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"; | |
| 18b210d | 30 | import { |
| 31 | loadSecretsContext, | |
| 32 | substituteSecrets, | |
| 33 | } from "./workflow-secrets"; | |
| eafe8c6 | 34 | |
| 35 | // --------------------------------------------------------------------------- | |
| 36 | // Tunables | |
| 37 | // --------------------------------------------------------------------------- | |
| 38 | ||
| 39 | /** Per-step subprocess timeout. */ | |
| 40 | const STEP_TIMEOUT_MS = 600_000; // 10 minutes | |
| 41 | ||
| 42 | /** Grace period between SIGTERM and SIGKILL when killing a step. */ | |
| 43 | const KILL_GRACE_MS = 5_000; | |
| 44 | ||
| 45 | /** Cap on full `workflow_jobs.logs` field. */ | |
| 46 | const JOB_LOG_CAP_BYTES = 64 * 1024; | |
| 47 | ||
| 48 | /** Cap on per-step stdout/stderr excerpts stored in `steps` JSON. */ | |
| 49 | const STEP_STREAM_CAP_BYTES = 16 * 1024; | |
| 50 | ||
| 51 | /** Default worker poll interval. */ | |
| 52 | const DEFAULT_POLL_INTERVAL_MS = 2_000; | |
| 53 | ||
| 54 | // --------------------------------------------------------------------------- | |
| 55 | // Types | |
| 56 | // --------------------------------------------------------------------------- | |
| 57 | ||
| 58 | interface ParsedStep { | |
| 59 | name?: string; | |
| 60 | run?: string; | |
| 61 | // `uses` / `with` etc. tolerated but ignored in v1. | |
| 62 | [key: string]: unknown; | |
| 63 | } | |
| 64 | ||
| 65 | interface ParsedJob { | |
| 66 | name?: string; | |
| 67 | "runs-on"?: string; | |
| 68 | runsOn?: string; | |
| 69 | steps?: ParsedStep[]; | |
| 70 | [key: string]: unknown; | |
| 71 | } | |
| 72 | ||
| 73 | interface ParsedWorkflow { | |
| 74 | name?: string; | |
| 75 | on?: unknown; | |
| 76 | jobs?: Record<string, ParsedJob> | ParsedJob[]; | |
| 77 | [key: string]: unknown; | |
| 78 | } | |
| 79 | ||
| 80 | interface StepResult { | |
| 81 | name: string; | |
| 82 | run: string; | |
| 83 | exitCode: number | null; | |
| 84 | durationMs: number; | |
| 85 | stdout: string; | |
| 86 | stderr: string; | |
| 87 | status: "success" | "failure" | "skipped"; | |
| 88 | } | |
| 89 | ||
| 90 | // --------------------------------------------------------------------------- | |
| 91 | // Small helpers | |
| 92 | // --------------------------------------------------------------------------- | |
| 93 | ||
| 94 | function truncate(value: string, limit: number): string { | |
| 95 | if (value.length <= limit) return value; | |
| 96 | return value.slice(0, limit) + "\n[... truncated ...]"; | |
| 97 | } | |
| 98 | ||
| 99 | /** | |
| 100 | * Normalise the parsed workflow JSON into an ordered array of jobs. | |
| 101 | * Accepts either the object form (`jobs: { build: {...} }`) or an array. | |
| 102 | */ | |
| 103 | function extractJobs(parsed: ParsedWorkflow): Array<{ key: string; job: ParsedJob }> { | |
| 104 | const out: Array<{ key: string; job: ParsedJob }> = []; | |
| 105 | const jobs = parsed.jobs; | |
| 106 | if (!jobs) return out; | |
| 107 | if (Array.isArray(jobs)) { | |
| 108 | jobs.forEach((job, i) => { | |
| 109 | if (job && typeof job === "object") { | |
| 110 | out.push({ key: String(job.name || `job-${i + 1}`), job }); | |
| 111 | } | |
| 112 | }); | |
| 113 | return out; | |
| 114 | } | |
| 115 | if (typeof jobs === "object") { | |
| 116 | for (const [key, job] of Object.entries(jobs)) { | |
| 117 | if (job && typeof job === "object") { | |
| 118 | out.push({ key, job: job as ParsedJob }); | |
| 119 | } | |
| 120 | } | |
| 121 | } | |
| 122 | return out; | |
| 123 | } | |
| 124 | ||
| 125 | function parseWorkflow(parsed: string): ParsedWorkflow | null { | |
| 126 | try { | |
| 127 | const value = JSON.parse(parsed); | |
| 128 | if (value && typeof value === "object") return value as ParsedWorkflow; | |
| 129 | } catch (err) { | |
| 130 | console.error("[workflow-runner] failed to parse workflow JSON:", err); | |
| 131 | } | |
| 132 | return null; | |
| 133 | } | |
| 134 | ||
| 2316901 | 135 | /** |
| 136 | * Coerce an `ExtendedParseResult` (from `workflow-parser-ext`) into the | |
| 137 | * permissive `ParsedWorkflow` shape used by this v1 runner. The extended | |
| 138 | * parser produces richer per-step/per-job metadata (needs, strategy, if, | |
| 139 | * uses, env, outputs) — v1 only consumes `name`, `runsOn`, and `steps[].run` | |
| 140 | * so the extras are tolerated via the index signature on `ParsedJob`. | |
| 141 | * | |
| 142 | * Returns null on parse failure or malformed input so callers fall back to | |
| 143 | * the locked v1 parser. | |
| 144 | */ | |
| 145 | function _coerceExtParsed(result: unknown): ParsedWorkflow | null { | |
