fix(convex): reduce hot rate-limit and catalog reads

convex/leaderboards.ts

@@ -13,7 +13,7 @@ import {
 
 const MAX_TRENDING_LIMIT = 200;
 const KEEP_LEADERBOARD_ENTRIES = 3;
-const DAILY_STATS_PAGE_SIZE = 5_000;
+const DAILY_STATS_PAGE_SIZE = 1_000;
 
 // ---------------------------------------------------------------------------
 // Action → Query → Mutation pattern (avoids 32K document-read limit)

convex/lib/httpRateLimit.test.ts

@@ -173,6 +173,35 @@ describe("applyRateLimit headers", () => {
     expect(headers.get("Retry-After")).toBeNull();
   });
 
+  it("converts shard write conflicts into a rate-limit response", async () => {
+    vi.spyOn(Date, "now").mockReturnValue(2_500_000);
+    const ctx = {
+      runQuery: vi.fn().mockResolvedValue({
+        allowed: true,
+        remaining: 19,
+        limit: 20,
+        resetAt: 2_530_000,
+      }),
+      runMutation: vi
+        .fn()
+        .mockRejectedValue(
+          new Error(
+            'Document in table "rateLimitShards" changed while this mutation was being run',
+          ),
+        ),
+    } as unknown as Parameters<typeof applyRateLimit>[0];
+    const request = new Request("https://example.com", {
+      headers: { "cf-connecting-ip": "203.0.113.1" },
+    });
+
+    const result = await applyRateLimit(ctx, request, "download");
+
+    expect(result.ok).toBe(false);
+    if (result.ok) return;
+    expect(result.response.status).toBe(429);
+    expect(result.response.headers.get("Retry-After")).toBe("30");
+  });
+
   it("allows authenticated users when user bucket is healthy and shared ip bucket is exhausted", async () => {
     vi.spyOn(Date, "now").mockReturnValue(3_000_000);
     const ctx = makeRateLimitCtx({

convex/lib/httpRateLimit.ts

@@ -4,7 +4,7 @@ import { getOptionalApiTokenUserId } from "./apiTokenAuth";
 import { corsHeaders, mergeHeaders } from "./httpHeaders";
 
 const RATE_LIMIT_WINDOW_MS = 60_000;
-const RATE_LIMIT_SHARDS = 16;
+const RATE_LIMIT_SHARDS = 64;
 export const RATE_LIMITS = {
   read: { ip: 600, key: 2400 },
   write: { ip: 45, key: 180 },
@@ -235,7 +235,7 @@ function shouldTrustForwardedIps() {
 function isRateLimitWriteConflict(error: unknown) {
   if (!(error instanceof Error)) return false;
   return (
-    error.message.includes("rateLimits") &&
+    (error.message.includes("rateLimits") || error.message.includes("rateLimitShards")) &&
     error.message.includes("changed while this mutation was being run")
   );
 }

convex/llmEval.ts

@@ -554,7 +554,7 @@ export const backfillLlmEval: ReturnType<typeof internalAction> = internalAction
       return { error: "OPENAI_API_KEY not configured" };
     }
 
-    const requestedBatchSize = Math.max(1, Math.floor(args.batchSize ?? 25));
+    const requestedBatchSize = Math.max(1, Math.min(Math.floor(args.batchSize ?? 25), 50));
     const maxToSchedule =
       args.maxToSchedule === undefined ? undefined : Math.max(0, Math.floor(args.maxToSchedule));
     const cursor = args.cursor ?? 0;

convex/rateLimits.test.ts

@@ -22,14 +22,12 @@ const consumeHandler = (
 )._handler;
 
 describe("rate limit sharding", () => {
-  it("sums all shards when checking status", async () => {
+  it("sums shard rows without reading the legacy rateLimits table", async () => {
     const ctx = {
       db: {
-        query: vi.fn((table: string) => ({
+        query: vi.fn(() => ({
           withIndex: vi.fn(() => ({
-            collect: vi.fn(async () =>
-              table === "rateLimits" ? [{ count: 3 }] : [{ count: 4 }, { count: 5 }],
-            ),
+            collect: vi.fn(async () => [{ count: 4 }, { count: 5 }]),
           })),
         })),
       },
@@ -42,7 +40,9 @@ describe("rate limit sharding", () => {
     });
 
     expect(result.allowed).toBe(true);
-    expect(result.remaining).toBe(8);
+    expect(result.remaining).toBe(11);
+    expect(ctx.db.query).toHaveBeenCalledTimes(1);
+    expect(ctx.db.query).toHaveBeenCalledWith("rateLimitShards");
   });
 
   it("writes only the selected shard when consuming", async () => {
@@ -55,7 +55,7 @@ describe("rate limit sharding", () => {
           })),
         })),
       });
-      return { unique: vi.fn(async () => null) };
+      return { first: vi.fn(async () => null) };
     });
     const ctx = {
       db: {

convex/rateLimits.ts

@@ -19,16 +19,12 @@ export const getRateLimitStatusInternal = internalQuery({
       return { allowed: false, remaining: 0, limit: args.limit, resetAt };
     }
 
-    const legacyRows = await ctx.db
-      .query("rateLimits")
-      .withIndex("by_key_window", (q) => q.eq("key", args.key).eq("windowStart", windowStart))
-      .collect();
     const shardRows = await ctx.db
       .query("rateLimitShards")
       .withIndex("by_key_window", (q) => q.eq("key", args.key).eq("windowStart", windowStart))
       .collect();
 
-    const count = [...legacyRows, ...shardRows].reduce((sum, row) => sum + row.count, 0);
+    const count = shardRows.reduce((sum, row) => sum + row.count, 0);
     const allowed = count < args.limit;
     return {
       allowed,
@@ -61,7 +57,7 @@ export const consumeRateLimitInternal = internalMutation({
       .withIndex("by_key_window_shard", (q) =>
         q.eq("key", args.key).eq("windowStart", windowStart).eq("shard", shard),
       )
-      .unique();
+      .first();
 
     if (!existing) {
       await ctx.db.insert("rateLimitShards", {

convex/skills.ts

@@ -124,8 +124,8 @@ const MAX_LIST_LIMIT = 50;
 const MAX_PUBLIC_LIST_LIMIT = 200;
 const MAX_LIST_BULK_LIMIT = 200;
 const MAX_LIST_TAKE = 1000;
-const MAX_SKILL_CATALOG_SCAN_DOCUMENTS = 2_000;
-const MAX_SKILL_CATALOG_SCAN_PAGES = 20;
+const MAX_SKILL_CATALOG_SCAN_DOCUMENTS = 500;
+const MAX_SKILL_CATALOG_SCAN_PAGES = 6;
 const MAX_SKILL_CATALOG_SEARCH_PAGE_SIZE = 200;
 const HARD_DELETE_BATCH_SIZE = 100;
 const HARD_DELETE_VERSION_BATCH_SIZE = 10;
@@ -3906,7 +3906,7 @@ export const getActiveSkillBatchForLlmBackfillInternal = internalQuery({
     batchSize: v.optional(v.number()),
   },
   handler: async (ctx, args) => {
-    const batchSize = args.batchSize ?? 10;
+    const batchSize = clampInt(args.batchSize ?? 10, 1, 50);
     const cursor = args.cursor ?? 0;
 
     // Use built-in by_creation_time index for stable cursor-based pagination