@kici-dev/agent 0.4.0 → 0.5.0

package/dist/index.js

@@ -748,7 +748,7 @@ function isAbsoluteRel(rel) {
 * env vars — the install runs with `--ignore-scripts` whenever a private
 * registry is configured.
 */
-const logger$2 = createLogger({ prefix: "dep-installer" });
+const logger$3 = createLogger({ prefix: "dep-installer" });
 const execFileAsync = promisify(execFile);
 /** Install subprocess timeout (10 min) and stdout/stderr buffer (128 MiB). */
 const INSTALL_TIMEOUT_MS = 6e5;
@@ -793,7 +793,7 @@ async function installDeps(kiciDir, opts = {}) {
 	const repoRoot = opts.repoRoot ?? dirname(kiciDir);
 	const packageManager = await detectKiciPackageManager(repoRoot, kiciDir);
 	const yarnFlavor = packageManager === PackageManager.Yarn ? await detectKiciYarnFlavor(repoRoot, kiciDir) : YarnFlavor.Classic;
-	logger$2.info("Installing deps inline", {
+	logger$3.info("Installing deps inline", {
 		packageManager,
 		yarnFlavor,
 		dir: kiciDir
@@ -851,7 +851,7 @@ async function installDeps(kiciDir, opts = {}) {
 	if (packageManager === PackageManager.Yarn) await buildYarnWorkspaceClosure(repoRoot, kiciDir, yarnFlavor);
 	const durationMs = Date.now() - startTime;
 	process.stderr.write(`[dep-installer:trace] install complete: ${durationMs}ms\n`);
-	logger$2.info("Deps installed inline", {
+	logger$3.info("Deps installed inline", {
 		packageManager,
 		durationMs
 	});
@@ -1143,10 +1143,25 @@ var init_dep_restore = __esmMin((() => {
 * dep-restore.ts and workflow-loader.ts.
 */
 var download_exports = /* @__PURE__ */ __exportAll({
+	UPLOAD_MAX_RETRIES: () => 2,
 	downloadUrl: () => downloadUrl,
 	uploadToPresignedUrl: () => uploadToPresignedUrl
 });
 /**
+* Whether a failed upload attempt is worth repeating.
+*
+* A transport failure (connection refused, reset, DNS) never reached a
+* responder, and 5xx / 429 are the object-storage overload signals AWS
+* documents as retry-with-backoff (S3 answers `SlowDown` with 503). Every other
+* status is a decision the server will repeat: a 403 from an expired or
+* malformed signature, a 400 from a malformed request. Retrying those burns the
+* ceiling without a chance of success and delays the real error.
+*/
+function isRetryableUploadFailure(err) {
+	if (!(err instanceof PresignedUploadHttpError)) return true;
+	return err.statusCode >= 500 || err.statusCode === 429;
+}
+/**
 * Download content from an HTTP/HTTPS URL.
 *
 * Includes a 5-minute timeout to prevent the agent from hanging indefinitely
@@ -1170,21 +1185,10 @@ function downloadUrl(url) {
 		}).on("error", reject);
 	});
 }
-/**
-* Upload a buffer to a pre-signed S3 URL via HTTP PUT.
-*
-* Used for direct-to-S3 uploads of bundles and dep tarballs. Localhost /
-* 127.0.0.1 URLs are rewritten via `resolveOrchestratorUrl` so the
-* filesystem cache backend's signed URLs work from container agents that
-* can't reach the orchestrator's host loopback directly.
-*
-* @param url - The pre-signed URL to upload to
-* @param data - The buffer to upload
-*/
-function uploadToPresignedUrl(url, data) {
+/** One PUT of the whole buffer. Rejects with {@link PresignedUploadHttpError} on a non-2xx. */
+function putOnce(resolvedUrl, data, timeoutMs) {
 	return new Promise((resolve, reject) => {
-		const resolved = resolveOrchestratorUrl(url);
-		const parsed = new URL(resolved);
+		const parsed = new URL(resolvedUrl);
 		const req = (parsed.protocol === "https:" ? https : http).request({
 			hostname: parsed.hostname,
 			port: parsed.port,
@@ -1193,7 +1197,7 @@ function uploadToPresignedUrl(url, data) {
 			headers: { "Content-Length": data.length }
 		}, (res) => {
 			if (res.statusCode && (res.statusCode < 200 || res.statusCode >= 300)) {
-				reject(/* @__PURE__ */ new Error(`HTTP ${res.statusCode} uploading to pre-signed URL`));
+				reject(new PresignedUploadHttpError(res.statusCode));
 				res.resume();
 				return;
 			}
@@ -1201,14 +1205,80 @@ function uploadToPresignedUrl(url, data) {
 			res.on("end", () => resolve());
 			res.on("error", reject);
 		});
+		req.setTimeout(timeoutMs, () => {
+			req.destroy(/* @__PURE__ */ new Error(`Pre-signed upload timed out after ${timeoutMs}ms`));
+		});
 		req.on("error", reject);
 		req.end(data);
 	});
 }
-var DOWNLOAD_TIMEOUT_MS$1;
+/**
+* Upload a buffer to a pre-signed S3 URL via HTTP PUT, retrying a transient
+* failure.
