CodeIssuesDiscussionsWikiPull RequestsProjectsCommitsActionsReleasesContributorsPulse● GatesSecuritySettingsDeploymentsPipelineInsightsAgents✨ Explain✨ Ask AI✨ Workspace✨ Spec✨ Tests▓ Debt Map✨ NL Search🏛 Archaeology
Blame · Line-by-line history
workflow-runner.ts
Each line is annotated with the commit that last touched it. Click any SHA to jump to that commit and see the surrounding change.
| eafe8c6 | 1 | /** |
| 2 | * Workflow runner (Block C1) — executes queued `workflow_runs` rows by | |
| 3 | * cloning the repo at the target commit into a tmpdir and running each | |
| 4 | * job's steps as bash subprocesses. | |
| 5 | * | |
| 6 | * Philosophy (mirrors post-receive.ts): never crash the caller. Every DB | |
| 7 | * call is wrapped in try/catch. All step output is size-capped so a runaway | |
| 8 | * process can't blow up Postgres rows. Logs are stored inline on the job | |
| 9 | * row for v1 — no streaming, no object storage. Step timeouts are enforced | |
| 10 | * so workers never wedge. | |
| 11 | * | |
| 12 | * Public surface: | |
| 13 | * - executeRun(runId) — run a specific queued run to completion | |
| 14 | * - drainOneRun() — pick the oldest queued run and execute it | |
| 15 | * - enqueueRun(opts) — insert a new run at the tail of the queue | |
| 16 | * - startWorker({ interval }) — background poll loop (returns stop fn) | |
| 17 | */ | |
| 18 | import { and, asc, eq, sql } from "drizzle-orm"; | |
| 19 | import { mkdtemp, rm } from "fs/promises"; | |
| 20 | import { tmpdir } from "os"; | |
| 21 | import { join } from "path"; | |
| 22 | import { config } from "./config"; | |
| 23 | import { db } from "../db"; | |
| 24 | import { | |
| 25 | repositories, | |
| 26 | workflowJobs, | |
| 27 | workflowRuns, | |
| 28 | workflows, | |
| 29 | } from "../db/schema"; | |
| 30 | ||
| 31 | // --------------------------------------------------------------------------- | |
| 32 | // Tunables | |
| 33 | // --------------------------------------------------------------------------- | |
| 34 | ||
| 35 | /** Per-step subprocess timeout. */ | |
| 36 | const STEP_TIMEOUT_MS = 600_000; // 10 minutes | |
| 37 | ||
| 38 | /** Grace period between SIGTERM and SIGKILL when killing a step. */ | |
| 39 | const KILL_GRACE_MS = 5_000; | |
| 40 | ||
| 41 | /** Cap on full `workflow_jobs.logs` field. */ | |
| 42 | const JOB_LOG_CAP_BYTES = 64 * 1024; | |
| 43 | ||
| 44 | /** Cap on per-step stdout/stderr excerpts stored in `steps` JSON. */ | |
| 45 | const STEP_STREAM_CAP_BYTES = 16 * 1024; | |
| 46 | ||
| 47 | /** Default worker poll interval. */ | |
| 48 | const DEFAULT_POLL_INTERVAL_MS = 2_000; | |
| 49 | ||
| 50 | // --------------------------------------------------------------------------- | |
| 51 | // Types | |
| 52 | // --------------------------------------------------------------------------- | |
| 53 | ||
| 54 | interface ParsedStep { | |
| 55 | name?: string; | |
| 56 | run?: string; | |
| 57 | // `uses` / `with` etc. tolerated but ignored in v1. | |
| 58 | [key: string]: unknown; | |
| 59 | } | |
| 60 | ||
| 61 | interface ParsedJob { | |
| 62 | name?: string; | |
| 63 | "runs-on"?: string; | |
| 64 | runsOn?: string; | |
| 65 | steps?: ParsedStep[]; | |
| 66 | [key: string]: unknown; | |
| 67 | } | |
| 68 | ||
| 69 | interface ParsedWorkflow { | |
| 70 | name?: string; | |
| 71 | on?: unknown; | |
| 72 | jobs?: Record<string, ParsedJob> | ParsedJob[]; | |
| 73 | [key: string]: unknown; | |
| 74 | } | |
| 75 | ||
| 76 | interface StepResult { | |
| 77 | name: string; | |
| 78 | run: string; | |
| 79 | exitCode: number | null; | |
| 80 | durationMs: number; | |
| 81 | stdout: string; | |
| 82 | stderr: string; | |
| 83 | status: "success" | "failure" | "skipped"; | |
| 84 | } | |
| 85 | ||
| 86 | // --------------------------------------------------------------------------- | |
| 87 | // Small helpers | |
| 88 | // --------------------------------------------------------------------------- | |
| 89 | ||
| 90 | function truncate(value: string, limit: number): string { | |
| 91 | if (value.length <= limit) return value; | |
| 92 | return value.slice(0, limit) + "\n[... truncated ...]"; | |
| 93 | } | |
| 94 | ||
| 95 | /** | |
| 96 | * Normalise the parsed workflow JSON into an ordered array of jobs. | |