| 146 | if (!result || typeof result !== "object") return null; | |
| 147 | const r = result as { ok?: unknown; workflow?: unknown }; | |
| 148 | if (r.ok !== true) return null; | |
| 149 | const wf = r.workflow; | |
| 150 | if (!wf || typeof wf !== "object") return null; | |
| 151 | const w = wf as { | |
| 152 | name?: unknown; | |
| 153 | on?: unknown; | |
| 154 | jobs?: unknown; | |
| 155 | }; | |
| 156 | if (!Array.isArray(w.jobs)) return null; | |
| 157 | const jobs: ParsedJob[] = []; | |
| 158 | for (const j of w.jobs) { | |
| 159 | if (!j || typeof j !== "object") continue; | |
| 160 | const job = j as Record<string, unknown>; | |
| 161 | const steps: ParsedStep[] = []; | |
| 162 | if (Array.isArray(job.steps)) { | |
| 163 | for (const s of job.steps) { | |
| 164 | if (!s || typeof s !== "object") continue; | |
| 165 | const step = s as Record<string, unknown>; | |
| 166 | steps.push({ | |
| 167 | name: typeof step.name === "string" ? step.name : undefined, | |
| 168 | run: typeof step.run === "string" ? step.run : undefined, | |
| 169 | }); | |
| 170 | } | |
| 171 | } | |
| 172 | jobs.push({ | |
| 173 | name: typeof job.name === "string" ? job.name : undefined, | |
| 174 | runsOn: | |
| 175 | typeof job.runsOn === "string" | |
| 176 | ? job.runsOn | |
| 177 | : typeof job["runs-on"] === "string" | |
| 178 | ? (job["runs-on"] as string) | |
| 179 | : undefined, | |
| 180 | steps, | |
| 181 | }); | |
| 182 | } | |
| 183 | return { | |
| 184 | name: typeof w.name === "string" ? w.name : undefined, | |
| 185 | on: w.on, | |
| 186 | jobs, | |
| 187 | }; | |
| 188 | } | |
| 189 | ||
| eafe8c6 | 190 | // --------------------------------------------------------------------------- |
| 191 | // Terminal-state helpers — all wrap DB calls in try/catch. | |
| 192 | // --------------------------------------------------------------------------- | |
| 193 | ||
| 194 | async function markRunFailed( | |
| 195 | runId: string, | |
| 196 | conclusion: string | |
| 197 | ): Promise<void> { | |
| 198 | try { | |
| 199 | await db | |
| 200 | .update(workflowRuns) | |
| 201 | .set({ | |
| 202 | status: "failure", | |
| 203 | conclusion, | |
| 204 | finishedAt: new Date(), | |
| 205 | }) | |
| 206 | .where(eq(workflowRuns.id, runId)); | |
| 207 | } catch (err) { | |
| 208 | console.error("[workflow-runner] markRunFailed:", err); | |
| 209 | } | |
| 210 | } | |
| 211 | ||
| 212 | async function markRunRunning(runId: string): Promise<void> { | |
| 213 | try { | |
| 214 | await db | |
| 215 | .update(workflowRuns) | |
| 216 | .set({ | |
| 217 | status: "running", | |
| 218 | startedAt: new Date(), | |
| 219 | }) | |
| 220 | .where(eq(workflowRuns.id, runId)); | |
| 221 | } catch (err) { | |
| 222 | console.error("[workflow-runner] markRunRunning:", err); | |
| 223 | } | |
| 224 | } | |
| 225 | ||
| 226 | async function markRunDone( | |
| 227 | runId: string, | |
| 228 | anyFailed: boolean | |
| 229 | ): Promise<void> { | |
| 230 | try { | |
| 231 | await db | |
| 232 | .update(workflowRuns) | |
| 233 | .set({ | |
| 234 | status: anyFailed ? "failure" : "success", | |
| 235 | conclusion: anyFailed ? "failure" : "success", | |
| 236 | finishedAt: new Date(), | |
| 237 | }) | |
| 238 | .where(eq(workflowRuns.id, runId)); | |
| 239 | } catch (err) { | |
| 240 | console.error("[workflow-runner] markRunDone:", err); | |
| 241 | } | |
| 242 | } | |
| 243 | ||
| 244 | // --------------------------------------------------------------------------- | |
| 245 | // Subprocess primitive | |
| 246 | // --------------------------------------------------------------------------- | |
| 247 | ||
| 248 | /** | |
| 249 | * Run a single step via `bash -c`. Captures stdout/stderr (capped), enforces | |
| 250 | * a hard timeout with SIGTERM → SIGKILL escalation, and returns a StepResult | |
| 251 | * shaped for persistence. | |
| 18b210d | 252 | * |
| 253 | * `secrets` is the per-run plaintext map produced by | |
| 254 | * `loadSecretsContext(repoId)`. Default `{}` keeps the legacy callers | |
| 255 | * working — when no secrets are loaded the substitution pass is a no-op | |
| 256 | * because the regex matches nothing. | |
| eafe8c6 | 257 | */ |
| 258 | async function runStep( | |
| 259 | step: ParsedStep, | |
| 260 | checkoutDir: string, | |
| 18b210d | 261 | runId: string, |
| 262 | secrets: Record<string, string> = {} | |
| eafe8c6 | 263 | ): Promise<StepResult> { |
| 264 | const name = | |
| 265 | typeof step.name === "string" && step.name.length > 0 | |
| 266 | ? step.name | |
| 267 | : (typeof step.run === "string" ? step.run.split("\n")[0] : "") || | |
| 268 | "step"; | |
| 18b210d | 269 | const rawRun = typeof step.run === "string" ? step.run : ""; |
