From 67a72b10239b71da2e9fc24235159718d9ad1ffd Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Marc-Andr=C3=A9=20Bombeck?= <mbombeck@gmail.com>
Date: Sun, 31 May 2026 22:42:11 +0200
Subject: [PATCH] perf(insights): parallelise the nightly per-user assessment
 warm pass
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

The warm pass forced the seven per-metric status generators strictly
back to back per user, so a fully configured account paid ~7× one LLM
round-trip serially. Bound the inner pass with a small `p-limit(3)` so
the cards warm a few at a time without saturating the Prisma pool or
fanning out an unbounded number of provider calls. The per-user budget
gate still bounds total cost per night.
---
 src/lib/jobs/insight-pregenerate.ts | 40 +++++++++++++++++++++--------
 1 file changed, 29 insertions(+), 11 deletions(-)

diff --git a/src/lib/jobs/insight-pregenerate.ts b/src/lib/jobs/insight-pregenerate.ts
index bbfd9c2bd..e70f9f3b3 100644
--- a/src/lib/jobs/insight-pregenerate.ts
+++ b/src/lib/jobs/insight-pregenerate.ts
@@ -45,6 +45,8 @@
  * Recurring pg-boss task — never runs inside an HTTP request and never
  * shells out to `tsx` (CLAUDE.md DO-NOTs).
  */
+import pLimit from "p-limit";
+
 import type { PrismaClient } from "@/generated/prisma/client";
 import { checkRateLimit } from "@/lib/rate-limit";
 import { getAssistantFlags } from "@/lib/feature-flags";
@@ -81,6 +83,17 @@ export const PREGENERATE_BATCH_CAP = 200;
 /** Per-user budget bucket window — one pre-generation per 20 h. */
 const PREGENERATE_BUDGET_WINDOW_MS = 20 * 60 * 60 * 1000;
 
+/**
+ * Concurrency for the per-user warm pass. The seven `*StatusForUser`
+ * generators are independent (each writes its own cache row), so they run
+ * a few at a time rather than strictly serially — a daily-weigher with
+ * every card configured was paying ~7× one LLM round-trip per user back
+ * to back. Capped low (3) so the warm pass cannot saturate the Prisma
+ * pool or fan out an unbounded number of concurrent provider calls; the
+ * per-user budget gate above still bounds total cost per night.
+ */
+const WARM_PASS_CONCURRENCY = 3;
+
 export interface InsightPregeneratePayload {
   triggeredAt: string;
 }
@@ -146,17 +159,22 @@ async function warmPerStatusCaches(
   locale: "de" | "en",
   generators: ReadonlyArray<StatusGenerator>,
 ): Promise<number> {
-  let warmed = 0;
-  for (const generate of generators) {
-    try {
-      const outcome = await generate(userId, { locale, force: true });
-      if (outcome.hasProvider && !outcome.cached) warmed++;
-    } catch {
-      // A single card's generation failing must not abort the rest of
-      // the warm pass for this user, nor the cron's user loop.
-    }
-  }
-  return warmed;
+  const limit = pLimit(WARM_PASS_CONCURRENCY);
+  const results = await Promise.all(
+    generators.map((generate) =>
+      limit(async () => {
+        try {
+          const outcome = await generate(userId, { locale, force: true });
+          return outcome.hasProvider && !outcome.cached ? 1 : 0;
+        } catch {
+          // A single card's generation failing must not abort the rest of
+          // the warm pass for this user, nor the cron's user loop.
+          return 0;
+        }
+      }),
+    ),
+  );
+  return results.reduce((sum: number, n) => sum + n, 0);
 }
 
 interface PregenerateCandidate {