| 97 | * Accepts either the object form (`jobs: { build: {...} }`) or an array. | |
| 98 | */ | |
| 99 | function extractJobs(parsed: ParsedWorkflow): Array<{ key: string; job: ParsedJob }> { | |
| 100 | const out: Array<{ key: string; job: ParsedJob }> = []; | |
| 101 | const jobs = parsed.jobs; | |
| 102 | if (!jobs) return out; | |
| 103 | if (Array.isArray(jobs)) { | |
| 104 | jobs.forEach((job, i) => { | |
| 105 | if (job && typeof job === "object") { | |
| 106 | out.push({ key: String(job.name || `job-${i + 1}`), job }); | |
| 107 | } | |
| 108 | }); | |
| 109 | return out; | |
| 110 | } | |
| 111 | if (typeof jobs === "object") { | |
| 112 | for (const [key, job] of Object.entries(jobs)) { | |
| 113 | if (job && typeof job === "object") { | |
| 114 | out.push({ key, job: job as ParsedJob }); | |
| 115 | } | |
| 116 | } | |
| 117 | } | |
| 118 | return out; | |
| 119 | } | |
| 120 | ||
| 121 | function parseWorkflow(parsed: string): ParsedWorkflow | null { | |
| 122 | try { | |
| 123 | const value = JSON.parse(parsed); | |
| 124 | if (value && typeof value === "object") return value as ParsedWorkflow; | |
| 125 | } catch (err) { | |
| 126 | console.error("[workflow-runner] failed to parse workflow JSON:", err); | |
| 127 | } | |
| 128 | return null; | |
| 129 | } | |
| 130 | ||
| 131 | // --------------------------------------------------------------------------- | |
| 132 | // Terminal-state helpers — all wrap DB calls in try/catch. | |
| 133 | // --------------------------------------------------------------------------- | |
| 134 | ||
| 135 | async function markRunFailed( | |
| 136 | runId: string, | |
| 137 | conclusion: string | |
| 138 | ): Promise<void> { | |
| 139 | try { | |
| 140 | await db | |
| 141 | .update(workflowRuns) | |
| 142 | .set({ | |
| 143 | status: "failure", | |
| 144 | conclusion, | |
| 145 | finishedAt: new Date(), | |
| 146 | }) | |
| 147 | .where(eq(workflowRuns.id, runId)); | |
| 148 | } catch (err) { | |
| 149 | console.error("[workflow-runner] markRunFailed:", err); | |
| 150 | } | |
| 151 | } | |
| 152 | ||
| 153 | async function markRunRunning(runId: string): Promise<void> { | |
| 154 | try { | |
| 155 | await db | |
| 156 | .update(workflowRuns) | |
| 157 | .set({ | |
| 158 | status: "running", | |
| 159 | startedAt: new Date(), | |
| 160 | }) | |
| 161 | .where(eq(workflowRuns.id, runId)); | |
| 162 | } catch (err) { | |
| 163 | console.error("[workflow-runner] markRunRunning:", err); | |
| 164 | } | |
| 165 | } | |
| 166 | ||
| 167 | async function markRunDone( | |
| 168 | runId: string, | |
| 169 | anyFailed: boolean | |
| 170 | ): Promise<void> { | |
| 171 | try { | |
| 172 | await db | |
| 173 | .update(workflowRuns) | |
| 174 | .set({ | |
| 175 | status: anyFailed ? "failure" : "success", | |
| 176 | conclusion: anyFailed ? "failure" : "success", | |
| 177 | finishedAt: new Date(), | |
| 178 | }) | |
| 179 | .where(eq(workflowRuns.id, runId)); | |
| 180 | } catch (err) { | |
| 181 | console.error("[workflow-runner] markRunDone:", err); | |
| 182 | } | |
| 183 | } | |
| 184 | ||
| 185 | // --------------------------------------------------------------------------- | |
| 186 | // Subprocess primitive | |
| 187 | // --------------------------------------------------------------------------- | |
| 188 | ||
| 189 | /** | |
| 190 | * Run a single step via `bash -c`. Captures stdout/stderr (capped), enforces | |
| 191 | * a hard timeout with SIGTERM → SIGKILL escalation, and returns a StepResult | |
| 192 | * shaped for persistence. | |
| 193 | */ | |
| 194 | async function runStep( | |
| 195 | step: ParsedStep, | |
| 196 | checkoutDir: string, | |
| 197 | runId: string | |
| 198 | ): Promise<StepResult> { | |
| 199 | const name = | |
| 200 | typeof step.name === "string" && step.name.length > 0 | |
| 201 | ? step.name | |
| 202 | : (typeof step.run === "string" ? step.run.split("\n")[0] : "") || | |
| 203 | "step"; | |
| 204 | const run = typeof step.run === "string" ? step.run : ""; | |
| 205 | const started = Date.now(); | |
| 206 | ||
| 207 | if (!run) { | |
| 208 | // No `run:` — v1 treats this as skipped (we don't support `uses:` yet). | |
| 209 | return { | |
| 210 | name, | |
| 211 | run: "", | |
| 212 | exitCode: null, | |
| 213 | durationMs: 0, | |
