest      (start a compile)
+ *   worker ──▶ main:  CompileJobResult       (success)
+ *   worker ──▶ main:  CompileJobError        (failure)
+ *
+ *   worker ──▶ main:  SchemaForTablesRequest (proxy schema fetch)
+ *   worker ──▶ main:  SchemaForSqlRequest    (proxy SQL block schema)
+ *   main ──▶ worker:  SchemaForTablesResponse / SchemaForSqlResponse
+ *
+ *   main ──▶ worker:  ShutdownRequest        (graceful drain & exit)
+ *
+ * The protocol intentionally uses plain structured-clonable POJOs so
+ * `parentPort.postMessage` and `worker.postMessage` can transfer them
+ * via V8's structured clone — much cheaper than JSON.stringify for
+ * the multi-MB `modelDef` payloads that come back from compile.
+ *
+ * All requests are correlated by an opaque `requestId` string so the
+ * receiver can match responses without relying on FIFO ordering.
+ */
+
+import type {
+   Annotation,
+   SQLSourceDef,
+   TableSourceDef,
+} from "@malloydata/malloy";
+
+// ──────────────────────────────────────────────────────────────────────
+// Direction: main ──▶ worker (compile job)
+// ──────────────────────────────────────────────────────────────────────
+
+/**
+ * Connection metadata the worker needs to construct a stub
+ * `InfoConnection`. Resolved lazily — the worker asks the main thread
+ * for these on the first `lookupConnection(name)` call (see
+ * {@link ConnectionMetadataRequest}). We don't ship the full list
+ * upfront because the caller layer doesn't always know it; Malloy
+ * sees connection names only as `connection.table('...')`
+ * references inside the model.
+ */
+export interface ConnectionMetadata {
+   name: string;
+   dialectName: string;
+   digest: string;
+}
+
+export interface CompileJobRequest {
+   type: "compile";
+   requestId: string;
+   /** Absolute path to the package directory on disk. */
+   packagePath: string;
+   /** Path of the model file relative to `packagePath`. */
+   modelPath: string;
+   /** Name of the default connection (e.g. "duckdb"), or null. */
+   defaultConnectionName: string | null;
+   /** Optional row-build manifest passed through to the Runtime. */
+   buildManifest?: unknown;
+}
+
+// ──────────────────────────────────────────────────────────────────────
+// Direction: worker ──▶ main (compile result)
+// ──────────────────────────────────────────────────────────────────────
+
+/**
+ * Wire shape of a successful compile. Mirrors the fields the
+ * server's `Model` constructor needs to fully describe a `.malloy`
+ * file without holding a `ModelMaterializer` reference.
+ *
+ * The materializer itself is intentionally NOT shipped back — it
+ * binds to a Runtime that holds live native connection handles and
+ * cannot cross a worker_threads boundary. The main thread builds
+ * its own materializer lazily on the first query (see
+ * `Model.ensureMaterializer`).
+ */
+export interface CompileJobResult {
+   type: "compile-result";
+   requestId: string;
+   /** Whatever `await modelMaterializer.getModel()`._modelDef returned. */
+   modelDef: unknown;
+   /** Source-info entries (from imports + local sources). */
+   sourceInfos: unknown[];
+   /** Pre-extracted API source descriptors. */
+   sources: unknown[];
+   /** Pre-extracted API query descriptors. */
+   queries: unknown[];
+   /** Parsed `#(filter)` map, keyed by source name. */
+   filterMap: Array<[string, unknown[]]>;
+   /** Givens declared on the model, already in API shape so the main
+    *  thread can stash them on the `Model` without further conversion. */
+   givens?: unknown[];
+   /** Accumulated dataStyles (from HackyDataStylesAccumulator). */
+   dataStyles: unknown;
+   /** Wall-clock ms inside the worker for the actual compile. */
+   compileDurationMs: number;
+}
+
+export interface CompileJobError {
+   type: "compile-error";
+   requestId: string;
+   /** Serialized error — the main thread reconstructs an Error. */
+   error: SerializedError;
+}
+
+/**
+ * Error wire-shape. We cannot transfer Error instances directly
+ * across postMessage cleanly (Bun/Node behaviour diverges on stack
+ * propagation), so we ship a structured payload and reconstitute on
+ * the main thread.
