track cache hit latency separately, show both on stats page

+3 -1

schema.sql

··· 45 45 CREATE TABLE IF NOT EXISTS metrics ( 46 46 hour INTEGER PRIMARY KEY, 47 47 searches INTEGER NOT NULL DEFAULT 0, 48 - total_ms REAL NOT NULL DEFAULT 0 48 + total_ms REAL NOT NULL DEFAULT 0, 49 + cache_hits INTEGER NOT NULL DEFAULT 0, 50 + cache_ms REAL NOT NULL DEFAULT 0 49 51 ); 50 52 51 53 CREATE TABLE IF NOT EXISTS snapshots (

+1 -1

src/enrichment.ts

··· 26 26 /** materialize stats data to KV so /stats never queries turso on the request path */ 27 27 export async function materializeStats(db: TursoDB, env: Env): Promise<void> { 28 28 const [metricsRes, snapshotRes, deltasRes] = await db.batch([ 29 - db.prepare("SELECT hour, searches, total_ms FROM metrics ORDER BY hour DESC LIMIT 2016"), 29 + db.prepare("SELECT hour, searches, total_ms, cache_hits, cache_ms FROM metrics ORDER BY hour DESC LIMIT 2016"), 30 30 db.prepare("SELECT hour, total, with_handles, with_avatars, hidden FROM snapshots ORDER BY hour ASC LIMIT 2000"), 31 31 db.prepare("SELECT bucket, actors_delta, handles_delta, avatars_delta FROM actor_deltas ORDER BY bucket DESC LIMIT 288"), 32 32 ], "read");

+4 -3

src/handlers/search.ts

··· 1 1 import type { TursoDB } from "../db"; 2 2 import type { ActorRow, Env } from "../types"; 3 3 import { json, sanitize, avatarUrl } from "../utils"; 4 - import { recordMetric, recordTrafficSource } from "../metrics"; 4 + import { recordMetric, recordCacheHit, recordTrafficSource } from "../metrics"; 5 5 import { throttledBackfill } from "../backfill"; 6 6 7 7 export async function handleSearch( ··· 28 28 return json({ actors: [] }); 29 29 } 30 30 31 + const t0 = Date.now(); 32 + 31 33 // cache API — edge cache for hot queries 32 34 const cacheKey = new Request( 33 35 `https://typeahead-cache/${encodeURIComponent(term)}:${limit}`, ··· 36 38 const cache = caches.default; 37 39 const cached = await cache.match(cacheKey); 38 40 if (cached) { 41 + ctx.waitUntil(recordCacheHit(db, Date.now() - t0)); 39 42 return cached; 40 43 } 41 - 42 - const t0 = Date.now(); 43 44 44 45 // 3-tier ranking: exact handle → handle prefix → FTS prefix 45 46 const ftsQuery = `"${term}"*`;

+6 -3

src/handlers/stats.ts

··· 19 19 20 20 // materialized data from KV (written by cron) — never hits turso 21 21 const kvData = await env.KV.get("stats_data", "json") as { 22 - metrics: { hour: number; searches: number; total_ms: number }[]; 22 + metrics: { hour: number; searches: number; total_ms: number; cache_hits: number; cache_ms: number }[]; 23 23 snapshots: { hour: number; total: number; with_handles: number; with_avatars: number; hidden: number }[]; 24 24 deltas: { bucket: number; actors_delta: number; handles_delta: number; avatars_delta: number }[]; 25 25 materialized_at: number; ··· 70 70 const withAvatars = latest.with_avatars; 71 71 const hiddenCount = dbSnapshots.length > 0 ? (dbSnapshots[dbSnapshots.length - 1].hidden ?? 0) : 0; 72 72 73 - const totalSearches = rows.reduce((s, r) => s + r.searches, 0); 73 + const totalSearches = rows.reduce((s, r) => s + r.searches + (r.cache_hits ?? 0), 0); 74 74 // avg latency over last 24h only (288 five-min buckets) so transient spikes age out quickly 75 75 const recentRows = rows.slice(0, 288); 76 76 const recentSearches = recentRows.reduce((s, r) => s + r.searches, 0); 77 77 const recentMs = recentRows.reduce((s, r) => s + r.total_ms, 0); 78 78 const avgLatency = recentSearches > 0 ? recentMs / recentSearches : 0; 79 + const recentCacheHits = recentRows.reduce((s, r) => s + (r.cache_hits ?? 0), 0); 80 + const recentCacheMs = recentRows.reduce((s, r) => s + (r.cache_ms ?? 0), 0); 81 + const avgCacheLatency = recentCacheHits > 0 ? recentCacheMs / recentCacheHits : 0; 79 82 const handlePct = total > 0 ? ((withHandles / total) * 100).toFixed(1) : "0"; 80 83 const avatarPct = total > 0 ? ((withAvatars / total) * 100).toFixed(1) : "0"; 81 84 82 85 const tProcess = performance.now(); 83 86 84 - const body = statsPage({ total, hiddenCount, rows, totalSearches, avgLatency, handlePct, avatarPct, snapshots, trafficSources }); 87 + const body = statsPage({ total, hiddenCount, rows, totalSearches, avgLatency, avgCacheLatency, handlePct, avatarPct, snapshots, trafficSources }); 85 88 86 89 const tRender = performance.now(); 87 90

+13

src/metrics.ts

··· 14 14 .run(); 15 15 } 16 16 17 + export async function recordCacheHit(db: TursoDB, ms: number): Promise<void> { 18 + const bucket = Math.floor(Date.now() / 300_000); 19 + await db.prepare( 20 + `INSERT INTO metrics (hour, cache_hits, cache_ms) 21 + VALUES (?1, 1, ?2) 22 + ON CONFLICT(hour) DO UPDATE SET 23 + cache_hits = cache_hits + 1, 24 + cache_ms = cache_ms + ?2` 25 + ) 26 + .bind(bucket, ms) 27 + .run(); 28 + } 29 + 17 30 /** fire-and-forget: record actor count deltas at 5-min granularity */ 18 31 export async function recordActorDelta( 19 32 db: TursoDB,

+4 -3

src/pages/stats.ts

··· 10 10 export interface StatsData { 11 11 total: number; 12 12 hiddenCount: number; 13 - rows: { hour: number; searches: number; total_ms: number }[]; 13 + rows: { hour: number; searches: number; total_ms: number; cache_hits: number; cache_ms: number }[]; 14 14 totalSearches: number; 15 15 avgLatency: number; 16 + avgCacheLatency: number; 16 17 handlePct: string; 17 18 avatarPct: string; 18 19 snapshots: SnapshotPoint[]; ··· 187 188 <div class="value">${d.totalSearches.toLocaleString()}</div> 188 189 </div> 189 190 <div class="metric"> 190 - <div class="label" data-tip="average response time for uncached searches (last 24h)">avg latency</div> 191 - <div class="value">${d.avgLatency.toFixed(1)} ms</div> 191 + <div class="label" data-tip="last 24h — cached (edge) vs uncached (turso)">avg latency</div> 192 + <div class="value" style="font-size:1.3rem">${d.avgCacheLatency.toFixed(0)} ms <span style="opacity:.4;font-size:.85rem">cached</span> · ${d.avgLatency.toFixed(0)} ms <span style="opacity:.4;font-size:.85rem">cold</span></div> 192 193 </div> 193 194 </div> 194 195

Configure Feed

Configure Feed