perf(stats): cut the dashboard tick's page misses by 43%

[henry40408]

Summary

While a dashboard is open, the server builds its stats snapshot every 10 seconds. On a 370k-row database each tick read 8 304 pages cold. It now reads 4 755 (−43%).

• One scan for the summary. count_queries_multi_since (totals, blocks) and cache_stats_multi_since (cache hits, latency) each walked the same 30 days of idx_query_logs_ts_metrics. summary_multi_since answers both in one statement by putting the allowed-only filter inside each CASE instead of the WHERE. The two old functions had no other callers and are removed.
• A covering index for top upstreams. upstream and response_ms were in no index, so the query looked up the table row for every forwarded query in the last 24h. Migration 14 adds idx_query_logs_ts_upstream ON (timestamp, upstream, response_ms) WHERE upstream IS NOT NULL, and the query names it with INDEXED BY.

Why this index shape

Measured with page misses on the same database. Four candidates were compared: timestamp-first vs upstream-first, full vs partial. The "pages written" column counts WAL frames for one committed 500-row batch, averaged over three batches.

index	read (24h)	size	pages written / batch
none (baseline)	1608	—	53
(upstream, timestamp, response_ms)	328	9.1 MiB	71
(timestamp, upstream, response_ms)	317	9.1 MiB	58
(upstream, …) WHERE upstream IS NOT NULL	220	5.7 MiB	65
(timestamp, …) WHERE upstream IS NOT NULL	211	5.7 MiB	56

Partial, because 56% of rows (blocked or cached) have no upstream. Timestamp-first, because the logger appends at the newest end; an upstream-first index would split every batch across one insertion point per upstream.

One tick, before and after

Each statement runs on a fresh connection, as the read pool's 2 MiB caches effectively leave them. The new statements return exactly the same results as the old ones (diffed).

before count_1m 3
before count_multi 2152
before cache_multi 2152
before domains 1831
before clients 558
before up_old 1608
after  count_1m 3
after  summary 2152
after  domains 1831
after  clients 558
after  up_new 211
TICK before 8304 after 4755

Tests

• stats_page_miss_test: the_dashboard_summary_is_one_metrics_scan and the_top_upstreams_never_read_the_log_table both failed before the change (225 pages vs 111 for one scan; 1470 of 1892 pages) and pass after. The shared seed now forwards half its rows so the second test can measure anything.
• stats_db_test: summary_multi_since gives each window's figures and leaves blocked rows out of the latency average. An empty window reports zeroes.
• db::tests::migration_v14_adds_the_upstream_index: a version-13 database gains the index, and top upstreams still answers correctly through it.

E2E was not run locally; the dashboard's rendering is unchanged and CI runs the suite.

🤖 Generated with Claude Code

[codecov]

Codecov Report

❌ Patch coverage is 97.75281% with 2 lines in your changes missing coverage. Please review.
✅ Project coverage is 91.27%. Comparing base (55c68ea) to head (bba141b).

Files with missing lines	Patch %	Lines
src/db.rs	97.22%	2 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #276      +/-   ##
==========================================
+ Coverage   91.26%   91.27%   +0.01%     
==========================================
  Files          31       31              
  Lines       11329    11342      +13     
==========================================
+ Hits        10339    10352      +13     
  Misses        990      990              

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.
• 📦 JS Bundle Analysis: Save yourself from yourself by tracking and limiting bundle sizes in JS merges.