perf: remove redundant per-statement clock reads (6 -> 3) by jkaczman · Pull Request #1890 · pgdogdev/pgdog

jkaczman · 2026-10-23T11:38:14Z

Reduces clock reads per statement from six to three.

Now we:

Re-use the request's arrival time to set the transaction start time
Only compute the buffered-request elapsed time when debug logging is enabled
Skip the clock read when the server transitions to Active right after checkout, since checkout already started the query and transaction timers

The remaining three are from request arrival, pool checkout, and ReadyForQuery.

codecov · 2026-10-23T11:54:17Z

Codecov Report

❌ Patch coverage is 65.11628% with 15 lines in your changes missing coverage. Please review.

Files with missing lines	Patch %	Lines
pgdog/src/frontend/client/mod.rs	31.25%	11 Missing ⚠️
pgdog/src/frontend/client/query_engine/query.rs	66.66%	4 Missing ⚠️

📢 Thoughts on this report? Let us know!

github-actions · 2026-10-23T12:01:14Z

Metal bench passed. All 150 measures within bounds.

suite	throughput	cpu/txn	user cycles/txn	user instr/txn	kernel cycles/txn	ctx/txn
pooler		28.38 (+0.8% ± 1.0%)	29,497 (+0.8% ± 1.5%)	35,348 (-1.2% ± 0.02%)	38,295 (+0.8% ± 1.3%)
lb		29.25 (+0.5% ± 0.7%)	30,937 (+0.6% ± 0.8%)	36,215 (-1.2% ± 0.03%)	38,958 (+0.4% ± 1.4%)
sharding-w2		53.04 (+0.6%)	70,043 (+1.7%)	72,609 (-0.6%)	59,664 (-1.3%)	1.399 (-0.1%)
sharding-w8	156,287 (+1.4%)	46.9 (-1.6%)	68,399 (-2.1%)	56,300 (-0.7%)	52,456 (-1.1%)	0.1461 (+2.1%)

Counters per txn

suite	br mispred	l1i refill	backend stall
pooler	79.58 (-2.9% ± 2.5%)	3,427 (+1.5% ± 1.3%)	2,337 (-6.2% ± 1.9%)
lb	85.02 (-1.4% ± 1.4%)	3,510 (+0.9% ± 0.7%)	2,497 (-6.2% ± 2.4%)
sharding-w2	329 (+4.8%)	5,614 (+2.0%)	6,168 (-7.5%)
sharding-w8	334 (+0.9%)	5,249 (+0.8%)	8,571 (-10.6%)