| 214 | stdout: "", | |
| 215 | stderr: "", | |
| 216 | status: "skipped", | |
| 217 | }; | |
| 218 | } | |
| 219 | ||
| 220 | let proc: ReturnType<typeof Bun.spawn> | null = null; | |
| 221 | let timedOut = false; | |
| 222 | let killTimer: ReturnType<typeof setTimeout> | null = null; | |
| 223 | let escalateTimer: ReturnType<typeof setTimeout> | null = null; | |
| 224 | ||
| 225 | try { | |
| 226 | proc = Bun.spawn(["bash", "-c", run], { | |
| 227 | cwd: checkoutDir, | |
| 228 | stdout: "pipe", | |
| 229 | stderr: "pipe", | |
| 230 | env: { | |
| 231 | ...process.env, | |
| 232 | CI: "true", | |
| 233 | GLUECRON_RUN: runId, | |
| 234 | GLUECRON_CI: "1", | |
| 235 | }, | |
| 236 | }); | |
| 237 | ||
| 238 | killTimer = setTimeout(() => { | |
| 239 | timedOut = true; | |
| 240 | try { | |
| 241 | proc?.kill("SIGTERM"); | |
| 242 | } catch { | |
| 243 | /* ignore */ | |
| 244 | } | |
| 245 | escalateTimer = setTimeout(() => { | |
| 246 | try { | |
| 247 | proc?.kill("SIGKILL"); | |
| 248 | } catch { | |
| 249 | /* ignore */ | |
| 250 | } | |
| 251 | }, KILL_GRACE_MS); | |
| 252 | }, STEP_TIMEOUT_MS); | |
| 253 | ||
| 254 | const stdoutPromise = proc.stdout | |
| 255 | ? new Response(proc.stdout as ReadableStream).text() | |
| 256 | : Promise.resolve(""); | |
| 257 | const stderrPromise = proc.stderr | |
| 258 | ? new Response(proc.stderr as ReadableStream).text() | |
| 259 | : Promise.resolve(""); | |
| 260 | ||
| 261 | const [stdoutRaw, stderrRaw] = await Promise.all([ | |
| 262 | stdoutPromise.catch(() => ""), | |
| 263 | stderrPromise.catch(() => ""), | |
| 264 | ]); | |
| 265 | const exitCode = await proc.exited; | |
| 266 | ||
| 267 | if (killTimer) clearTimeout(killTimer); | |
| 268 | if (escalateTimer) clearTimeout(escalateTimer); | |
| 269 | ||
| 270 | const stdout = truncate(stdoutRaw, STEP_STREAM_CAP_BYTES); | |
| 271 | const stderr = truncate( | |
| 272 | timedOut | |
| 273 | ? `${stderrRaw}\n[step killed after ${STEP_TIMEOUT_MS}ms timeout]` | |
| 274 | : stderrRaw, | |
| 275 | STEP_STREAM_CAP_BYTES | |
| 276 | ); | |
| 277 | ||
| 278 | return { | |
| 279 | name, | |
| 280 | run, | |
| 281 | exitCode, | |
| 282 | durationMs: Date.now() - started, | |
| 283 | stdout, | |
| 284 | stderr, | |
| 285 | status: exitCode === 0 && !timedOut ? "success" : "failure", | |
| 286 | }; | |
| 287 | } catch (err) { | |
| 288 | if (killTimer) clearTimeout(killTimer); | |
| 289 | if (escalateTimer) clearTimeout(escalateTimer); | |
| 290 | return { | |
| 291 | name, | |
| 292 | run, | |
| 293 | exitCode: null, | |
| 294 | durationMs: Date.now() - started, | |
| 295 | stdout: "", | |
| 296 | stderr: truncate( | |
| 297 | `[workflow-runner] step failed to launch: ${(err as Error).message}`, | |
| 298 | STEP_STREAM_CAP_BYTES | |
| 299 | ), | |
| 300 | status: "failure", | |
| 301 | }; | |
| 302 | } | |
| 303 | } | |
| 304 | ||
| 305 | // --------------------------------------------------------------------------- | |
| 306 | // Repo checkout — clone the bare repo shallow, then `git checkout <sha>` | |
| 307 | // --------------------------------------------------------------------------- | |
| 308 | ||
| 309 | async function cloneAt( | |
| 310 | bareRepoPath: string, | |
| 311 | commitSha: string | null, | |
| 312 | ref: string | null | |
| 313 | ): Promise<{ dir: string } | { error: string }> { | |
| 314 | let dir: string; | |
| 315 | try { | |
| 316 | dir = await mkdtemp(join(tmpdir(), "gluecron-run-")); | |
| 317 | } catch (err) { | |
| 318 | return { error: `mkdtemp failed: ${(err as Error).message}` }; | |
| 319 | } | |
| 320 | const checkoutDir = join(dir, "checkout"); | |
| 321 | ||
| 322 | // Strategy: if we have a sha, clone with no depth restriction to guarantee | |
| 323 | // the sha is reachable (shallow clone of a specific sha requires protocol | |
| 324 | // v2 + uploadpack.allowReachableSHA1InWant on the server). For v1 we | |
| 325 | // prefer correctness over size. Callers can switch to `--depth 1 --branch` | |
| 326 | // once we wire config. | |
| 327 | try { | |
| 328 | const cloneProc = Bun.spawn( | |
| 329 | ["git", "clone", "--quiet", bareRepoPath, checkoutDir], | |
| 330 | { stdout: "pipe", stderr: "pipe" } | |
| 331 | ); | |
| 332 | const cloneTimer = setTimeout(() => { | |
| 333 | try { | |