+*
+* Used for direct-to-S3 uploads of bundles and dep tarballs. Localhost /
+* 127.0.0.1 URLs are rewritten via `resolveOrchestratorUrl` so the
+* filesystem cache backend's signed URLs work from container agents that
+* can't reach the orchestrator's host loopback directly.
+*
+* **Why retrying is safe here.** A pre-signed PUT writes one whole object at a
+* single key: there is no multipart session, no append, and no
+* server-generated identity, so a repeat attempt writes the same bytes to the
+* same key and the last write wins. S3 also only makes an object visible once
+* the body has been received in full, so an attempt that died mid-body left
+* nothing behind. A retry therefore cannot double-write or produce a torn
+* object — which is why every AWS SDK retries PUTs by default.
+*
+* Only a failure that can plausibly differ next time is repeated — see
+* {@link isRetryableUploadFailure}.
+*
+* @param url - The pre-signed URL to upload to
+* @param data - The buffer to upload
+* @param opts.baseDelayMs - Backoff before the first retry (doubles thereafter)
+* @param opts.timeoutMs - Per-attempt socket-inactivity timeout (see
+*   {@link UPLOAD_TIMEOUT_MS}); an override exists so a test can drive the
+*   stall path without waiting out the production budget.
+*/
+async function uploadToPresignedUrl(url, data, opts) {
+	const resolved = resolveOrchestratorUrl(url);
+	const baseDelayMs = opts?.baseDelayMs ?? UPLOAD_RETRY_BASE_DELAY_MS;
+	const timeoutMs = opts?.timeoutMs ?? UPLOAD_TIMEOUT_MS;
+	let lastError;
+	for (let attempt = 0; attempt <= 2; attempt++) {
+		if (attempt > 0) {
+			const delayMs = baseDelayMs * 2 ** (attempt - 1);
+			logger$1.warn("Retrying pre-signed upload", {
+				attempt,
+				delayMs,
+				error: lastError?.message
+			});
+			await new Promise((r) => setTimeout(r, delayMs));
+		}
+		try {
+			await putOnce(resolved, data, timeoutMs);
+			return;
+		} catch (err) {
+			lastError = err instanceof Error ? err : new Error(String(err));
+			if (!isRetryableUploadFailure(lastError)) throw lastError;
+		}
+	}
+	throw new Error(`Pre-signed upload failed after 3 attempts: ${lastError?.message}`);
+}
+var logger$1, DOWNLOAD_TIMEOUT_MS$1, UPLOAD_TIMEOUT_MS, UPLOAD_RETRY_BASE_DELAY_MS, PresignedUploadHttpError;
 var init_download = __esmMin((() => {
 	init_dep_restore();
+	logger$1 = createLogger({ prefix: "agent:download" });
 	DOWNLOAD_TIMEOUT_MS$1 = 300 * 1e3;
+	UPLOAD_TIMEOUT_MS = 300 * 1e3;
+	UPLOAD_RETRY_BASE_DELAY_MS = 500;
+	PresignedUploadHttpError = class extends Error {
+		statusCode;
+		constructor(statusCode) {
+			super(`HTTP ${statusCode} uploading to pre-signed URL`);
+			this.statusCode = statusCode;
+			this.name = "PresignedUploadHttpError";
+		}
+	};
 }));
 //#endregion
 //#region src/execution/cache/cache-engine.ts

package/dist/metrics/prometheus.d.ts

@@ -2,40 +2,40 @@
  * Total completed jobs.
  * Labels:
  * - status: success | failed | cancelled
- * - scaler: injected by orch-side AgentMetricsAggregator (Phase 5b); `stateful` for static agents, backend name (`container` / `firecracker` / `bare-metal`) for scaler-managed
+ * - scaler: injected by orch-side AgentMetricsAggregator; `stateful` for static agents, backend type (`container` / `firecracker` / `bare-metal`) for scaler-managed
  */
 export declare const jobsTotal: import("@opentelemetry/api").Counter<import("@opentelemetry/api").Attributes>;
 /**
  * Currently running jobs.
  * Labels:
- * - scaler: injected by orch-side AgentMetricsAggregator (Phase 5b); `stateful` for static agents, backend name (`container` / `firecracker` / `bare-metal`) for scaler-managed
+ * - scaler: injected by orch-side AgentMetricsAggregator; `stateful` for static agents, backend type (`container` / `firecracker` / `bare-metal`) for scaler-managed
  */
 export declare const jobsActive: import("@opentelemetry/api").UpDownCounter<import("@opentelemetry/api").Attributes>;
 /**
  * Total completed steps.
