gsd-pi 2.78.1-dev.b0759e59b → 2.78.1-dev.d8826a445

package/src/resources/extensions/gsd/tests/metrics-lock-retry-sleep.test.ts

@@ -0,0 +1,287 @@
+// GSD-2 + metrics-lock-retry-sleep.test.ts: verify sleep between lock acquire retries (M3 follow-up)
+/**
+ * Verifies that acquireLock sleeps between non-stale-evicting retries:
+ *
+ *   1. Under contention: a child process holds the lock for 100ms; the main
+ *      process acquireLock attempt should make at most ~30 sleepy retries
+ *      (100ms contention / 5ms sleep) — not thousands as would occur without
+ *      the sleep.
+ *
+ *   2. Stale-lock eviction path: when a stale lock is detected and forcibly
+ *      removed, the subsequent acquire attempt does NOT sleep — the sleepy
+ *      retry counter stays at zero and the acquire succeeds immediately.
+ *
+ *   3. Regression: M3 lock-hardening tests still pass (invoked separately
+ *      as part of the test suite — tested here by exercising the same
+ *      stale-lock + PID-stamp code paths through snapshotUnitMetrics).
+ */
+
+import { describe, test, beforeEach, afterEach } from "node:test";
+import assert from "node:assert/strict";
+import {
+  mkdtempSync,
+  mkdirSync,
+  rmSync,
+  writeFileSync,
+  existsSync,
+  utimesSync,
+} from "node:fs";
+import { join } from "node:path";
+import { tmpdir } from "node:os";
+import { spawnSync, spawn } from "node:child_process";
+
+import {
+  initMetrics,
+  resetMetrics,
+  snapshotUnitMetrics,
+  STALE_LOCK_THRESHOLD_MS,
+  LOCK_RETRY_INTERVAL_MS,
+  getLockSleepyRetries,
+  resetLockSleepyRetries,
+} from "../metrics.js";
+
+// ─── Helpers ─────────────────────────────────────────────────────────────────
+
+function makeProjectDir(): string {
+  const dir = mkdtempSync(join(tmpdir(), "gsd-metrics-sleep-"));
+  mkdirSync(join(dir, ".gsd"), { recursive: true });
+  return dir;
+}
+
+function metricsPath(base: string): string {
+  return join(base, ".gsd", "metrics.json");
+}
+
+function lockPath(base: string): string {
+  return metricsPath(base) + ".lock";
+}
+
+function assistantCtx(): any {
+  return {
+    sessionManager: {
+      getEntries: () => [
+        {
+          type: "message",
+          id: "entry-0",
+          parentId: null,
+          timestamp: new Date().toISOString(),
+          message: {
+            role: "assistant",
+            content: [{ type: "text", text: "Done" }],
+            usage: {
+              input: 100,
+              output: 50,
+              cacheRead: 0,
+              cacheWrite: 0,
+              totalTokens: 150,
+              cost: 0.001,
+            },
+          },
+        },
+      ],
+    },
+  };
+}
+
+// Worker that acquires the lock using O_EXCL and holds it for holdMs, then releases.
+// Writes its PID to stdout once the lock is acquired so the caller can synchronize.
+const LOCK_HOLDER_WORKER = `
+const { openSync, closeSync, writeFileSync, unlinkSync } = require('node:fs');
+const lockPath = process.env.GSD_TEST_LOCK_PATH;
+const holdMs = parseInt(process.env.GSD_TEST_HOLD_MS || '100', 10);
+
+const deadline = Date.now() + 3000;
+let acquired = false;
+while (Date.now() < deadline) {
+  try {
+    const fd = openSync(lockPath, 'wx');
+    closeSync(fd);
+    writeFileSync(lockPath, process.pid + '\\n' + new Date().toISOString() + '\\n', 'utf-8');
+    acquired = true;
+    break;
+  } catch { /* retry */ }
+}
+
+if (!acquired) {
+  process.stderr.write('Worker failed to acquire lock\\n');
+  process.exit(1);
+}
+
+// Signal that the lock is held.
+process.stdout.write(String(process.pid) + '\\n');
+
+// Hold the lock.