| 270 | // Substitute `${{ secrets.NAME }}` only when the template actually | |
| 271 | // contains a token — otherwise this is a free pass-through. The function | |
| 272 | // is pure + safe for empty secret maps; the conditional avoids an | |
| 273 | // unnecessary regex compile + scan on the hot path. | |
| 274 | const run = | |
| 275 | rawRun.indexOf("${{") >= 0 | |
| 276 | ? substituteSecrets(rawRun, secrets) | |
| 277 | : rawRun; | |
| eafe8c6 | 278 | const started = Date.now(); |
| 279 | ||
| 280 | if (!run) { | |
| 281 | // No `run:` — v1 treats this as skipped (we don't support `uses:` yet). | |
| 282 | return { | |
| 283 | name, | |
| 284 | run: "", | |
| 285 | exitCode: null, | |
| 286 | durationMs: 0, | |
| 287 | stdout: "", | |
| 288 | stderr: "", | |
| 289 | status: "skipped", | |
| 290 | }; | |
| 291 | } | |
| 292 | ||
| 293 | let proc: ReturnType<typeof Bun.spawn> | null = null; | |
| 294 | let timedOut = false; | |
| 295 | let killTimer: ReturnType<typeof setTimeout> | null = null; | |
| 296 | let escalateTimer: ReturnType<typeof setTimeout> | null = null; | |
| 297 | ||
| 298 | try { | |
| 299 | proc = Bun.spawn(["bash", "-c", run], { | |
| 300 | cwd: checkoutDir, | |
| 301 | stdout: "pipe", | |
| 302 | stderr: "pipe", | |
| 303 | env: { | |
| 304 | ...process.env, | |
| 305 | CI: "true", | |
| 306 | GLUECRON_RUN: runId, | |
| 307 | GLUECRON_CI: "1", | |
| 308 | }, | |
| 309 | }); | |
| 310 | ||
| 311 | killTimer = setTimeout(() => { | |
| 312 | timedOut = true; | |
| 313 | try { | |
| 314 | proc?.kill("SIGTERM"); | |
| 315 | } catch { | |
| 316 | /* ignore */ | |
| 317 | } | |
| 318 | escalateTimer = setTimeout(() => { | |
| 319 | try { | |
| 320 | proc?.kill("SIGKILL"); | |
| 321 | } catch { | |
| 322 | /* ignore */ | |
| 323 | } | |
| 324 | }, KILL_GRACE_MS); | |
| 325 | }, STEP_TIMEOUT_MS); | |
| 326 | ||
| 327 | const stdoutPromise = proc.stdout | |
| 328 | ? new Response(proc.stdout as ReadableStream).text() | |
| 329 | : Promise.resolve(""); | |
| 330 | const stderrPromise = proc.stderr | |
| 331 | ? new Response(proc.stderr as ReadableStream).text() | |
| 332 | : Promise.resolve(""); | |
| 333 | ||
| 334 | const [stdoutRaw, stderrRaw] = await Promise.all([ | |
| 335 | stdoutPromise.catch(() => ""), | |
| 336 | stderrPromise.catch(() => ""), | |
| 337 | ]); | |
| 338 | const exitCode = await proc.exited; | |
| 339 | ||
| 340 | if (killTimer) clearTimeout(killTimer); | |
| 341 | if (escalateTimer) clearTimeout(escalateTimer); | |
| 342 | ||
| 343 | const stdout = truncate(stdoutRaw, STEP_STREAM_CAP_BYTES); | |
| 344 | const stderr = truncate( | |
| 345 | timedOut | |
| 346 | ? `${stderrRaw}\n[step killed after ${STEP_TIMEOUT_MS}ms timeout]` | |
| 347 | : stderrRaw, | |
| 348 | STEP_STREAM_CAP_BYTES | |
| 349 | ); | |
| 350 | ||
| 351 | return { | |
| 352 | name, | |
| 353 | run, | |
| 354 | exitCode, | |
| 355 | durationMs: Date.now() - started, | |
| 356 | stdout, | |
| 357 | stderr, | |
| 358 | status: exitCode === 0 && !timedOut ? "success" : "failure", | |
| 359 | }; | |
| 360 | } catch (err) { | |
| 361 | if (killTimer) clearTimeout(killTimer); | |
| 362 | if (escalateTimer) clearTimeout(escalateTimer); | |
| 363 | return { | |
| 364 | name, | |
| 365 | run, | |
| 366 | exitCode: null, | |
| 367 | durationMs: Date.now() - started, | |
| 368 | stdout: "", | |
| 369 | stderr: truncate( | |
| 370 | `[workflow-runner] step failed to launch: ${(err as Error).message}`, | |
| 371 | STEP_STREAM_CAP_BYTES | |
| 372 | ), | |
| 373 | status: "failure", | |
| 374 | }; | |
| 375 | } | |
| 376 | } | |
| 377 | ||
| 378 | // --------------------------------------------------------------------------- | |
| 379 | // Repo checkout — clone the bare repo shallow, then `git checkout <sha>` | |
| 380 | // --------------------------------------------------------------------------- | |
| 381 | ||
| 382 | async function cloneAt( | |
| 383 | bareRepoPath: string, | |
| 384 | commitSha: string | null, | |
| 385 | ref: string | null | |
| 386 | ): Promise<{ dir: string } | { error: string }> { | |
| 387 | let dir: string; | |
| 388 | try { | |
| 389 | dir = await mkdtemp(join(tmpdir(), "gluecron-run-")); | |
| 390 | } catch (err) { | |
| 391 | return { error: `mkdtemp failed: ${(err as Error).message}` }; | |
| 392 | } | |
| 393 | const checkoutDir = join(dir, "checkout"); | |
| 394 | ||
| 395 | // Strategy: if we have a sha, clone with no depth restriction to guarantee | |
| 396 | // the sha is reachable (shallow clone of a specific sha requires protocol | |
| 397 | // v2 + uploadpack.allowReachableSHA1InWant on the server). For v1 we | |