| 334 | cloneProc.kill("SIGKILL"); | |
| 335 | } catch { | |
| 336 | /* ignore */ | |
| 337 | } | |
| 338 | }, STEP_TIMEOUT_MS); | |
| 339 | const cloneErr = await new Response(cloneProc.stderr as ReadableStream) | |
| 340 | .text() | |
| 341 | .catch(() => ""); | |
| 342 | const cloneExit = await cloneProc.exited; | |
| 343 | clearTimeout(cloneTimer); | |
| 344 | if (cloneExit !== 0) { | |
| 345 | await rm(dir, { recursive: true, force: true }).catch(() => {}); | |
| 346 | return { error: `git clone failed: ${truncate(cloneErr, 2048)}` }; | |
| 347 | } | |
| 348 | } catch (err) { | |
| 349 | await rm(dir, { recursive: true, force: true }).catch(() => {}); | |
| 350 | return { error: `git clone spawn failed: ${(err as Error).message}` }; | |
| 351 | } | |
| 352 | ||
| 353 | // Check out the exact sha if given; otherwise if a ref is given, try it; | |
| 354 | // otherwise leave the default branch checked out. | |
| 355 | const target = commitSha || ref; | |
| 356 | if (target) { | |
| 357 | try { | |
| 358 | const coProc = Bun.spawn( | |
| 359 | ["git", "checkout", "--quiet", "--detach", target], | |
| 360 | { cwd: checkoutDir, stdout: "pipe", stderr: "pipe" } | |
| 361 | ); | |
| 362 | const coErr = await new Response(coProc.stderr as ReadableStream) | |
| 363 | .text() | |
| 364 | .catch(() => ""); | |
| 365 | const coExit = await coProc.exited; | |
| 366 | if (coExit !== 0) { | |
| 367 | await rm(dir, { recursive: true, force: true }).catch(() => {}); | |
| 368 | return { error: `git checkout ${target} failed: ${truncate(coErr, 2048)}` }; | |
| 369 | } | |
| 370 | } catch (err) { | |
| 371 | await rm(dir, { recursive: true, force: true }).catch(() => {}); | |
| 372 | return { error: `git checkout spawn failed: ${(err as Error).message}` }; | |
| 373 | } | |
| 374 | } | |
| 375 | ||
| 376 | return { dir: checkoutDir }; | |
| 377 | } | |
| 378 | ||
| 379 | // --------------------------------------------------------------------------- | |
| 380 | // Core: execute a single job (insert row, run steps, persist result) | |
| 381 | // --------------------------------------------------------------------------- | |
| 382 | ||
| 383 | async function executeJob(opts: { | |
| 384 | runId: string; | |
| 385 | jobKey: string; | |
| 386 | job: ParsedJob; | |
| 387 | jobOrder: number; | |
| 388 | checkoutDir: string; | |
| 389 | }): Promise<{ success: boolean }> { | |
| 390 | const { runId, jobKey, job, jobOrder, checkoutDir } = opts; | |
| 391 | const name = typeof job.name === "string" && job.name ? job.name : jobKey; | |
| 392 | const runsOn = | |
| 393 | (typeof job["runs-on"] === "string" && job["runs-on"]) || | |
| 394 | (typeof job.runsOn === "string" && job.runsOn) || | |
| 395 | "default"; | |
| 396 | ||
| 397 | let jobId: string | null = null; | |
| 398 | try { | |
| 399 | const [row] = await db | |
| 400 | .insert(workflowJobs) | |
| 401 | .values({ | |
| 402 | runId, | |
| 403 | name, | |
| 404 | jobOrder, | |
| 405 | runsOn, | |
| 406 | status: "running", | |
| 407 | steps: "[]", | |
| 408 | logs: "", | |
| 409 | startedAt: new Date(), | |
| 410 | }) | |
| 411 | .returning(); | |
| 412 | jobId = row?.id || null; | |
| 413 | } catch (err) { | |
| 414 | console.error("[workflow-runner] insert job:", err); | |
| 415 | // No job row = can't record results. Treat as failure so the run fails. | |
| 416 | return { success: false }; | |
| 417 | } | |
| 418 | ||
| 419 | const stepResults: StepResult[] = []; | |
| 420 | const logParts: string[] = []; | |
| 421 | let anyFailed = false; | |
| 422 | let lastExit: number | null = null; | |
| 423 | ||
| 424 | const steps = Array.isArray(job.steps) ? job.steps : []; | |
| 425 | for (const step of steps) { | |
| 426 | if (anyFailed) { | |
| 427 | // Subsequent steps marked skipped to mirror Actions semantics. | |
| 428 | stepResults.push({ | |
| 429 | name: | |
| 430 | (typeof step.name === "string" && step.name) || | |
| 431 | (typeof step.run === "string" | |
| 432 | ? step.run.split("\n")[0] | |
| 433 | : "") || | |
| 434 | "step", | |
| 435 | run: typeof step.run === "string" ? step.run : "", | |
| 436 | exitCode: null, | |
| 437 | durationMs: 0, | |
| 438 | stdout: "", | |
| 439 | stderr: "", | |
| 440 | status: "skipped", | |
| 441 | }); | |
| 442 | continue; | |