+const releaseAt = Date.now() + holdMs;
+while (Date.now() < releaseAt) { /* spin for short hold */ }
+
+try { unlinkSync(lockPath); } catch {}
+`;
+
+// ─── Tests ────────────────────────────────────────────────────────────────────
+
+describe("metrics lock retry sleep (M3 follow-up)", () => {
+  let tmpDir: string;
+
+  beforeEach(() => {
+    tmpDir = makeProjectDir();
+    resetLockSleepyRetries();
+  });
+
+  afterEach(() => {
+    resetMetrics();
+    rmSync(tmpDir, { recursive: true, force: true });
+  });
+
+  // ── Test 1: Under contention, sleep caps sleepy retries ───────────────────
+
+  test("sleepy retry count is bounded under lock contention (5ms sleep, 500ms timeout)", () => {
+    const lp = lockPath(tmpDir);
+
+    // Spawn a child that holds the lock for 100ms.
+    // We use spawnSync with a timeout > holdMs so it completes before our check.
+    // But we need the child to hold first THEN we attempt. Since spawnSync blocks,
+    // we pre-create the lock file instead to simulate contention for 100ms.
+
+    // Simulate: lock is held for 100ms from now, then released.
+    // We create the lock file manually (as if another process holds it) and
+    // schedule its removal after 100ms using a child process that holds then deletes.
+    //
+    // Strategy: use a background worker via spawnSync with hold=100ms,
+    // but we can't overlap with spawnSync. Instead, we directly test with
+    // snapshotUnitMetrics + a pre-placed lock + a child that will remove it.
+    //
+    // Simplest approach: place the lock file, run snapshotUnitMetrics
+    // which calls saveLedger → acquireLock. acquireLock will retry for 100ms
+    // (lock file just sits there), then we remove it from a thread... but
+    // there are no threads in Node main.
+    //
+    // The correct approach is to spawn a background child that holds the lock
+    // for 100ms, and run acquireLock from the main process concurrently via
+    // snapshotUnitMetrics (which is synchronous). The child is started as a
+    // background process using spawn (not spawnSync), then we call
+    // snapshotUnitMetrics synchronously and block until it acquires or times out.
+
+    const child = spawn(process.execPath, ["-e", LOCK_HOLDER_WORKER], {
+      env: {
+        ...process.env,
+        GSD_TEST_LOCK_PATH: lp,
+        GSD_TEST_HOLD_MS: "100",
+      },
+    });
+
+    // Wait until the child has acquired the lock (it writes PID to stdout).
+    // Poll until the lock file exists (the child acquired it).
+    const waitStart = Date.now();
+    while (!existsSync(lp) && Date.now() - waitStart < 2000) {
+      // Busy-wait for child to acquire the lock — this is test setup, not
+      // production code. Short window (child acquires almost immediately).
+      const arr = new Int32Array(new SharedArrayBuffer(4));
+      Atomics.wait(arr, 0, 0, 5);
+    }
+
+    assert.ok(existsSync(lp), "child must have acquired the lock before we attempt");
+
+    // Now call snapshotUnitMetrics synchronously. It will call saveLedger →
+    // acquireLock, which retries with 5ms sleep until the child releases (~100ms).
+    initMetrics(tmpDir);
+    const ctx = assistantCtx();
+    const start = Date.now();
+    const unit = snapshotUnitMetrics(ctx, "execute-task", "M001/S01/T01", Date.now() - 500, "test-model");
+    const elapsed = Date.now() - start;
+
+    // Clean up child.
+    child.kill();
+
+    // The operation must succeed (either by acquiring after child releases, or
+    // by timeout-fallback write in saveLedger).
+    assert.ok(unit !== null, "snapshotUnitMetrics must succeed under contention");
+
+    const retries = getLockSleepyRetries();
+
+    // With 5ms sleep and ~100ms contention, expect roughly 20 sleepy retries.
+    // Upper bound: 500ms timeout / 5ms = 100. With 2s default timeout, upper
+    // bound is 2000ms / 5ms = 400. But we want far fewer than the ~20,000
+    // that would occur without any sleep.
+    //
+    // Conservative assertion: fewer than 200 sleepy retries across the wait
+    // (vs ~20,000 without sleep). This is the key regression guard.
+    assert.ok(
+      retries < 200,
+      `Expected < 200 sleepy retries with 5ms sleep, got ${retries}. ` +
+      `Elapsed: ${elapsed}ms. LOCK_RETRY_INTERVAL_MS=${LOCK_RETRY_INTERVAL_MS}`,
+    );
+
+    // Sanity: at least a few retries happened (lock was actually contested).