+ */
+export interface SerializedError {
+   name: string;
+   message: string;
+   stack?: string;
+   /** Set when the error originated as a Malloy `MalloyError`. */
+   malloyProblems?: unknown[];
+   /** Set when the error originated as `ModelCompilationError`. */
+   isCompilationError?: boolean;
+}
+
+// ──────────────────────────────────────────────────────────────────────
+// Direction: worker ──▶ main (proxy connection metadata)
+// ──────────────────────────────────────────────────────────────────────
+
+export interface ConnectionMetadataRequest {
+   type: "connection-metadata";
+   requestId: string;
+   jobId: string;
+   connectionName: string;
+}
+
+export interface ConnectionMetadataResponse {
+   type: "connection-metadata-response";
+   requestId: string;
+   ok: true;
+   metadata: ConnectionMetadata;
+}
+
+// ──────────────────────────────────────────────────────────────────────
+// Direction: worker ──▶ main (proxy schema fetches)
+// ──────────────────────────────────────────────────────────────────────
+
+export interface SchemaForTablesRequest {
+   type: "schema-for-tables";
+   requestId: string;
+   /** Job this RPC belongs to (so main routes to the right config). */
+   jobId: string;
+   connectionName: string;
+   tables: Record<string, string>;
+   options: {
+      refreshTimestamp?: number;
+      modelAnnotation?: Annotation;
+   };
+}
+
+export interface SchemaForTablesResponse {
+   type: "schema-for-tables-response";
+   requestId: string;
+   ok: true;
+   schemas: Record<string, TableSourceDef>;
+   errors: Record<string, string>;
+}
+
+export interface SchemaForSqlRequest {
+   type: "schema-for-sql";
+   requestId: string;
+   jobId: string;
+   connectionName: string;
+   sentence: unknown;
+   options: {
+      refreshTimestamp?: number;
+      modelAnnotation?: Annotation;
+   };
+}
+
+export interface SchemaForSqlResponse {
+   type: "schema-for-sql-response";
+   requestId: string;
+   ok: true;
+   structDef?: SQLSourceDef;
+   error?: string;
+}
+
+export interface RpcErrorResponse {
+   type: "rpc-error";
+   requestId: string;
+   ok: false;
+   error: SerializedError;
+}
+
+// ──────────────────────────────────────────────────────────────────────
+// Direction: worker ──▶ main (file read for imports)
+// ──────────────────────────────────────────────────────────────────────
+
+/**
+ * Workers read most files directly via fs (they run in the same
+ * filesystem namespace). This RPC exists for the rare case where the
+ * package URL reader has host-specific behaviour (e.g. virtual files,
+ * remote URLs) — we delegate back to the main thread's URL reader so
+ * compile semantics stay identical to the in-process path.
+ */
+export interface ReadUrlRequest {
+   type: "read-url";
+   requestId: string;
+   jobId: string;
+   url: string;
+}
+
+export interface ReadUrlResponse {
+   type: "read-url-response";
+   requestId: string;
+   ok: true;
+   contents: string;
+   invalidationKey?: string | number | null;
+}
+
+// ──────────────────────────────────────────────────────────────────────
+// Lifecycle
+// ──────────────────────────────────────────────────────────────────────
+
+export interface ShutdownRequest {
+   type: "shutdown";
+}
+
+export interface ReadyMessage {
+   type: "ready";
+}
+
+// ──────────────────────────────────────────────────────────────────────
+// Union types for routing
+// ──────────────────────────────────────────────────────────────────────
+
+export type MainToWorkerMessage =
+   | CompileJobRequest
+   | ConnectionMetadataResponse
+   | SchemaForTablesResponse
+   | SchemaForSqlResponse
+   | ReadUrlResponse
+   | RpcErrorResponse
+   | ShutdownRequest;
+
+export type WorkerToMainMessage =
+   | CompileJobResult
+   | CompileJobError
+   | ConnectionMetadataRequest
+   | SchemaForTablesRequest
+   | SchemaForSqlRequest
+   | ReadUrlRequest
+   | ReadyMessage;

package/src/config.spec.ts

@@ -856,3 +856,309 @@ describe("Config Environment Variable Substitution", () => {
       });
    });
 });
+
+// TODO: Remove this during projects cleanup
+describe("Config legacy 'projects' key back-compat", () => {
+   const testServerRoot = path.join(process.cwd(), "test-temp-legacy-config");
+   const configPath = path.join(testServerRoot, PUBLISHER_CONFIG_NAME);
+
+   beforeEach(() => {
+      if (!fs.existsSync(testServerRoot)) {
+         fs.mkdirSync(testServerRoot, { recursive: true });
+      }
+   });
+
+   afterEach(() => {
+      if (fs.existsSync(configPath)) {
+         fs.unlinkSync(configPath);
+      }
+      if (fs.existsSync(testServerRoot)) {
+         fs.rmdirSync(testServerRoot, { recursive: true });
+      }
+   });
+
+   it("reads from legacy 'projects' key when 'environments' is absent", async () => {
+      // Pre-rename on-disk shape: top-level key is `projects`, not
+      // `environments`. Without back-compat this silently parses as empty.
