@malloy-publisher/server 0.0.198 → 0.0.199

package/package.json

@@ -1,7 +1,7 @@
 {
    "name": "@malloy-publisher/server",
    "description": "Malloy Publisher Server",
-   "version": "0.0.198",
+   "version": "0.0.199",
    "main": "dist/server.mjs",
    "bin": {
       "malloy-publisher": "dist/server.mjs"
@@ -51,7 +51,6 @@
       "@opentelemetry/sdk-metrics": "^2.0.0",
       "@opentelemetry/sdk-node": "^0.200.0",
       "@opentelemetry/sdk-trace-node": "^2.0.0",
-      "adm-zip": "^0.5.16",
       "async-mutex": "^0.5.0",
       "aws-sdk": "^2.1692.0",
       "body-parser": "^1.20.2",
@@ -61,6 +60,7 @@
       "cors": "^2.8.5",
       "duckdb": "1.4.4",
       "express": "^4.21.0",
+      "extract-zip": "^2.0.1",
       "globals": "^15.9.0",
       "handlebars": "^4.7.8",
       "http-proxy-middleware": "^3.0.5",
@@ -76,7 +76,6 @@
       "@eslint/eslintrc": "^3.3.1",
       "@eslint/js": "^9.23.0",
       "@faker-js/faker": "^9.4.0",
-      "@types/adm-zip": "^0.5.7",
       "@types/bun": "^1.2.20",
       "@types/cors": "^2.8.12",
       "@types/express": "^4.17.14",

package/src/controller/compile.controller.ts

@@ -1,4 +1,4 @@
-import type { LogMessage } from "@malloydata/malloy";
+import type { GivenValue, LogMessage } from "@malloydata/malloy";
 import { EnvironmentStore } from "../service/environment_store";
 
 export class CompileController {
@@ -14,6 +14,7 @@ export class CompileController {
       modelName: string,
       source: string,
       includeSql: boolean = false,
+      givens?: Record<string, GivenValue>,
    ): Promise<{ status: string; problems: LogMessage[]; sql?: string }> {
       const environment = await this.environmentStore.getEnvironment(
          environmentName,
@@ -24,6 +25,7 @@ export class CompileController {
          modelName,
          source,
          includeSql,
+         givens,
       );
 
       // Determine overall status based on presence of errors

package/src/controller/model.controller.ts

@@ -2,6 +2,7 @@ import { components } from "../api";
 import { ModelNotFoundError } from "../errors";
 import { EnvironmentStore } from "../service/environment_store";
 import type { FilterParams } from "../service/filter";
+import type { GivenValue } from "@malloydata/malloy";
 
 type ApiNotebook = components["schemas"]["Notebook"];
 type ApiModel = components["schemas"]["Model"];
@@ -97,6 +98,7 @@ export class ModelController {
       cellIndex: number,
       filterParams?: FilterParams,
       bypassFilters?: boolean,
+      givens?: Record<string, GivenValue>,
    ): Promise<{
       type: "code" | "markdown";
       text: string;
@@ -117,6 +119,11 @@ export class ModelController {
          throw new ModelNotFoundError(`${notebookPath} is a model`);
       }
 
-      return model.executeNotebookCell(cellIndex, filterParams, bypassFilters);
+      return model.executeNotebookCell(
+         cellIndex,
+         filterParams,
+         bypassFilters,
+         givens,
+      );
    }
 }

package/src/controller/query.controller.ts

@@ -4,6 +4,7 @@ import { API_PREFIX } from "../constants";
 import { ModelNotFoundError } from "../errors";
 import { EnvironmentStore } from "../service/environment_store";
 import type { FilterParams } from "../service/filter";
+import type { GivenValue } from "@malloydata/malloy";
 
 type ApiQuery = components["schemas"]["QueryResult"];
 