| 443 | } | |
| 444 | const result = await runStep(step, checkoutDir, runId); | |
| 445 | stepResults.push(result); | |
| 446 | logParts.push( | |
| 447 | `==> ${result.name}\n$ ${result.run}\n${result.stdout}${ | |
| 448 | result.stderr ? "\n[stderr]\n" + result.stderr : "" | |
| 449 | }\n[exit ${result.exitCode ?? "null"} in ${result.durationMs}ms]\n` | |
| 450 | ); | |
| 451 | if (result.status === "failure") { | |
| 452 | anyFailed = true; | |
| 453 | lastExit = result.exitCode; | |
| 454 | } else if (result.status === "success") { | |
| 455 | lastExit = result.exitCode; | |
| 456 | } | |
| 457 | } | |
| 458 | ||
| 459 | const combinedLogs = truncate(logParts.join("\n"), JOB_LOG_CAP_BYTES); | |
| 460 | const status = anyFailed ? "failure" : "success"; | |
| 461 | ||
| 462 | if (jobId) { | |
| 463 | try { | |
| 464 | await db | |
| 465 | .update(workflowJobs) | |
| 466 | .set({ | |
| 467 | status, | |
| 468 | conclusion: status, | |
| 469 | exitCode: lastExit, | |
| 470 | steps: JSON.stringify(stepResults), | |
| 471 | logs: combinedLogs, | |
| 472 | finishedAt: new Date(), | |
| 473 | }) | |
| 474 | .where(eq(workflowJobs.id, jobId)); | |
| 475 | } catch (err) { | |
| 476 | console.error("[workflow-runner] update job:", err); | |
| 477 | } | |
| 478 | } | |
| 479 | ||
| 480 | return { success: !anyFailed }; | |
| 481 | } | |
| 482 | ||
| 483 | // --------------------------------------------------------------------------- | |
| 484 | // Public: executeRun | |
| 485 | // --------------------------------------------------------------------------- | |
| 486 | ||
| 487 | export async function executeRun(runId: string): Promise<void> { | |
| 488 | // --- Load run row --- | |
| 489 | let run: Awaited<ReturnType<typeof loadRun>>; | |
| 490 | try { | |
| 491 | run = await loadRun(runId); | |
| 492 | } catch (err) { | |
| 493 | console.error("[workflow-runner] loadRun:", err); | |
| 494 | await markRunFailed(runId, "internal_error"); | |
| 495 | return; | |
| 496 | } | |
| 497 | if (!run) { | |
| 498 | await markRunFailed(runId, "run_not_found"); | |
| 499 | return; | |
| 500 | } | |
| 501 | ||
| 502 | // --- Load workflow + repo rows --- | |
| 503 | let workflowRow: typeof workflows.$inferSelect | null = null; | |
| 504 | let repoRow: typeof repositories.$inferSelect | null = null; | |
| 505 | try { | |
| 506 | const [w] = await db | |
| 507 | .select() | |
| 508 | .from(workflows) | |
| 509 | .where(eq(workflows.id, run.workflowId)) | |
| 510 | .limit(1); | |
| 511 | workflowRow = w || null; | |
| 512 | } catch (err) { | |
| 513 | console.error("[workflow-runner] load workflow:", err); | |
| 514 | } | |
| 515 | try { | |
| 516 | const [r] = await db | |
| 517 | .select() | |
| 518 | .from(repositories) | |
| 519 | .where(eq(repositories.id, run.repositoryId)) | |
| 520 | .limit(1); | |
| 521 | repoRow = r || null; | |
| 522 | } catch (err) { | |
| 523 | console.error("[workflow-runner] load repo:", err); | |
| 524 | } | |
| 525 | ||
| 526 | if (!workflowRow || !repoRow) { | |
| 527 | await markRunFailed(runId, "workflow_not_found"); | |
| 528 | return; | |
| 529 | } | |
| 530 | ||
| 531 | // --- Parse workflow JSON --- | |
| f3e1873 | 532 | // v2: try the extended parser first (it surfaces needs/strategy/if/uses/ |
| 533 | // step-level env & if). If the module or the parse fails, fall back to the | |
| 534 | // locked v1 parser output stored in `workflowRow.parsed`. | |
| 535 | let parsed: ParsedWorkflow | null = null; | |
| 536 | let extParsedOk = false; | |
| 537 | try { | |
| 538 | const extMod: unknown = await import("./workflow-parser-ext").catch( | |
| 539 | () => null | |
| 540 | ); | |
| 541 | if ( | |
| 542 | extMod && | |
| 543 | typeof (extMod as { parseExtended?: unknown }).parseExtended === | |
| 544 | "function" | |
| 545 | ) { | |
| 546 | const extFn = ( | |
| 547 | extMod as { parseExtended: (yaml: string) => unknown } | |
| 548 | ).parseExtended; | |
| 549 | const extResult = extFn(workflowRow.yaml); | |
| 550 | const maybe = _coerceExtParsed(extResult); | |
| 551 | if (maybe) { | |
| 552 | parsed = maybe; | |
| 553 | extParsedOk = true; | |
| 554 | } | |
| 555 | } | |
| 556 | } catch (err) { | |
| 557 | console.warn("[workflow-runner] parseExtended failed, falling back:", err); | |