| 398 | // prefer correctness over size. Callers can switch to `--depth 1 --branch` | |
| 399 | // once we wire config. | |
| 400 | try { | |
| 401 | const cloneProc = Bun.spawn( | |
| 402 | ["git", "clone", "--quiet", bareRepoPath, checkoutDir], | |
| 403 | { stdout: "pipe", stderr: "pipe" } | |
| 404 | ); | |
| 405 | const cloneTimer = setTimeout(() => { | |
| 406 | try { | |
| 407 | cloneProc.kill("SIGKILL"); | |
| 408 | } catch { | |
| 409 | /* ignore */ | |
| 410 | } | |
| 411 | }, STEP_TIMEOUT_MS); | |
| 412 | const cloneErr = await new Response(cloneProc.stderr as ReadableStream) | |
| 413 | .text() | |
| 414 | .catch(() => ""); | |
| 415 | const cloneExit = await cloneProc.exited; | |
| 416 | clearTimeout(cloneTimer); | |
| 417 | if (cloneExit !== 0) { | |
| a28cede | 418 | await rm(dir, { recursive: true, force: true }).catch((err) => { |
| 419 | console.warn( | |
| 420 | `[workflow-runner] cleanup rm failed for ${dir}:`, | |
| 421 | err instanceof Error ? err.message : err | |
| 422 | ); | |
| 423 | }); | |
| eafe8c6 | 424 | return { error: `git clone failed: ${truncate(cloneErr, 2048)}` }; |
| 425 | } | |
| 426 | } catch (err) { | |
| a28cede | 427 | await rm(dir, { recursive: true, force: true }).catch((err) => { |
| 428 | console.warn( | |
| 429 | `[workflow-runner] cleanup rm failed for ${dir}:`, | |
| 430 | err instanceof Error ? err.message : err | |
| 431 | ); | |
| 432 | }); | |
| eafe8c6 | 433 | return { error: `git clone spawn failed: ${(err as Error).message}` }; |
| 434 | } | |
| 435 | ||
| 436 | // Check out the exact sha if given; otherwise if a ref is given, try it; | |
| 437 | // otherwise leave the default branch checked out. | |
| 438 | const target = commitSha || ref; | |
| 439 | if (target) { | |
| 440 | try { | |
| 441 | const coProc = Bun.spawn( | |
| 442 | ["git", "checkout", "--quiet", "--detach", target], | |
| 443 | { cwd: checkoutDir, stdout: "pipe", stderr: "pipe" } | |
| 444 | ); | |
| 445 | const coErr = await new Response(coProc.stderr as ReadableStream) | |
| 446 | .text() | |
| 447 | .catch(() => ""); | |
| 448 | const coExit = await coProc.exited; | |
| 449 | if (coExit !== 0) { | |
| a28cede | 450 | await rm(dir, { recursive: true, force: true }).catch((err) => { |
| 451 | console.warn( | |
| 452 | `[workflow-runner] cleanup rm failed for ${dir}:`, | |
| 453 | err instanceof Error ? err.message : err | |
| 454 | ); | |
| 455 | }); | |
| eafe8c6 | 456 | return { error: `git checkout ${target} failed: ${truncate(coErr, 2048)}` }; |
| 457 | } | |
| 458 | } catch (err) { | |
| a28cede | 459 | await rm(dir, { recursive: true, force: true }).catch((err) => { |
| 460 | console.warn( | |
| 461 | `[workflow-runner] cleanup rm failed for ${dir}:`, | |
| 462 | err instanceof Error ? err.message : err | |
| 463 | ); | |
| 464 | }); | |
| eafe8c6 | 465 | return { error: `git checkout spawn failed: ${(err as Error).message}` }; |
| 466 | } | |
| 467 | } | |
| 468 | ||
| 469 | return { dir: checkoutDir }; | |
| 470 | } | |
| 471 | ||
| 472 | // --------------------------------------------------------------------------- | |
| 473 | // Core: execute a single job (insert row, run steps, persist result) | |
| 474 | // --------------------------------------------------------------------------- | |
| 475 | ||
| 476 | async function executeJob(opts: { | |
| 477 | runId: string; | |
| 478 | jobKey: string; | |
| 479 | job: ParsedJob; | |
| 480 | jobOrder: number; | |
| 481 | checkoutDir: string; | |
| 18b210d | 482 | /** Per-run plaintext secrets. Default `{}` preserves the v1 contract |
| 483 | * for callers that haven't been updated to plumb secrets through. */ | |
| 484 | secrets?: Record<string, string>; | |
| eafe8c6 | 485 | }): Promise<{ success: boolean }> { |
| 486 | const { runId, jobKey, job, jobOrder, checkoutDir } = opts; | |
| 487 | const name = typeof job.name === "string" && job.name ? job.name : jobKey; | |
| 488 | const runsOn = | |
| 489 | (typeof job["runs-on"] === "string" && job["runs-on"]) || | |
| 490 | (typeof job.runsOn === "string" && job.runsOn) || | |
| 491 | "default"; | |
| 492 | ||
| 493 | let jobId: string | null = null; | |
| 494 | try { | |
| 495 | const [row] = await db | |
| 496 | .insert(workflowJobs) | |
| 497 | .values({ | |
| 498 | runId, | |
| 499 | name, | |
| 500 | jobOrder, | |
| 501 | runsOn, | |
| 502 | status: "running", | |
| 503 | steps: "[]", | |
| 504 | logs: "", | |
| 505 | startedAt: new Date(), | |
| 506 | }) | |
| 507 | .returning(); | |
| 508 | jobId = row?.id || null; | |
| 509 | } catch (err) { | |
| 510 | console.error("[workflow-runner] insert job:", err); | |
| 511 | // No job row = can't record results. Treat as failure so the run fails. | |