+      const legacyConfig = {
+         frozenConfig: false,
+         projects: [
+            {
+               name: "legacy-env",
+               packages: [
+                  {
+                     name: "p1",
+                     location: "./packages/p1",
+                  },
+               ],
+            },
+         ],
+      };
+
+      fs.writeFileSync(configPath, JSON.stringify(legacyConfig, null, 2));
+
+      // Spy on logger.warn so we can assert the deprecation message fired.
+      const { logger } = await import("./logger");
+      const originalWarn = logger.warn;
+      const warnings: string[] = [];
+      logger.warn = ((msg: unknown, ..._rest: unknown[]) => {
+         warnings.push(typeof msg === "string" ? msg : String(msg));
+         return logger;
+      }) as typeof logger.warn;
+
+      try {
+         const result = getPublisherConfig(testServerRoot);
+
+         expect(result.environments.length).toBe(1);
+         expect(result.environments[0].name).toBe("legacy-env");
+         expect(result.environments[0].packages[0].name).toBe("p1");
+
+         expect(
+            warnings.some((w) => w.includes('uses deprecated "projects" key')),
+         ).toBe(true);
+      } finally {
+         logger.warn = originalWarn;
+      }
+   });
+
+   it("prefers the new 'environments' key when both are present", async () => {
+      const config = {
+         frozenConfig: false,
+         environments: [
+            {
+               name: "new-env",
+               packages: [{ name: "p1", location: "./packages/p1" }],
+            },
+         ],
+         projects: [
+            {
+               name: "should-be-ignored",
+               packages: [{ name: "p2", location: "./packages/p2" }],
+            },
+         ],
+      };
+
+      fs.writeFileSync(configPath, JSON.stringify(config, null, 2));
+
+      const { logger } = await import("./logger");
+      const originalWarn = logger.warn;
+      const warnings: string[] = [];
+      logger.warn = ((msg: unknown, ..._rest: unknown[]) => {
+         warnings.push(typeof msg === "string" ? msg : String(msg));
+         return logger;
+      }) as typeof logger.warn;
+
+      try {
+         const result = getPublisherConfig(testServerRoot);
+
+         expect(result.environments.length).toBe(1);
+         expect(result.environments[0].name).toBe("new-env");
+
+         // No deprecation warning should fire when `environments` is present.