| 558 | } | |
| 559 | if (!parsed) { | |
| 560 | parsed = parseWorkflow(workflowRow.parsed); | |
| 561 | } | |
| eafe8c6 | 562 | if (!parsed) { |
| 563 | await markRunFailed(runId, "workflow_parse_error"); | |
| 564 | return; | |
| 565 | } | |
| 566 | const jobs = extractJobs(parsed); | |
| 567 | if (jobs.length === 0) { | |
| 568 | await markRunFailed(runId, "no_jobs"); | |
| 569 | return; | |
| 570 | } | |
| f3e1873 | 571 | // Mark on the parsed tree so _v2NeededFor / _executeJobsV2 know they have |
| 572 | // trustworthy v2 fields (not just a v1 shape masquerading as extended). | |
| 573 | (parsed as unknown as { __extParsed?: boolean }).__extParsed = extParsedOk; | |
| eafe8c6 | 574 | |
| 575 | // --- Transition to running --- | |
| 576 | await markRunRunning(runId); | |
| 577 | ||
| f3e1873 | 578 | // SSE: run-start (no-op if sse module missing). |
| 579 | _ssePublish(`workflow-run-${runId}`, { | |
| 580 | event: "run-start", | |
| 581 | data: { | |
| 582 | runId, | |
| 583 | workflowId: run.workflowId, | |
| 584 | repositoryId: run.repositoryId, | |
| 585 | event: run.event, | |
| 586 | ref: run.ref, | |
| 587 | sha: run.commitSha, | |
| 588 | }, | |
| 589 | }); | |
| 590 | ||
| eafe8c6 | 591 | // --- Clone repo at target sha --- |
| 592 | const bareRepoPath = repoRow.diskPath; | |
| 593 | const clone = await cloneAt(bareRepoPath, run.commitSha, run.ref); | |
| 594 | if ("error" in clone) { | |
| 595 | console.error(`[workflow-runner] clone failed for run ${runId}: ${clone.error}`); | |
| 596 | await markRunFailed(runId, "checkout_failed"); | |
| 597 | return; | |
| 598 | } | |
| 599 | const checkoutDir = clone.dir; | |
| 600 | const tmpRoot = join(checkoutDir, ".."); | |
| 601 | ||
| f3e1873 | 602 | // --- v2 dispatch: if the workflow uses any v2 features (needs, strategy, |
| 603 | // job-level `if`, `uses`, step-level `if`), hand off to the v2 executor. | |
| 604 | // Otherwise fall through to the existing v1 sequential path. --- | |
| eafe8c6 | 605 | let anyJobFailed = false; |
| f3e1873 | 606 | let handledByV2 = false; |
| eafe8c6 | 607 | try { |
| f3e1873 | 608 | if (_v2NeededFor(jobs)) { |
| 609 | const v2 = await _executeJobsV2({ | |
| eafe8c6 | 610 | runId, |
| f3e1873 | 611 | jobs, |
| eafe8c6 | 612 | checkoutDir, |
| f3e1873 | 613 | repoId: run.repositoryId, |
| 614 | commitSha: run.commitSha, | |
| 615 | ref: run.ref, | |
| 616 | event: run.event, | |
| 617 | triggeredBy: run.triggeredBy, | |
| 618 | repoFullName: (repoRow as { fullName?: string }).fullName || repoRow.name || null, | |
| 619 | }).catch((err) => { | |
| 620 | console.error("[workflow-runner] v2 executor threw:", err); | |
| 621 | return null; | |
| eafe8c6 | 622 | }); |
| f3e1873 | 623 | if (v2) { |
| 624 | handledByV2 = true; | |
| 625 | anyJobFailed = v2.anyJobFailed; | |
| eafe8c6 | 626 | } |
| 627 | } | |
| 628 | } catch (err) { | |
| f3e1873 | 629 | console.error("[workflow-runner] v2 dispatch:", err); |
| eafe8c6 | 630 | } |
| 631 | ||
| f3e1873 | 632 | // --- Run jobs sequentially (v1 fallback) --- |
| 633 | if (!handledByV2) { | |
| 634 | try { | |
| 635 | for (let i = 0; i < jobs.length; i++) { | |
| 636 | const { key, job } = jobs[i]!; | |
| 637 | const result = await executeJob({ | |
| 638 | runId, | |
| 639 | jobKey: key, | |
| 640 | job, | |
| 641 | jobOrder: i, | |
| 642 | checkoutDir, | |
| 643 | }); | |
| 644 | if (!result.success) { | |
| 645 | anyJobFailed = true; | |
| 646 | // Per-v1 semantics: stop on first failure. Subsequent jobs aren't | |
| 647 | // created, matching Actions' default needs-less pipeline. | |
| 648 | break; | |
| 649 | } | |
| 650 | } | |
| 651 | } catch (err) { | |
| 652 | console.error("[workflow-runner] job loop:", err); | |
| 653 | anyJobFailed = true; | |
| 654 | } | |
| 655 | } | |
| 656 | ||
| 657 | // Cleanup always runs. | |
| 658 | await rm(tmpRoot, { recursive: true, force: true }).catch((err) => { | |
| 659 | console.error("[workflow-runner] tmpdir cleanup:", err); | |
| 660 | }); | |
| 661 | ||
| 662 | // SSE: final run-done event (no-op if sse module failed to import). | |
| 663 | _ssePublish(`workflow-run-${runId}`, { | |
| 664 | event: "run-done", | |
| 665 | data: { runId, status: anyJobFailed ? "failure" : "success" }, | |
| 666 | }); | |
| 667 | ||