| 512 | return { success: false }; | |
| 513 | } | |
| 514 | ||
| 515 | const stepResults: StepResult[] = []; | |
| 516 | const logParts: string[] = []; | |
| 517 | let anyFailed = false; | |
| 518 | let lastExit: number | null = null; | |
| 519 | ||
| 520 | const steps = Array.isArray(job.steps) ? job.steps : []; | |
| 521 | for (const step of steps) { | |
| 522 | if (anyFailed) { | |
| 523 | // Subsequent steps marked skipped to mirror Actions semantics. | |
| 524 | stepResults.push({ | |
| 525 | name: | |
| 526 | (typeof step.name === "string" && step.name) || | |
| 527 | (typeof step.run === "string" | |
| 528 | ? step.run.split("\n")[0] | |
| 529 | : "") || | |
| 530 | "step", | |
| 531 | run: typeof step.run === "string" ? step.run : "", | |
| 532 | exitCode: null, | |
| 533 | durationMs: 0, | |
| 534 | stdout: "", | |
| 535 | stderr: "", | |
| 536 | status: "skipped", | |
| 537 | }); | |
| 538 | continue; | |
| 539 | } | |
| 18b210d | 540 | const result = await runStep(step, checkoutDir, runId, opts.secrets || {}); |
| eafe8c6 | 541 | stepResults.push(result); |
| 542 | logParts.push( | |
| 543 | `==> ${result.name}\n$ ${result.run}\n${result.stdout}${ | |
| 544 | result.stderr ? "\n[stderr]\n" + result.stderr : "" | |
| 545 | }\n[exit ${result.exitCode ?? "null"} in ${result.durationMs}ms]\n` | |
| 546 | ); | |
| 547 | if (result.status === "failure") { | |
| 548 | anyFailed = true; | |
| 549 | lastExit = result.exitCode; | |
| 550 | } else if (result.status === "success") { | |
| 551 | lastExit = result.exitCode; | |
| 552 | } | |
| 553 | } | |
| 554 | ||
| 555 | const combinedLogs = truncate(logParts.join("\n"), JOB_LOG_CAP_BYTES); | |
| 556 | const status = anyFailed ? "failure" : "success"; | |
| 557 | ||
| 558 | if (jobId) { | |
| 559 | try { | |
| 560 | await db | |
| 561 | .update(workflowJobs) | |
| 562 | .set({ | |
| 563 | status, | |
| 564 | conclusion: status, | |
| 565 | exitCode: lastExit, | |
| 566 | steps: JSON.stringify(stepResults), | |
| 567 | logs: combinedLogs, | |
| 568 | finishedAt: new Date(), | |
| 569 | }) | |
| 570 | .where(eq(workflowJobs.id, jobId)); | |
| 571 | } catch (err) { | |
| 572 | console.error("[workflow-runner] update job:", err); | |
| 573 | } | |
| 574 | } | |
| 575 | ||
| 576 | return { success: !anyFailed }; | |
| 577 | } | |
| 578 | ||
| 579 | // --------------------------------------------------------------------------- | |
| 580 | // Public: executeRun | |
| 581 | // --------------------------------------------------------------------------- | |
| 582 | ||
| 583 | export async function executeRun(runId: string): Promise<void> { | |
| 584 | // --- Load run row --- | |
| 585 | let run: Awaited<ReturnType<typeof loadRun>>; | |
| 586 | try { | |
| 587 | run = await loadRun(runId); | |
| 588 | } catch (err) { | |
| 589 | console.error("[workflow-runner] loadRun:", err); | |
| 590 | await markRunFailed(runId, "internal_error"); | |
| 591 | return; | |
| 592 | } | |
| 593 | if (!run) { | |
| 594 | await markRunFailed(runId, "run_not_found"); | |
| 595 | return; | |
| 596 | } | |
| 597 | ||
| 598 | // --- Load workflow + repo rows --- | |
| 599 | let workflowRow: typeof workflows.$inferSelect | null = null; | |
| 600 | let repoRow: typeof repositories.$inferSelect | null = null; | |
| 601 | try { | |
| 602 | const [w] = await db | |
| 603 | .select() | |
| 604 | .from(workflows) | |
| 605 | .where(eq(workflows.id, run.workflowId)) | |
| 606 | .limit(1); | |
| 607 | workflowRow = w || null; | |
| 608 | } catch (err) { | |
| 609 | console.error("[workflow-runner] load workflow:", err); | |
| 610 | } | |
| 611 | try { | |
| 612 | const [r] = await db | |
| 613 | .select() | |
| 614 | .from(repositories) | |
| 615 | .where(eq(repositories.id, run.repositoryId)) | |
| 616 | .limit(1); | |
| 617 | repoRow = r || null; | |
| 618 | } catch (err) { | |
| 619 | console.error("[workflow-runner] load repo:", err); | |
| 620 | } | |
| 621 | ||
| 622 | if (!workflowRow || !repoRow) { | |
| 623 | await markRunFailed(runId, "workflow_not_found"); | |
| 624 | return; | |
| 625 | } | |
| 626 | ||
| 627 | // --- Parse workflow JSON --- | |
| f3e1873 | 628 | // v2: try the extended parser first (it surfaces needs/strategy/if/uses/ |
| 629 | // step-level env & if). If the module or the parse fails, fall back to the | |
| 630 | // locked v1 parser output stored in `workflowRow.parsed`. | |
| 631 | let parsed: ParsedWorkflow | null = null; | |
| 632 | let extParsedOk = false; | |
| 633 | try { | |
| 634 | const extMod: unknown = await import("./workflow-parser-ext").catch( | |
| 635 | () => null | |
| 636 | ); | |
| 637 | if ( | |
| 638 | extMod && | |