@@ -32,6 +33,7 @@ export class QueryController {
       compactJson: boolean = false,
       filterParams?: FilterParams,
       bypassFilters?: boolean,
+      givens?: Record<string, GivenValue>,
    ): Promise<ApiQuery> {
       const environment = await this.environmentStore.getEnvironment(
          environmentName,
@@ -49,6 +51,7 @@ export class QueryController {
             query,
             filterParams,
             bypassFilters,
+            givens,
          );
          const renderLogs = validateRenderTags(result);
          return {

package/src/health.spec.ts

@@ -0,0 +1,90 @@
+import { afterEach, beforeEach, describe, expect, it, spyOn } from "bun:test";
+import { Server } from "http";
+import { performGracefulShutdownAfterDrain } from "./health";
+import { logger } from "./logger";
+
+// Regression test for the graceful-shutdown ordering bug that caused
+//   [winston] Attempt to write logs with no transports: {"message":"Waiting 50 seconds..."}
+// to appear in production logs. logger.close() must run after every
+// logger.* call, including the "Waiting ... seconds after server close
+// before exit..." message.
+//
+// Tests call performGracefulShutdownAfterDrain directly rather than
+// emitting SIGTERM, so module-level operationalState is not mutated
+// and the spec stays isolated from sibling tests in the same process.
+describe("performGracefulShutdownAfterDrain: shutdown ordering", () => {
+   const originalExit = process.exit;
+   let callOrder: string[];
+
+   beforeEach(() => {
+      callOrder = [];
+
+      spyOn(logger, "info").mockImplementation(((msg: string) => {
+         callOrder.push(`info:${msg}`);
+         return logger;
+      }) as never);
+      spyOn(logger, "close").mockImplementation((() => {
+         callOrder.push("close");
+         return logger;
+      }) as never);
+      // Silence warn/error calls so spec output stays clean. They are
+      // not load-bearing for these assertions.
+      spyOn(logger, "warn").mockImplementation((() => logger) as never);
+      spyOn(logger, "error").mockImplementation((() => logger) as never);
+
+      process.exit = ((_code?: number) => {
+         callOrder.push("exit");
+      }) as never;
+   });
+
+   afterEach(() => {
+      process.exit = originalExit;
+   });
+
+   const fakeServer = (): Server => ({ listening: false }) as unknown as Server;
+
+   it("logs the 'Waiting ...' message before closing the logger", async () => {
+      await performGracefulShutdownAfterDrain(fakeServer(), fakeServer(), 0.05);
+
+      const waitingIdx = callOrder.findIndex((entry) =>
+         entry.startsWith("info:Waiting"),
+      );
+      const closeIdx = callOrder.indexOf("close");
+      const exitIdx = callOrder.indexOf("exit");
+
+      expect(waitingIdx).toBeGreaterThanOrEqual(0);
+      expect(closeIdx).toBeGreaterThanOrEqual(0);
+      expect(exitIdx).toBeGreaterThanOrEqual(0);
+      expect(waitingIdx).toBeLessThan(closeIdx);
+      expect(closeIdx).toBeLessThan(exitIdx);
+   });
+
+   it("emits no logger.info calls after logger.close", async () => {
+      await performGracefulShutdownAfterDrain(fakeServer(), fakeServer(), 0.05);
+
+      const closeIdx = callOrder.indexOf("close");
+      const lateInfoIdx = callOrder.findIndex(
+         (entry, idx) => idx > closeIdx && entry.startsWith("info:"),
+      );
+      expect(closeIdx).toBeGreaterThanOrEqual(0);
+      expect(lateInfoIdx).toBe(-1);
+   });
+
+   it("closes the logger exactly once", async () => {
+      await performGracefulShutdownAfterDrain(fakeServer(), fakeServer(), 0.05);
+
+      const closes = callOrder.filter((entry) => entry === "close").length;
+      expect(closes).toBe(1);
+   });
+
+   it("skips the 'Waiting ...' message when gracefulCloseTimeoutSeconds is 0", async () => {
+      await performGracefulShutdownAfterDrain(fakeServer(), fakeServer(), 0);
+
+      const waitingCalls = callOrder.filter((entry) =>
+         entry.startsWith("info:Waiting"),
+      );
+      expect(waitingCalls.length).toBe(0);
+      expect(callOrder.indexOf("close")).toBeGreaterThanOrEqual(0);
+      expect(callOrder.indexOf("exit")).toBeGreaterThanOrEqual(0);
+   });
+});

package/src/health.ts

@@ -57,8 +57,8 @@ export function markNotReady(): void {
  * 2. Waits shutdownDrainDurationSeconds to allow in-flight requests to complete
  * 3. Sets preGracefulShutdownCompleted flag (enables drainingGuard middleware to reject new requests)
  * 4. Closes main server and MCP server (stops accepting new connections)
- * 5. Closes logger
- * 6. Waits shutdownGracefulCloseTimeoutSeconds (if > 0) for final cleanup
+ * 5. Waits shutdownGracefulCloseTimeoutSeconds (if > 0) for final cleanup
+ * 6. Closes logger (last, so any logs emitted during cleanup are flushed)
  * 7. Exits process
  *
  * Note: drainingGuard only rejects requests after step 3 completes. During step 2,
@@ -92,51 +92,94 @@ export function registerSignalHandlers(
          }, shutdownDrainDurationSeconds * 1000),
       );
 
