iteration-cli 0.7.1 → 0.7.3

package/dist/collector-fetch.js

@@ -2,13 +2,12 @@ import http from "node:http";
 import https from "node:https";
 const MAX_RESPONSE_BYTES = 1024 * 1024;
 const NULL_BODY_STATUSES = new Set([101, 103, 204, 205, 304]);
-// Node's fetch can negotiate HTTP/2 and, after a timed-out request breaks the
-// TLS connection, keep handing every later request the destroyed session until
-// the process restarts. The collector runs for days, so it uses HTTP/1.1
-// connections that are discarded when they fail.
+// Reused connections failed uploads in production: Node's HTTP/2 fetch kept a
+// destroyed session, and kept-alive sockets timed out on ~40 s server waits.
+// See docs/architecture/trace-ingestion.md.
 const agents = {
-    "http:": new http.Agent({ keepAlive: true, maxSockets: 2 }),
-    "https:": new https.Agent({ keepAlive: true, maxSockets: 2 }),
+    "http:": new http.Agent({ keepAlive: false }),
+    "https:": new https.Agent({ keepAlive: false }),
 };
 export const collectorFetch = (input, init = {}) => {
     const url = new URL(input);

package/dist/collector.js

@@ -12,6 +12,10 @@ const MAX_UPLOAD_BYTES = 3 * 1024 * 1024;
 const MAX_EVENT_BYTES = MAX_UPLOAD_BYTES - 1024;
 const MAX_UPLOADS_PER_TICK = 5;
 const MAX_REQUEST_BYTES = 4 * 1024 * 1024;
+// Production stores some batches in about 40 seconds. Abandoning a request
+// earlier does not stop the server, so the retry only waits behind it and the
+// collector reports a failure for work that succeeded.
+const UPLOAD_TIMEOUT_MS = 90_000;
 // A tick can queue up to MAX_SCAN_BYTES_PER_TICK of new events, so scanning
 // pauses well below the outbox capacity and resumes once uploads drain it.
 const SCAN_PAUSE_OUTBOX_BYTES = 32 * 1024 * 1024;
@@ -182,8 +186,8 @@ async function uploadPendingOnce(config, paths, fetcher = collectorFetch, signal
                 },
                 body,
                 signal: signal
-                    ? AbortSignal.any([signal, AbortSignal.timeout(15_000)])
-                    : AbortSignal.timeout(15_000),
+                    ? AbortSignal.any([signal, AbortSignal.timeout(UPLOAD_TIMEOUT_MS)])
+                    : AbortSignal.timeout(UPLOAD_TIMEOUT_MS),
             });
         }
         catch (error) {

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "iteration-cli",
-  "version": "0.7.1",
+  "version": "0.7.3",
   "type": "module",
   "repository": {
     "type": "git",