+    assert.ok(
+      retries >= 1,
+      `Expected at least 1 sleepy retry (lock was held by child), got ${retries}`,
+    );
+  });
+
+  // ── Test 2: Stale-lock eviction does NOT sleep ────────────────────────────
+
+  test("stale-lock eviction path retries immediately without incrementing sleepy counter", () => {
+    const lp = lockPath(tmpDir);
+
+    // Create a stale lock (mtime older than STALE_LOCK_THRESHOLD_MS).
+    writeFileSync(lp, `999999\n${new Date(Date.now() - STALE_LOCK_THRESHOLD_MS - 500).toISOString()}\n`, "utf-8");
+    const staleTime = (Date.now() - STALE_LOCK_THRESHOLD_MS - 500) / 1000;
+    utimesSync(lp, staleTime, staleTime);
+
+    assert.ok(existsSync(lp), "stale lock file must exist before acquire");
+
+    initMetrics(tmpDir);
+    const ctx = assistantCtx();
+
+    const start = Date.now();
+    const unit = snapshotUnitMetrics(ctx, "execute-task", "M002/S01/T01", Date.now() - 200, "test-model");
+    const elapsed = Date.now() - start;
+
+    assert.ok(unit !== null, "snapshotUnitMetrics must succeed after stale-lock eviction");
+
+    const retries = getLockSleepyRetries();
+
+    // The stale-lock path uses `continue` (no sleep), so the sleepy retry
+    // counter must be zero: the lock was evicted and the very next openSync
+    // call succeeded — no sleepy retry was needed.
+    assert.equal(
+      retries,
+      0,
+      `Expected 0 sleepy retries after stale-lock eviction, got ${retries}. ` +
+      `Elapsed: ${elapsed}ms`,
+    );
+
+    // Should complete very quickly (no artificial delay).
+    assert.ok(
+      elapsed < 200,
+      `Stale-lock recovery should complete quickly, took ${elapsed}ms`,
+    );
+  });
+
+  // ── Test 3: M3 regression — stale lock recovers + metrics written ─────────
+
+  test("M3 regression: stale lock from dead process is cleared and metrics are written", () => {
+    const lp = lockPath(tmpDir);
+
+    // Backdate the lock file to simulate a crashed process.
+    const stalePid = 9999999;
+    writeFileSync(
+      lp,
+      `${stalePid}\n${new Date(Date.now() - STALE_LOCK_THRESHOLD_MS - 1000).toISOString()}\n`,
+      "utf-8",
+    );
+    const staleTime = (Date.now() - STALE_LOCK_THRESHOLD_MS - 1000) / 1000;
+    utimesSync(lp, staleTime, staleTime);
+
+    initMetrics(tmpDir);
+    const ctx = assistantCtx();
+    const unit = snapshotUnitMetrics(ctx, "execute-task", "M003/S01/T01", Date.now() - 300, "test-model");
+
+    assert.ok(unit !== null, "snapshotUnitMetrics must succeed after stale lock from dead process");
+    assert.equal(unit!.id, "M003/S01/T01");
+    assert.ok(
+      existsSync(metricsPath(tmpDir)),
+      "metrics.json must exist after recovery",
+    );
+  });
+});

package/src/resources/extensions/gsd/tests/metrics-prune-cache-invalidation.test.ts

@@ -0,0 +1,149 @@
+// GSD-2 + metrics prune cache invalidation: pruned units must not reappear after snapshotUnitMetricsByScope
+
+import { describe, test, beforeEach, afterEach } from "node:test";
+import assert from "node:assert/strict";
+import {
+  mkdtempSync,
+  mkdirSync,
+  readFileSync,
+  rmSync,
+  writeFileSync,
+} from "node:fs";
+import { join } from "node:path";
+import { tmpdir } from "node:os";
+
+import {
+  initMetricsByScope,
+  resetMetricsByScope,
+  snapshotUnitMetricsByScope,
+  pruneMetricsLedger,
+  type MetricsLedger,
+} from "../metrics.js";
+import { createWorkspace, scopeMilestone } from "../workspace.ts";
+
+// ─── Helpers ─────────────────────────────────────────────────────────────────
+
+function makeProjectDir(): string {
+  const dir = mkdtempSync(join(tmpdir(), "gsd-metrics-prune-"));
+  mkdirSync(join(dir, ".gsd"), { recursive: true });
+  return dir;
+}
+
+function metricsPath(base: string): string {
+  return join(base, ".gsd", "metrics.json");