-      const closeServer = (server: Server, name: string) =>
-         new Promise<void>((resolve) => {
-            if (server && server.listening) {
-               server.close((err) => {
-                  if (err) {
-                     logger.error(`${name} close error:`, err);
-                  } else {
-                     logger.info(`${name} closed`);
-                  }
-                  resolve();
-               });
-            } else {
-               resolve();
-            }
-         });
-
-      await Promise.all([
-         closeServer(server, "Main server"),
-         closeServer(mcpServer, "MCP server"),
-      ]);
-
-      try {
-         await shutdownSDK();
-         logger.info("OpenTelemetry SDK shut down");
-      } catch (_error) {
-         /* do nothing */
-      }
-
-      try {
-         logger.close();
-      } catch (_error) {
-         /* do nothing */
-      }
-
-      if (shutdownGracefulCloseTimeoutSeconds > 0) {
-         logger.info(
-            `Waiting ${shutdownGracefulCloseTimeoutSeconds} seconds after server close before exit...`,
-         );
-         await new Promise((resolve) =>
-            setTimeout(resolve, shutdownGracefulCloseTimeoutSeconds * 1000),
-         );
-      }
-      process.exit(0);
+      await performGracefulShutdownAfterDrain(
+         server,
+         mcpServer,
+         shutdownGracefulCloseTimeoutSeconds,
+      );
    });
 }
+
+/**
+ * Performs the post-drain shutdown work: closes both HTTP servers,
+ * shuts down the OpenTelemetry SDK, waits the optional graceful-close
+ * window so any in-flight cleanup can finish logging, closes the
+ * winston logger, and exits the process.
+ *
+ * Exported so unit tests can exercise the close + log + exit ordering
+ * without emitting SIGTERM (which would leave module-level
+ * operationalState stuck in "draining" and leak into sibling specs).
+ */
+export async function performGracefulShutdownAfterDrain(
+   server: Server,
+   mcpServer: Server,
+   shutdownGracefulCloseTimeoutSeconds: number,
+): Promise<void> {
+   const closeServer = (server: Server, name: string) =>
+      new Promise<void>((resolve) => {
+         if (server && server.listening) {
+            server.close((err) => {
+               if (err) {
+                  logger.error(`${name} close error:`, err);
+               } else {
+                  logger.info(`${name} closed`);
+               }
+               resolve();
+            });
+         } else {
+            resolve();
+         }
+      });
+
+   await Promise.all([
+      closeServer(server, "Main server"),
+      closeServer(mcpServer, "MCP server"),
+   ]);
+
+   try {
+      await shutdownSDK();
+      logger.info("OpenTelemetry SDK shut down");
+   } catch (_error) {
+      /* do nothing */
+   }
+
+   try {
+      // Drain in-flight compiles and terminate worker_threads before
+      // we exit so a slow compile doesn't leave orphan worker
+      // processes. Lazy-imported to avoid pulling the pool module
+      // into the health.ts dep graph for tests that don't exercise
+      // the compile path.
+      const { getPackageLoadPool } = await import(
+         "./package_load/package_load_pool"
+      );
+      await getPackageLoadPool().shutdown();
+      logger.info("Package-load worker pool shut down");
+   } catch (_error) {
+      /* do nothing */
+   }
+
+   if (shutdownGracefulCloseTimeoutSeconds > 0) {
+      logger.info(
+         `Waiting ${shutdownGracefulCloseTimeoutSeconds} seconds after server close before exit...`,
+      );
+      await new Promise((resolve) =>
+         setTimeout(resolve, shutdownGracefulCloseTimeoutSeconds * 1000),
+      );
+   }
+
+   // Close the logger last so anything emitted during the wait window
+   // above (or by other shutdown paths still running) reaches its
+   // transports. Closing earlier triggers winston's
+   // "Attempt to write logs with no transports" warning on any
+   // subsequent logger call.
+   try {
+      logger.close();
+   } catch (_error) {
+      /* do nothing */
+   }
+
+   process.exit(0);
+}
 /**
  * Middleware that returns 503 for non-health and metrics requests when service is draining.
  * Must be registered before application routes.