+         expect(
+            warnings.some((w) => w.includes('uses deprecated "projects" key')),
+         ).toBe(false);
+      } finally {
+         logger.warn = originalWarn;
+      }
+   });
+});
+
+describe("Committed example configs", () => {
+   const serverDir = path.resolve(__dirname, "..");
+
+   it.each([
+      ["publisher.config.json", false],
+      ["publisher.config.example.duckdb.json", false],
+      ["publisher.config.example.bigquery.json", true],
+   ])(
+      "%s parses as a valid PublisherConfig",
+      (filename, expectsBigQueryConnection) => {
+         const filePath = path.join(serverDir, filename);
+         expect(fs.existsSync(filePath)).toBe(true);
+
+         const raw = fs.readFileSync(filePath, "utf-8");
+         const parsed = JSON.parse(raw) as PublisherConfig;
+
+         expect(parsed).toHaveProperty("environments");
+         expect(Array.isArray(parsed.environments)).toBe(true);
+         expect(parsed.environments.length).toBeGreaterThan(0);
+
+         const env = parsed.environments[0];
+         expect(env.name).toBeTruthy();
+         expect(Array.isArray(env.packages)).toBe(true);
+         expect(env.packages.length).toBeGreaterThan(0);
+
+         if (expectsBigQueryConnection) {
+            expect(
+               (env.connections ?? []).some((c) => c.type === "bigquery"),
+            ).toBe(true);
+            expect(
+               env.packages.some((p) => p.name === "bigquery-hackernews"),
+            ).toBe(true);
+         } else {
+            expect(
+               env.packages.some((p) => p.name === "bigquery-hackernews"),
+            ).toBe(false);
+         }
+      },
+   );
+});
+
+describe("Config path resolution (--config and bundled default)", () => {
+   const emptyServerRoot = path.join(process.cwd(), "test-empty-server-root");
+   const customConfigPath = path.join(
+      process.cwd(),
+      "test-custom-publisher.config.json",
+   );
+
+   beforeEach(() => {
+      if (!fs.existsSync(emptyServerRoot)) {
+         fs.mkdirSync(emptyServerRoot, { recursive: true });
+      }
+      delete process.env.PUBLISHER_CONFIG_PATH;
+      delete process.env.PUBLISHER_USE_BUNDLED_DEFAULT;
+   });
+
+   afterEach(() => {
+      delete process.env.PUBLISHER_CONFIG_PATH;
+      delete process.env.PUBLISHER_USE_BUNDLED_DEFAULT;
+      if (fs.existsSync(customConfigPath)) {
+         fs.unlinkSync(customConfigPath);
+      }
+      if (fs.existsSync(emptyServerRoot)) {
+         fs.rmSync(emptyServerRoot, { recursive: true, force: true });
+      }
+   });
+
+   it("falls back to the bundled default config when serverRoot has no publisher.config.json and bundled-default opt-in is set", () => {
+      // The bundled default is opt-in (server.ts sets the env var when
+      // the user passed neither --server_root nor --config) so embedded
+      // callers don't get a surprise filesystem read.
+      process.env.PUBLISHER_USE_BUNDLED_DEFAULT = "true";
+      const result = getPublisherConfig(emptyServerRoot);
+
+      expect(result.environments.length).toBeGreaterThan(0);
+      const env = result.environments[0];
+      expect(env.name).toBe("malloy-samples");
+      expect(env.packages.some((p) => p.name === "ecommerce")).toBe(true);
+      expect(env.packages.some((p) => p.name === "bigquery-hackernews")).toBe(
+         false,
+      );
+   });
+
+   it("does NOT fall back to the bundled default when opt-in flag is unset (programmatic construction)", () => {
+      // No PUBLISHER_USE_BUNDLED_DEFAULT, no PUBLISHER_CONFIG_PATH, no
+      // file in emptyServerRoot — result should be the original empty
+      // shape, preserving prior behavior for embeds and tests.
+      const result = getPublisherConfig(emptyServerRoot);
+      expect(result.environments).toEqual([]);
+   });
+
+   it("honors PUBLISHER_CONFIG_PATH (set by --config) over the server root", () => {
+      const customConfig = {
+         frozenConfig: false,
+         environments: [
+            { name: "custom-env", packages: [{ name: "foo", location: "/x" }] },
+         ],
+      };
+      fs.writeFileSync(customConfigPath, JSON.stringify(customConfig));
+      process.env.PUBLISHER_CONFIG_PATH = customConfigPath;
+
+      const result = getPublisherConfig(emptyServerRoot);
+      expect(result.environments[0].name).toBe("custom-env");
+   });
+
+   it("returns empty environments (not the bundled default) when --config points at a missing file", () => {
+      process.env.PUBLISHER_CONFIG_PATH = path.join(
+         process.cwd(),
+         "does-not-exist.json",
+      );
+      // Even with bundled-default opt-in, --config with a missing target
+      // is a user error and we don't paper over it.
