lightflow-engine 0.2.12 → 0.2.13

package/BENCHMARKS.md

@@ -160,6 +160,7 @@ same Postgres (16, local) and the same harness (200 runs x 5 steps @ conc 20;
 | 0.2.8 | 455.2 | 2,276 | 8,695 |
 | 0.2.9 | 474.1 | 2,371 | 9,008 |
 | 0.2.10 | 455.3 | 2,277 | 9,368 |
+| 0.2.12 | 457.0 | 2,285 | 9,943 |
 
 ### What changed per version (all perf-neutral by design)
 
@@ -209,3 +210,13 @@ ln -s <path-to>/lightflow/node_modules <dir>/node_modules
 PKG=<dir> LIGHTFLOW_PG_URL=postgres://... BENCH_N=200 RN=5 npx tsx bench-pkg.mts
 PKG=<dir> LIGHTFLOW_PG_URL=postgres://... BENCH_N=50 RN=400 npx tsx bench-pkg.mts
 ```
+
+### 0.2.12 — bug-fix release (cancel/reaper/chunk)
+
+- **cancel is terminal**: `cancel()` now writes `status='cancelled'` + flag atomically; `finishRun` refuses to overwrite a cancelled run (was: racing replay resurrected the run as completed/failed). `RunStatus` gains `"cancelled"`.
+- **Lease-aware stale reaper**: `staleRuns` only reaps runs whose lease expired — a run inside a long step is no longer double-executed.
+- **Stream close marker**: `close()` uses a monotonic chunk index (was `ctx.writes`, which could collide and silently drop the done marker → hanging stream).
+- **O(1) nextChunkIndex**: `lightflow_runs.chunk_count` counter replaces a full chunk scan per write — long-run steps/s 9,368 → ~9,900.
+- Tidy: single `getCurrent()` per call site (was double call + `!`).
+
+Short: 200×5 conc 20 → **457 runs/s / 2,285 steps/s** (within noise of 0.2.9/0.2.10). Long: 50×400 → **9,943 steps/s** (best recorded).

package/dist/index.d.ts

@@ -49,6 +49,7 @@ export interface Store {
     isCancelled?(runId: string): Promise<boolean>;
     /** Next monotonic chunk index (max+1), atomic per run. */
     nextChunkIndex?(runId: string): Promise<number>;
+    pruneEvents?(runId: string, beforeSeq: number): Promise<void>;
     /** Run lease: claim (atomically) / release before executing a replay. */
     claimRun?(runId: string, leaseMs?: number): Promise<{
         ok: boolean;

package/dist/index.js

@@ -161,6 +161,10 @@ async function maybeSnapshot(ctx) {
         runId: ctx.runId, seq: ctx.seq, type: "snapshot",
         payload: { memo }, createdAt: ctx.now(),
     });
+    // Folded events below the snapshot are dead weight — drop them so the log
+    // stays O(recent activity), not O(total steps). Chunks are kept (stream
+    // replay reads them); run_completed (seq 0) is protected by beforeSeq > 0.
+    await ctx.store.pruneEvents?.(ctx.runId, ctx.seq - 1);
 }
 /** Durable sleep. Accepts milliseconds or an absolute Date. */
 export async function sleep(until) {

package/dist/pg-store.js

@@ -300,6 +300,15 @@ export async function createPostgresStore(url, opts = {}) {
                 payload: { ...(r.rows[0].payload ?? {}), resolved: Boolean(r.rows[0].resolved) },
             };
         },
+        async pruneEvents(runId, beforeSeq) {
+            // Remove events fully folded into the latest snapshot. Chunk events are
+            // KEPT: getReadable/getTailIndex stream chunks from the event log, and
+            // the snapshot memo is not a chunk source. run_completed (seq 0) is
+            // protected by beforeSeq > 0.
+            await pool.query(`DELETE FROM lightflow_events
+         WHERE run_id=$1 AND seq < $2
+           AND type IN ('step_completed','step_failed','sleep_created','sleep_completed')`, [runId, beforeSeq]);
+        },
         async setStatus(runId, status, output) {
             await pool.query(`UPDATE lightflow_runs SET status=$2, output=$3, updated_at=$4 WHERE run_id=$1`, [runId, status, output === undefined ? null : JSON.stringify(output), Date.now()]);
         },

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "lightflow-engine",
-  "version": "0.2.12",
+  "version": "0.2.13",
   "description": "A tiny durable workflow engine for Node.js and Postgres — with a Vercel Workflow drop-in compat layer.",
   "license": "MIT",
   "type": "module",