+}
+
+function makeUnit(id: string, startedAt: number): any {
+  return {
+    type: "execute-task",
+    id,
+    model: "test-model",
+    startedAt,
+    finishedAt: startedAt + 1000,
+    tokens: { input: 100, output: 50, cacheRead: 0, cacheWrite: 0, total: 150 },
+    cost: 0.001,
+    toolCalls: 1,
+    assistantMessages: 1,
+    userMessages: 1,
+  };
+}
+
+function makeProjectLedger(units: any[]): MetricsLedger {
+  return { version: 1, projectStartedAt: 1000, units } as MetricsLedger;
+}
+
+function assistantCtx(): any {
+  const entries = [
+    {
+      type: "message",
+      id: "e0",
+      parentId: null,
+      timestamp: new Date().toISOString(),
+      message: {
+        role: "assistant",
+        content: [{ type: "text", text: "Done" }],
+        usage: {
+          input: 100,
+          output: 50,
+          cacheRead: 0,
+          cacheWrite: 0,
+          totalTokens: 150,
+          cost: 0.001,
+        },
+      },
+    },
+  ];
+  return { sessionManager: { getEntries: () => entries } };
+}
+
+// ─── Tests ───────────────────────────────────────────────────────────────────
+
+describe("pruneMetricsLedger: invalidates scoped ledger cache", () => {
+  let tmpDir: string;
+  let ws: ReturnType<typeof createWorkspace>;
+  let scope: ReturnType<typeof scopeMilestone>;
+
+  beforeEach(() => {
+    tmpDir = makeProjectDir();
+    mkdirSync(join(tmpDir, ".gsd", "milestones"), { recursive: true });
+    ws = createWorkspace(tmpDir);
+    scope = scopeMilestone(ws, "M001");
+  });
+
+  afterEach(() => {
+    resetMetricsByScope(scope);
+    rmSync(tmpDir, { recursive: true, force: true });
+  });
+
+  test("pruned units do not reappear in subsequent snapshotUnitMetricsByScope call", () => {
+    // 1. Write a ledger with 5 units to disk.
+    const oldUnits = [
+      makeUnit("M001/S01/T01", 1000),
+      makeUnit("M001/S01/T02", 2000),
+      makeUnit("M001/S01/T03", 3000),
+      makeUnit("M001/S01/T04", 4000),
+      makeUnit("M001/S01/T05", 5000),
+    ];
+    writeFileSync(metricsPath(tmpDir), JSON.stringify(makeProjectLedger(oldUnits), null, 2));
+
+    // 2. Load the scoped cache — this populates scopedLedgers with all 5 units.
+    initMetricsByScope(scope);
+
+    // 3. Prune to keepCount=2 — should evict 3 old units from disk AND clear scopedLedgers.
+    const removed = pruneMetricsLedger(tmpDir, 2);
+    assert.equal(removed, 3, "pruneMetricsLedger should report 3 removed units");
+
+    // 4. Snapshot a new unit via scope. This exercises the lazy-reload path
+    //    (scopedLedgers was cleared by prune) and writes the result to disk.
+    const ctx = assistantCtx();
+    const newUnit = snapshotUnitMetricsByScope(
+      scope,
+      ctx,
+      "execute-task",
+      "M001/S02/T01",
+      Date.now(),
+      "test-model",
+    );
+    assert.ok(newUnit !== null, "snapshotUnitMetricsByScope should return a unit");
+
+    // 5. Read the on-disk result and verify pruned units did NOT reappear.
+    const raw = readFileSync(metricsPath(tmpDir), "utf-8");
+    const ledger: MetricsLedger = JSON.parse(raw);
+
+    const unitIds = ledger.units.map((u) => u.id);
+
+    // The pruned units must not be in the output.
+    const prunedIds = ["M001/S01/T01", "M001/S01/T02", "M001/S01/T03"];
+    for (const id of prunedIds) {
+      assert.ok(
+        !unitIds.includes(id),
+        `pruned unit "${id}" must not reappear after prune + snapshot`,
+      );
+    }
+
+    // The 2 kept units and the new snapshot unit must be present.
+    assert.ok(unitIds.includes("M001/S01/T04"), "kept unit T04 should still be present");
+    assert.ok(unitIds.includes("M001/S01/T05"), "kept unit T05 should still be present");
+    assert.ok(unitIds.includes("M001/S02/T01"), "new snapshot unit should be present");
+  });
+});