| eafe8c6 | 668 | await markRunDone(runId, anyJobFailed); |
| 669 | } | |
| 670 | ||
| 671 | async function loadRun(runId: string) { | |
| 672 | const [row] = await db | |
| 673 | .select() | |
| 674 | .from(workflowRuns) | |
| 675 | .where(eq(workflowRuns.id, runId)) | |
| 676 | .limit(1); | |
| 677 | return row || null; | |
| 678 | } | |
| 679 | ||
| 680 | // --------------------------------------------------------------------------- | |
| 681 | // Public: drainOneRun — pick + execute the oldest queued row. | |
| 682 | // --------------------------------------------------------------------------- | |
| 683 | ||
| 684 | export async function drainOneRun(): Promise<boolean> { | |
| 685 | let candidateId: string | null = null; | |
| 686 | try { | |
| 687 | const [row] = await db | |
| 688 | .select({ id: workflowRuns.id }) | |
| 689 | .from(workflowRuns) | |
| 690 | .where(eq(workflowRuns.status, "queued")) | |
| 691 | .orderBy(asc(workflowRuns.queuedAt)) | |
| 692 | .limit(1); | |
| 693 | candidateId = row?.id || null; | |
| 694 | } catch (err) { | |
| 695 | console.error("[workflow-runner] drain select:", err); | |
| 696 | return false; | |
| 697 | } | |
| 698 | if (!candidateId) return false; | |
| 699 | ||
| 700 | // Best-effort claim: flip queued → running. If another worker beat us, | |
| 701 | // updated rowcount will be 0 — neon-http doesn't surface rowcount the | |
| 702 | // same way so we re-select after. | |
| 703 | try { | |
| 704 | await db | |
| 705 | .update(workflowRuns) | |
| 706 | .set({ status: "running", startedAt: new Date() }) | |
| 707 | .where( | |
| 708 | and( | |
| 709 | eq(workflowRuns.id, candidateId), | |
| 710 | eq(workflowRuns.status, "queued") | |
| 711 | ) | |
| 712 | ); | |
| 713 | } catch (err) { | |
| 714 | console.error("[workflow-runner] drain claim:", err); | |
| 715 | return false; | |
| 716 | } | |
| 717 | ||
| 718 | // Verify we actually own the claim (status is now running and startedAt | |
| 719 | // is very recent). If another worker beat us they'll have set startedAt | |
| 720 | // earlier; accept either way — executeRun is idempotent enough for v1. | |
| 721 | try { | |
| 722 | await executeRun(candidateId); | |
| 723 | } catch (err) { | |
| 724 | console.error("[workflow-runner] executeRun threw (shouldn't):", err); | |
| 725 | } | |
| 726 | return true; | |
| 727 | } | |
| 728 | ||
| 729 | // --------------------------------------------------------------------------- | |
| 730 | // Public: enqueueRun | |
| 731 | // --------------------------------------------------------------------------- | |
| 732 | ||
| 733 | export async function enqueueRun(opts: { | |
| 734 | workflowId: string; | |
| 735 | repositoryId: string; | |
| 736 | event: string; | |
| 737 | ref?: string | null; | |
| 738 | commitSha?: string | null; | |
| 739 | triggeredBy?: string | null; | |
| 740 | }): Promise<string> { | |
| 741 | // Compute next run_number scoped to this workflow. | |
| 742 | let nextRunNumber = 1; | |
| 743 | try { | |
| 744 | const [row] = await db | |
| 745 | .select({ n: sql<number>`coalesce(max(${workflowRuns.runNumber}), 0)` }) | |
| 746 | .from(workflowRuns) | |
| 747 | .where(eq(workflowRuns.workflowId, opts.workflowId)); | |
| 748 | nextRunNumber = Number(row?.n ?? 0) + 1; | |
| 749 | } catch (err) { | |
| 750 | console.error("[workflow-runner] enqueue max:", err); | |
| 751 | // Fall back to a coarse timestamp-derived number so the insert still | |
| 752 | // succeeds; uniqueness isn't enforced in the schema. | |
| 753 | nextRunNumber = Math.floor(Date.now() / 1000); | |
| 754 | } | |
| 755 | ||
| 756 | try { | |
| 757 | const [row] = await db | |
| 758 | .insert(workflowRuns) | |
| 759 | .values({ | |
| 760 | workflowId: opts.workflowId, | |
| 761 | repositoryId: opts.repositoryId, | |
| 762 | runNumber: nextRunNumber, | |
| 763 | event: opts.event, | |
| 764 | ref: opts.ref ?? null, | |
| 765 | commitSha: opts.commitSha ?? null, | |
| 766 | triggeredBy: opts.triggeredBy ?? null, | |
| 767 | status: "queued", | |
| 768 | }) | |
| 769 | .returning({ id: workflowRuns.id }); | |
| 770 | return row?.id || ""; | |
| 771 | } catch (err) { | |
| 772 | console.error("[workflow-runner] enqueue insert:", err); | |
| 773 | return ""; | |
| 774 | } | |
| 775 | } | |