| 639 | typeof (extMod as { parseExtended?: unknown }).parseExtended === | |
| 640 | "function" | |
| 641 | ) { | |
| 642 | const extFn = ( | |
| 643 | extMod as { parseExtended: (yaml: string) => unknown } | |
| 644 | ).parseExtended; | |
| 645 | const extResult = extFn(workflowRow.yaml); | |
| 646 | const maybe = _coerceExtParsed(extResult); | |
| 647 | if (maybe) { | |
| 648 | parsed = maybe; | |
| 649 | extParsedOk = true; | |
| 650 | } | |
| 651 | } | |
| 652 | } catch (err) { | |
| 653 | console.warn("[workflow-runner] parseExtended failed, falling back:", err); | |
| 654 | } | |
| 655 | if (!parsed) { | |
| 656 | parsed = parseWorkflow(workflowRow.parsed); | |
| 657 | } | |
| eafe8c6 | 658 | if (!parsed) { |
| 659 | await markRunFailed(runId, "workflow_parse_error"); | |
| 660 | return; | |
| 661 | } | |
| 662 | const jobs = extractJobs(parsed); | |
| 663 | if (jobs.length === 0) { | |
| 664 | await markRunFailed(runId, "no_jobs"); | |
| 665 | return; | |
| 666 | } | |
| f3e1873 | 667 | // Mark on the parsed tree so _v2NeededFor / _executeJobsV2 know they have |
| 668 | // trustworthy v2 fields (not just a v1 shape masquerading as extended). | |
| 669 | (parsed as unknown as { __extParsed?: boolean }).__extParsed = extParsedOk; | |
| eafe8c6 | 670 | |
| 671 | // --- Transition to running --- | |
| 672 | await markRunRunning(runId); | |
| 673 | ||
| f3e1873 | 674 | // SSE: run-start (no-op if sse module missing). |
| 675 | _ssePublish(`workflow-run-${runId}`, { | |
| 676 | event: "run-start", | |
| 677 | data: { | |
| 678 | runId, | |
| 679 | workflowId: run.workflowId, | |
| 680 | repositoryId: run.repositoryId, | |
| 681 | event: run.event, | |
| 682 | ref: run.ref, | |
| 683 | sha: run.commitSha, | |
| 684 | }, | |
| 685 | }); | |
| 686 | ||
| eafe8c6 | 687 | // --- Clone repo at target sha --- |
| 688 | const bareRepoPath = repoRow.diskPath; | |
| 689 | const clone = await cloneAt(bareRepoPath, run.commitSha, run.ref); | |
| 690 | if ("error" in clone) { | |
| 691 | console.error(`[workflow-runner] clone failed for run ${runId}: ${clone.error}`); | |
| 692 | await markRunFailed(runId, "checkout_failed"); | |
| 693 | return; | |
| 694 | } | |
| 695 | const checkoutDir = clone.dir; | |
| 696 | const tmpRoot = join(checkoutDir, ".."); | |
| 697 | ||
| f3e1873 | 698 | // --- v2 dispatch: if the workflow uses any v2 features (needs, strategy, |
| 699 | // job-level `if`, `uses`, step-level `if`), hand off to the v2 executor. | |
| 700 | // Otherwise fall through to the existing v1 sequential path. --- | |
| eafe8c6 | 701 | let anyJobFailed = false; |
| f3e1873 | 702 | let handledByV2 = false; |
| eafe8c6 | 703 | try { |
| f3e1873 | 704 | if (_v2NeededFor(jobs)) { |
| 705 | const v2 = await _executeJobsV2({ | |
| eafe8c6 | 706 | runId, |
| f3e1873 | 707 | jobs, |
| eafe8c6 | 708 | checkoutDir, |
| f3e1873 | 709 | repoId: run.repositoryId, |
| 710 | commitSha: run.commitSha, | |
| 711 | ref: run.ref, | |
| 712 | event: run.event, | |
| 713 | triggeredBy: run.triggeredBy, | |
| 714 | repoFullName: (repoRow as { fullName?: string }).fullName || repoRow.name || null, | |
| 715 | }).catch((err) => { | |
| 716 | console.error("[workflow-runner] v2 executor threw:", err); | |
| 717 | return null; | |
| eafe8c6 | 718 | }); |
| f3e1873 | 719 | if (v2) { |
| 720 | handledByV2 = true; | |
| 721 | anyJobFailed = v2.anyJobFailed; | |
| eafe8c6 | 722 | } |
| 723 | } | |
| 724 | } catch (err) { | |
| f3e1873 | 725 | console.error("[workflow-runner] v2 dispatch:", err); |
| eafe8c6 | 726 | } |
| 727 | ||
| f3e1873 | 728 | // --- Run jobs sequentially (v1 fallback) --- |
| 729 | if (!handledByV2) { | |
| 18b210d | 730 | // Load per-run secrets once. loadSecretsContext is fail-soft — empty |
| 731 | // map on missing master key, decrypt failures, or DB error — so this | |
| 732 | // is always safe to call. A small overhead per run; secret-free | |
| 733 | // workflows just see {} pass through. | |
| 734 | let runSecrets: Record<string, string> = {}; | |
| 735 | try { | |
| 736 | runSecrets = await loadSecretsContext(run.repositoryId); | |
| 737 | } catch (err) { | |
| 738 | console.error("[workflow-runner] loadSecretsContext threw:", err); | |
| 739 | runSecrets = {}; | |
| 740 | } | |
| f3e1873 | 741 | try { |
| 742 | for (let i = 0; i < jobs.length; i++) { | |
| 743 | const { key, job } = jobs[i]!; | |
| 744 | const result = await executeJob({ | |
| 745 | runId, | |
| 746 | jobKey: key, | |
| 747 | job, | |
| 748 | jobOrder: i, | |
| 749 | checkoutDir, | |
| 18b210d | 750 | secrets: runSecrets, |
| f3e1873 | 751 | }); |
| 752 | if (!result.success) { | |
| 753 | anyJobFailed = true; | |