+      process.env.PUBLISHER_USE_BUNDLED_DEFAULT = "true";
+      const result = getPublisherConfig(emptyServerRoot);
+      expect(result.environments).toEqual([]);
+   });
+});
+
+describe("getMemoryGovernorConfig", () => {
+   const GOVERNOR_ENV_VARS = [
+      "PUBLISHER_MAX_MEMORY_BYTES",
+      "PUBLISHER_MEMORY_HIGH_WATER_FRACTION",
+      "PUBLISHER_MEMORY_LOW_WATER_FRACTION",
+      "PUBLISHER_MEMORY_CHECK_INTERVAL_MS",
+      "PUBLISHER_MEMORY_BACKPRESSURE",
+   ];
+
+   beforeEach(() => {
+      for (const v of GOVERNOR_ENV_VARS) delete process.env[v];
+   });
+   afterEach(() => {
+      for (const v of GOVERNOR_ENV_VARS) delete process.env[v];
+   });
+
+   it("returns null when PUBLISHER_MAX_MEMORY_BYTES is unset", async () => {
+      const { getMemoryGovernorConfig } = await import("./config");
+      expect(getMemoryGovernorConfig()).toBeNull();
+   });
+
+   it("parses defaults when only PUBLISHER_MAX_MEMORY_BYTES is set", async () => {
+      process.env.PUBLISHER_MAX_MEMORY_BYTES = String(2 * 1024 * 1024 * 1024);
+      const { getMemoryGovernorConfig } = await import("./config");
+      const cfg = getMemoryGovernorConfig();
+      expect(cfg).not.toBeNull();
+      expect(cfg!.maxMemoryBytes).toBe(2 * 1024 * 1024 * 1024);
+      expect(cfg!.backpressureEnabled).toBe(true);
+      expect(cfg!.highWaterFraction).toBeGreaterThan(cfg!.lowWaterFraction);
+   });
+
+   it("honours fraction and interval overrides", async () => {
+      process.env.PUBLISHER_MAX_MEMORY_BYTES = "1000000000";
+      process.env.PUBLISHER_MEMORY_HIGH_WATER_FRACTION = "0.85";
+      process.env.PUBLISHER_MEMORY_LOW_WATER_FRACTION = "0.7";
+      process.env.PUBLISHER_MEMORY_CHECK_INTERVAL_MS = "10000";
+      process.env.PUBLISHER_MEMORY_BACKPRESSURE = "false";
+      const { getMemoryGovernorConfig } = await import("./config");
+      const cfg = getMemoryGovernorConfig();
+      expect(cfg).not.toBeNull();
+      expect(cfg!.highWaterFraction).toBe(0.85);
+      expect(cfg!.lowWaterFraction).toBe(0.7);
+      expect(cfg!.checkIntervalMs).toBe(10000);
+      expect(cfg!.backpressureEnabled).toBe(false);
+   });
+
+   it("treats PUBLISHER_MAX_MEMORY_BYTES=0 as disabled (returns null)", async () => {
+      process.env.PUBLISHER_MAX_MEMORY_BYTES = "0";
+      const { getMemoryGovernorConfig } = await import("./config");
+      expect(getMemoryGovernorConfig()).toBeNull();
+   });
+
+   it("rejects a negative PUBLISHER_MAX_MEMORY_BYTES", async () => {
+      process.env.PUBLISHER_MAX_MEMORY_BYTES = "-1";
+      const { getMemoryGovernorConfig } = await import("./config");
+      expect(() => getMemoryGovernorConfig()).toThrow();
+   });
+
+   it("rejects low >= high", async () => {
+      process.env.PUBLISHER_MAX_MEMORY_BYTES = "1000000000";
+      process.env.PUBLISHER_MEMORY_HIGH_WATER_FRACTION = "0.7";
+      process.env.PUBLISHER_MEMORY_LOW_WATER_FRACTION = "0.8";
+      const { getMemoryGovernorConfig } = await import("./config");
+      expect(() => getMemoryGovernorConfig()).toThrow();
+   });
+
+   it("rejects an out-of-range fraction", async () => {
+      process.env.PUBLISHER_MAX_MEMORY_BYTES = "1000000000";
+      process.env.PUBLISHER_MEMORY_HIGH_WATER_FRACTION = "1.5";
+      const { getMemoryGovernorConfig } = await import("./config");
+      expect(() => getMemoryGovernorConfig()).toThrow();
+   });
+
+   it("rejects a check interval below the safety floor", async () => {
+      process.env.PUBLISHER_MAX_MEMORY_BYTES = "1000000000";
+      process.env.PUBLISHER_MEMORY_CHECK_INTERVAL_MS = "10";
+      const { getMemoryGovernorConfig } = await import("./config");
+      expect(() => getMemoryGovernorConfig()).toThrow();
+   });
+});