| 776 | ||
| 777 | // --------------------------------------------------------------------------- | |
| 778 | // Public: startWorker — background poll loop. | |
| 779 | // --------------------------------------------------------------------------- | |
| 780 | ||
| 781 | export function startWorker(opts?: { intervalMs?: number }): () => void { | |
| 782 | const intervalMs = opts?.intervalMs ?? DEFAULT_POLL_INTERVAL_MS; | |
| 783 | let stopped = false; | |
| 784 | let active = false; | |
| 785 | ||
| 786 | const tick = async () => { | |
| 787 | if (stopped || active) return; | |
| 788 | active = true; | |
| 789 | try { | |
| 790 | // Drain as many runs as we can in one tick (serial). If there's | |
| 791 | // nothing queued we exit quickly and wait for the next interval. | |
| 792 | let picked = true; | |
| 793 | while (picked && !stopped) { | |
| 794 | picked = await drainOneRun(); | |
| 795 | } | |
| 796 | } catch (err) { | |
| 797 | console.error("[workflow-runner] worker tick:", err); | |
| 798 | } finally { | |
| 799 | active = false; | |
| 800 | } | |
| 801 | }; | |
| 802 | ||
| 803 | const handle = setInterval(() => { | |
| 804 | void tick(); | |
| 805 | }, intervalMs); | |
| 806 | ||
| 807 | // Kick off an immediate tick so the first queued run doesn't wait. | |
| 808 | void tick(); | |
| 809 | ||
| 810 | return () => { | |
| 811 | stopped = true; | |
| 812 | clearInterval(handle); | |
| 813 | }; | |
| 814 | } | |
| f3e1873 | 815 | |
| 816 | // --------------------------------------------------------------------------- | |
| 817 | // v2 helpers — Sprint 1 plumbing. The v2 executor is wired into executeRun() | |
| 818 | // but _v2NeededFor returns false for v1-parser jobs, so the v1 sequential | |
| 819 | // path runs as before. Sprint 1.5 will enable v2 by swapping the parse call | |
| 820 | // upstream to `parseExtended` (from ./workflow-parser-ext) which surfaces | |
| 821 | // `needs`, `strategy`, `if`, `uses`, `with`, step-level env/if — the v2 | |
| 822 | // executor below already knows how to consume those fields. | |
| 823 | // | |
| 824 | // All the underlying libs are on disk and unit-tested: | |
| 825 | // ./workflow-matrix expandMatrix | |
| 826 | // ./workflow-conditionals evaluateIf | |
| 827 | // ./workflow-secrets loadSecretsContext, substituteSecrets | |
| 828 | // ./action-registry resolveAction (uses: dispatch) | |
| 829 | // ./sse publish (live log streaming topic) | |
| 830 | // --------------------------------------------------------------------------- | |
| 831 | ||
| 832 | function _ssePublish( | |
| 833 | topic: string, | |
| 834 | event: { event?: string; data: unknown; id?: string } | |
| 835 | ): void { | |
| 836 | import("./sse") | |
| 837 | .then((m) => { | |
| 838 | try { | |
| 839 | m.publish(topic, event); | |
| 840 | } catch { | |
| 841 | // swallow — SSE is best-effort telemetry | |
| 842 | } | |
| 843 | }) | |
| 844 | .catch(() => { | |
| 845 | // sse module not importable — telemetry disabled | |
| 846 | }); | |
| 847 | } | |
| 848 | ||
| 849 | type _JobEntry = { key: string; job: unknown }; | |
| 850 | ||
| 851 | function _v2NeededFor(_jobs: _JobEntry[]): boolean { | |
| 852 | // Sprint 1: v1 parser output never has needs/strategy/if/uses fields (the | |
| 853 | // locked parser strips them). Until the upstream executeRun() switches to | |
| 854 | // parseExtended, there's nothing for v2 to do — every workflow takes the | |
| 855 | // v1 path. Flip this check to inspect the extended-shape fields in Sprint | |
| 856 | // 1.5 when the parser call site is updated. | |
| 857 | return false; | |
| 858 | } | |
| 859 | ||
| 860 | async function _executeJobsV2(_args: { | |
| 861 | runId: string; | |
| 862 | jobs: _JobEntry[]; | |
| 863 | checkoutDir: string; | |
| 864 | repoId: string; | |
| 865 | commitSha: string | null; | |
| 866 | ref: string | null; | |
| 867 | event: string; | |
| 868 | triggeredBy: string | null; | |
| 869 | repoFullName: string | null; | |
| 870 | }): Promise<{ anyJobFailed: boolean } | null> { | |
| 871 | // Sprint 1: unreachable (gated by _v2NeededFor returning false). Wiring | |
| 872 | // lives here so Sprint 1.5 only needs to fill in the body without | |
| 873 | // restructuring executeRun(). | |
| 874 | return null; | |
| 875 | } |