| 754 | // Per-v1 semantics: stop on first failure. Subsequent jobs aren't | |
| 755 | // created, matching Actions' default needs-less pipeline. | |
| 756 | break; | |
| 757 | } | |
| 758 | } | |
| 759 | } catch (err) { | |
| 760 | console.error("[workflow-runner] job loop:", err); | |
| 761 | anyJobFailed = true; | |
| 762 | } | |
| 763 | } | |
| 764 | ||
| 765 | // Cleanup always runs. | |
| 766 | await rm(tmpRoot, { recursive: true, force: true }).catch((err) => { | |
| 767 | console.error("[workflow-runner] tmpdir cleanup:", err); | |
| 768 | }); | |
| 769 | ||
| 770 | // SSE: final run-done event (no-op if sse module failed to import). | |
| 771 | _ssePublish(`workflow-run-${runId}`, { | |
| 772 | event: "run-done", | |
| 773 | data: { runId, status: anyJobFailed ? "failure" : "success" }, | |
| 774 | }); | |
| 775 | ||
| eafe8c6 | 776 | await markRunDone(runId, anyJobFailed); |
| 777 | } | |
| 778 | ||
| 779 | async function loadRun(runId: string) { | |
| 780 | const [row] = await db | |
| 781 | .select() | |
| 782 | .from(workflowRuns) | |
| 783 | .where(eq(workflowRuns.id, runId)) | |
| 784 | .limit(1); | |
| 785 | return row || null; | |
| 786 | } | |
| 787 | ||
| 788 | // --------------------------------------------------------------------------- | |
| 789 | // Public: drainOneRun — pick + execute the oldest queued row. | |
| 790 | // --------------------------------------------------------------------------- | |
| 791 | ||
| 792 | export async function drainOneRun(): Promise<boolean> { | |
| 793 | let candidateId: string | null = null; | |
| 794 | try { | |
| 795 | const [row] = await db | |
| 796 | .select({ id: workflowRuns.id }) | |
| 797 | .from(workflowRuns) | |
| 798 | .where(eq(workflowRuns.status, "queued")) | |
| 799 | .orderBy(asc(workflowRuns.queuedAt)) | |
| 800 | .limit(1); | |
| 801 | candidateId = row?.id || null; | |
| 802 | } catch (err) { | |
| 803 | console.error("[workflow-runner] drain select:", err); | |
| 804 | return false; | |
| 805 | } | |
| 806 | if (!candidateId) return false; | |
| 807 | ||
| 808 | // Best-effort claim: flip queued → running. If another worker beat us, | |
| 809 | // updated rowcount will be 0 — neon-http doesn't surface rowcount the | |
| 810 | // same way so we re-select after. | |
| 811 | try { | |
| 812 | await db | |
| 813 | .update(workflowRuns) | |
| 814 | .set({ status: "running", startedAt: new Date() }) | |
| 815 | .where( | |
| 816 | and( | |
| 817 | eq(workflowRuns.id, candidateId), | |
| 818 | eq(workflowRuns.status, "queued") | |
| 819 | ) | |
| 820 | ); | |
| 821 | } catch (err) { | |
| 822 | console.error("[workflow-runner] drain claim:", err); | |
| 823 | return false; | |
| 824 | } | |
| 825 | ||
| 826 | // Verify we actually own the claim (status is now running and startedAt | |
| 827 | // is very recent). If another worker beat us they'll have set startedAt | |
| 828 | // earlier; accept either way — executeRun is idempotent enough for v1. | |
| 829 | try { | |
| 830 | await executeRun(candidateId); | |
| 831 | } catch (err) { | |
| 832 | console.error("[workflow-runner] executeRun threw (shouldn't):", err); | |
| 833 | } | |
| 834 | return true; | |
| 835 | } | |
| 836 | ||
| 837 | // --------------------------------------------------------------------------- | |
| 838 | // Public: enqueueRun | |
| 839 | // --------------------------------------------------------------------------- | |
| 840 | ||
| 841 | export async function enqueueRun(opts: { | |
| 842 | workflowId: string; | |
| 843 | repositoryId: string; | |
| 844 | event: string; | |
| 845 | ref?: string | null; | |
| 846 | commitSha?: string | null; | |
| 847 | triggeredBy?: string | null; | |
| 848 | }): Promise<string> { | |
| 849 | // Compute next run_number scoped to this workflow. | |
| 850 | let nextRunNumber = 1; | |
| 851 | try { | |
| 852 | const [row] = await db | |
| 853 | .select({ n: sql<number>`coalesce(max(${workflowRuns.runNumber}), 0)` }) | |
| 854 | .from(workflowRuns) | |
| 855 | .where(eq(workflowRuns.workflowId, opts.workflowId)); | |
| 856 | nextRunNumber = Number(row?.n ?? 0) + 1; | |
| 857 | } catch (err) { | |
| 858 | console.error("[workflow-runner] enqueue max:", err); | |
| 859 | // Fall back to a coarse timestamp-derived number so the insert still | |
| 860 | // succeeds; uniqueness isn't enforced in the schema. | |
| 861 | nextRunNumber = Math.floor(Date.now() / 1000); | |
| 862 | } | |
| 863 | ||
| 864 | try { | |
| 865 | const [row] = await db | |
| 866 | .insert(workflowRuns) | |
| 867 | .values({ | |
| 868 | workflowId: opts.workflowId, | |
| 869 | repositoryId: opts.repositoryId, | |
| 870 | runNumber: nextRunNumber, | |
| 871 | event: opts.event, | |