  * Labels:
  * - status: success | failed | skipped
- * - scaler: injected by orch-side AgentMetricsAggregator (Phase 5b); `stateful` for static agents, backend name (`container` / `firecracker` / `bare-metal`) for scaler-managed
+ * - scaler: injected by orch-side AgentMetricsAggregator; `stateful` for static agents, backend type (`container` / `firecracker` / `bare-metal`) for scaler-managed
  */
 export declare const stepsTotal: import("@opentelemetry/api").Counter<import("@opentelemetry/api").Attributes>;
 /**
  * Step execution duration in seconds.
  * Advisory boundaries cover sub-second steps through 30-minute long-running steps.
  * Labels:
- * - scaler: injected by orch-side AgentMetricsAggregator (Phase 5b); `stateful` for static agents, backend name (`container` / `firecracker` / `bare-metal`) for scaler-managed
+ * - scaler: injected by orch-side AgentMetricsAggregator; `stateful` for static agents, backend type (`container` / `firecracker` / `bare-metal`) for scaler-managed
  */
 export declare const stepDurationSeconds: import("@opentelemetry/api").Histogram<import("@opentelemetry/api").Attributes>;
 /**
  * Git clone duration in seconds.
  * Advisory boundaries cover fast shallow clones through large repo clones.
  * Labels:
- * - scaler: injected by orch-side AgentMetricsAggregator (Phase 5b); `stateful` for static agents, backend name (`container` / `firecracker` / `bare-metal`) for scaler-managed
+ * - scaler: injected by orch-side AgentMetricsAggregator; `stateful` for static agents, backend type (`container` / `firecracker` / `bare-metal`) for scaler-managed
  */
 export declare const cloneDurationSeconds: import("@opentelemetry/api").Histogram<import("@opentelemetry/api").Attributes>;
 /**
  * Total log bytes streamed back to orchestrator.
  * Labels:
- * - scaler: injected by orch-side AgentMetricsAggregator (Phase 5b); `stateful` for static agents, backend name (`container` / `firecracker` / `bare-metal`) for scaler-managed
+ * - scaler: injected by orch-side AgentMetricsAggregator; `stateful` for static agents, backend type (`container` / `firecracker` / `bare-metal`) for scaler-managed
  */
 export declare const logBytesTotal: import("@opentelemetry/api").Counter<import("@opentelemetry/api").Attributes>;
 /**
@@ -51,7 +51,7 @@ export declare const logBytesTotal: import("@opentelemetry/api").Counter<import(
  * Labels:
  * - mode: `pause` (producer paused, no data loss) | `drop` (lines
  *   discarded with a `[N lines dropped due to backpressure]` marker)
- * - scaler: injected by orch-side AgentMetricsAggregator (Phase 5b); `stateful` for static agents, backend name (`container` / `firecracker` / `bare-metal`) for scaler-managed
+ * - scaler: injected by orch-side AgentMetricsAggregator; `stateful` for static agents, backend type (`container` / `firecracker` / `bare-metal`) for scaler-managed
  */
 export declare const logBackpressureEventsTotal: import("@opentelemetry/api").Counter<import("@opentelemetry/api").Attributes>;
 /**
@@ -64,7 +64,7 @@ export declare const logBackpressureEventsTotal: import("@opentelemetry/api").Co
  * producer.
  *
  * Labels:
- * - scaler: injected by orch-side AgentMetricsAggregator (Phase 5b); `stateful` for static agents, backend name (`container` / `firecracker` / `bare-metal`) for scaler-managed
+ * - scaler: injected by orch-side AgentMetricsAggregator; `stateful` for static agents, backend type (`container` / `firecracker` / `bare-metal`) for scaler-managed
  */
 export declare const logLinesDroppedTotal: import("@opentelemetry/api").Counter<import("@opentelemetry/api").Attributes>;
 /**
@@ -81,7 +81,7 @@ export declare const logLinesDroppedTotal: import("@opentelemetry/api").Counter<
  *
  * Labels:
  * - mode: `pause` | `drop`
- * - scaler: injected by orch-side AgentMetricsAggregator (Phase 5b); `stateful` for static agents, backend name (`container` / `firecracker` / `bare-metal`) for scaler-managed
+ * - scaler: injected by orch-side AgentMetricsAggregator; `stateful` for static agents, backend type (`container` / `firecracker` / `bare-metal`) for scaler-managed
  */
 export declare const logBackpressureActive: import("@opentelemetry/api").UpDownCounter<import("@opentelemetry/api").Attributes>;
 /**
@@ -89,7 +89,7 @@ export declare const logBackpressureActive: import("@opentelemetry/api").UpDownC
  * Use add(1) for connected, add(-1) for disconnected.
  *
  * Labels:
- * - scaler: injected by orch-side AgentMetricsAggregator (Phase 5b); `stateful` for static agents, backend name (`container` / `firecracker` / `bare-metal`) for scaler-managed
+ * - scaler: injected by orch-side AgentMetricsAggregator; `stateful` for static agents, backend type (`container` / `firecracker` / `bare-metal`) for scaler-managed
  */
 export declare const connectionStatus: import("@opentelemetry/api").UpDownCounter<import("@opentelemetry/api").Attributes>;
 //# sourceMappingURL=prometheus.d.ts.map