package/src/instrumentation.ts

@@ -1,3 +1,4 @@
+import { monitorEventLoopDelay } from "node:perf_hooks";
 import { metrics } from "@opentelemetry/api";
 import { getNodeAutoInstrumentations } from "@opentelemetry/auto-instrumentations-node";
 import { OTLPLogExporter } from "@opentelemetry/exporter-logs-otlp-proto";
@@ -116,6 +117,55 @@ const httpRequestCount = meter.createCounter("http_server_requests_total", {
    description: "Total number of HTTP requests",
 });
 
+// Event-loop-delay metrics. A blocked event loop is the only way the
+// /health/liveness probe (a pure synchronous 200 handler) can fail under K8s,
+// so we surface p50/p99/max so an operator can correlate liveness restarts
+// with sustained event-loop pressure (large Malloy compiles, GC, etc.).
+const eventLoopHistogram = monitorEventLoopDelay({ resolution: 20 });
+eventLoopHistogram.enable();
+
+const eventLoopLagP50 = meter.createObservableGauge(
+   "publisher_event_loop_lag_p50_ms",
+   {
+      description:
+         "Event loop delay p50 since the last scrape, in milliseconds",
+      unit: "ms",
+   },
+);
+const eventLoopLagP99 = meter.createObservableGauge(
+   "publisher_event_loop_lag_p99_ms",
+   {
+      description:
+         "Event loop delay p99 since the last scrape, in milliseconds",
+      unit: "ms",
+   },
+);
+const eventLoopLagMax = meter.createObservableGauge(
+   "publisher_event_loop_lag_max_ms",
+   {
+      description:
+         "Event loop delay max since the last scrape, in milliseconds",
+      unit: "ms",
+   },
+);
+
+// Sample all three in one batch so the histogram reset can't race the reads.
+meter.addBatchObservableCallback(
+   (observableResult) => {
+      observableResult.observe(
+         eventLoopLagP50,
+         eventLoopHistogram.percentile(50) / 1e6,
+      );
+      observableResult.observe(
+         eventLoopLagP99,
+         eventLoopHistogram.percentile(99) / 1e6,
+      );
+      observableResult.observe(eventLoopLagMax, eventLoopHistogram.max / 1e6);
+      eventLoopHistogram.reset();
+   },
+   [eventLoopLagP50, eventLoopLagP99, eventLoopLagMax],
+);
+
 const IGNORED_PATHS = new Set([
    "/health",
    "/health/liveness",

package/src/mcp/tools/execute_query_tool.ts

@@ -1,6 +1,7 @@
 import { McpServer } from "@modelcontextprotocol/sdk/server/mcp.js";
 import { ErrorCode, McpError } from "@modelcontextprotocol/sdk/types.js";
 import { z } from "zod";
+import type { GivenValue } from "@malloydata/malloy";
 import { logger } from "../../logger";
 import { EnvironmentStore } from "../../service/environment_store";
 import { getMalloyErrorDetails, type ErrorDetails } from "../error_messages";
@@ -30,6 +31,12 @@ const executeQueryShape = {
       .describe(
          "Filter parameter values keyed by filter name. Used with sources that declare #(filter) annotations.",
       ),
+   givens: z
+      .record(z.unknown())
+      .optional()
+      .describe(
+         "Per-query given values that override model defaults. Keys are given names declared in the model's given: block.",
+      ),
 };
 
 // Type inference is handled automatically by the MCP server based on the executeQueryShape
@@ -56,6 +63,7 @@ export function registerExecuteQueryTool(
             sourceName,
             queryName,
             filterParams,
+            givens,
          } = params;
 
          logger.info("[MCP Tool executeQuery] Received params:", { params });
@@ -128,6 +136,8 @@ export function registerExecuteQueryTool(
                   undefined,
                   query,
                   filterParams,
+                  undefined,
+                  givens as Record<string, GivenValue> | undefined,
                );
                const { validateRenderTags } = await import(
                   "@malloydata/render-validator"
@@ -174,6 +184,8 @@ export function registerExecuteQueryTool(
                   queryName,
                   undefined,
                   filterParams,
+                  undefined,
+                  givens as Record<string, GivenValue> | undefined,
                );
                const { validateRenderTags } = await import(
                   "@malloydata/render-validator"