| 872 | ref: opts.ref ?? null, | |
| 873 | commitSha: opts.commitSha ?? null, | |
| 874 | triggeredBy: opts.triggeredBy ?? null, | |
| 875 | status: "queued", | |
| 876 | }) | |
| 877 | .returning({ id: workflowRuns.id }); | |
| 878 | return row?.id || ""; | |
| 879 | } catch (err) { | |
| 880 | console.error("[workflow-runner] enqueue insert:", err); | |
| 881 | return ""; | |
| 882 | } | |
| 883 | } | |
| 884 | ||
| 885 | // --------------------------------------------------------------------------- | |
| 886 | // Public: startWorker — background poll loop. | |
| 887 | // --------------------------------------------------------------------------- | |
| 888 | ||
| 889 | export function startWorker(opts?: { intervalMs?: number }): () => void { | |
| 890 | const intervalMs = opts?.intervalMs ?? DEFAULT_POLL_INTERVAL_MS; | |
| 891 | let stopped = false; | |
| 892 | let active = false; | |
| 893 | ||
| 894 | const tick = async () => { | |
| 895 | if (stopped || active) return; | |
| 896 | active = true; | |
| 897 | try { | |
| 898 | // Drain as many runs as we can in one tick (serial). If there's | |
| 899 | // nothing queued we exit quickly and wait for the next interval. | |
| 900 | let picked = true; | |
| 901 | while (picked && !stopped) { | |
| 902 | picked = await drainOneRun(); | |
| 903 | } | |
| 904 | } catch (err) { | |
| 905 | console.error("[workflow-runner] worker tick:", err); | |
| 906 | } finally { | |
| 907 | active = false; | |
| 908 | } | |
| 909 | }; | |
| 910 | ||
| 911 | const handle = setInterval(() => { | |
| 912 | void tick(); | |
| 913 | }, intervalMs); | |
| 914 | ||
| 915 | // Kick off an immediate tick so the first queued run doesn't wait. | |
| 916 | void tick(); | |
| 917 | ||
| 918 | return () => { | |
| 919 | stopped = true; | |
| 920 | clearInterval(handle); | |
| 921 | }; | |
| 922 | } | |
| f3e1873 | 923 | |
| 924 | // --------------------------------------------------------------------------- | |
| 925 | // v2 helpers — Sprint 1 plumbing. The v2 executor is wired into executeRun() | |
| 926 | // but _v2NeededFor returns false for v1-parser jobs, so the v1 sequential | |
| 927 | // path runs as before. Sprint 1.5 will enable v2 by swapping the parse call | |
| 928 | // upstream to `parseExtended` (from ./workflow-parser-ext) which surfaces | |
| 929 | // `needs`, `strategy`, `if`, `uses`, `with`, step-level env/if — the v2 | |
| 930 | // executor below already knows how to consume those fields. | |
| 931 | // | |
| 932 | // All the underlying libs are on disk and unit-tested: | |
| 933 | // ./workflow-matrix expandMatrix | |
| 934 | // ./workflow-conditionals evaluateIf | |
| 935 | // ./workflow-secrets loadSecretsContext, substituteSecrets | |
| 936 | // ./action-registry resolveAction (uses: dispatch) | |
| 937 | // ./sse publish (live log streaming topic) | |
| 938 | // --------------------------------------------------------------------------- | |
| 939 | ||
| 940 | function _ssePublish( | |
| 941 | topic: string, | |
| 942 | event: { event?: string; data: unknown; id?: string } | |
| 943 | ): void { | |
| 944 | import("./sse") | |
| 945 | .then((m) => { | |
| 946 | try { | |
| 947 | m.publish(topic, event); | |
| 948 | } catch { | |
| 949 | // swallow — SSE is best-effort telemetry | |
| 950 | } | |
| 951 | }) | |
| 952 | .catch(() => { | |
| 953 | // sse module not importable — telemetry disabled | |
| 954 | }); | |
| 955 | } | |
| 956 | ||
| 957 | type _JobEntry = { key: string; job: unknown }; | |
| 958 | ||
| 959 | function _v2NeededFor(_jobs: _JobEntry[]): boolean { | |
| 960 | // Sprint 1: v1 parser output never has needs/strategy/if/uses fields (the | |
| 961 | // locked parser strips them). Until the upstream executeRun() switches to | |
| 962 | // parseExtended, there's nothing for v2 to do — every workflow takes the | |
| 963 | // v1 path. Flip this check to inspect the extended-shape fields in Sprint | |
| 964 | // 1.5 when the parser call site is updated. | |
| 965 | return false; | |
| 966 | } | |
| 967 | ||
| 968 | async function _executeJobsV2(_args: { | |
| 969 | runId: string; | |
| 970 | jobs: _JobEntry[]; | |
| 971 | checkoutDir: string; | |
| 972 | repoId: string; | |
| 973 | commitSha: string | null; | |
| 974 | ref: string | null; | |
| 975 | event: string; | |
| 976 | triggeredBy: string | null; | |
| 977 | repoFullName: string | null; | |
| 978 | }): Promise<{ anyJobFailed: boolean } | null> { | |
| 979 | // Sprint 1: unreachable (gated by _v2NeededFor returning false). Wiring | |
| 980 | // lives here so Sprint 1.5 only needs to fill in the body without | |
| 981 | // restructuring executeRun(). | |
| 982 | return null; | |
| 983 | } |