[jkaczman]

Previously, when a query ran on several shards, we only sorted the rows we got back from the shards by certain ORDER BY keys [columns, positions, vector distances]

If a different kind of key, one that wasn't a [column, position, vector distance], was present in ORDER BY, e.g., CASE ... END, we ignored it, even if the query also selected it, e.g., SELECT id, CASE ... END AS prio ... ORDER BY CASE ... END.

Now, we correctly handle such a sort key by adding it to the SELECT list as a helper column, sorting by it, and dropping it before sending the rows back

This also fixes two cases where we previously returned rows in the wrong order. Now, an explicit error is returned instead:

• If the expression has an aggregate in it, e.g., ORDER BY count(*).
• If the query has DISTINCT and sorts by an expression.

Fixes #1736

[codecov]

Codecov Report

❌ Patch coverage is 97.77778% with 2 lines in your changes missing coverage. Please review.

Files with missing lines	Patch %	Lines
...og/src/frontend/client/query_engine/route_query.rs	90.90%	1 Missing ⚠️
pgdog/src/frontend/router/parser/query/select.rs	83.33%	1 Missing ⚠️

📢 Thoughts on this report? Let us know!

[github-actions]

Bencher Report

Project	PgDog
Branch	jk-order-by-expressions
Testbed	Intel v1

⚠️ WARNING: Truncated view!

The full continuous benchmarking report exceeds the maximum length allowed on this platform.

🚨 2 Alerts

🐰 View full continuous benchmarking report in Bencher

[github-actions]

Metal bench failed. 1 of 150 measures outside bounds.

suite	throughput	cpu/txn	user cycles/txn	user instr/txn	kernel cycles/txn	ctx/txn
pooler		28.52 (+1.1% ± 0.9%)	29,944 (+1.5% ± 1.1%)	35,339 (-0.0% ± 0.03%)	38,369 (+1.2% ± 1.8%)
lb		29.47 (+0.2% ± 0.8%)	31,288 (-0.4% ± 1.0%)	36,205 (+0.0% ± 0.03%)	39,307 (+1.3% ± 1.5%)
sharding-w2		53.28 (+1.8%)	70,656 (+3.0%)	72,624 (+0.0%)	59,731 (+0.3%)	1.398 (-0.3%)
sharding-w8	155,002 (-0.1%)	47.25 (-0.8%)	69,299 (-1.8%)	56,295 (+0.0%)	52,493 (+0.4%)	0.1451 (+4.8%), bound +3%

Counters per txn

suite	frontend stall	l1i refill	backend stall
pooler	16,222 (+3.7% ± 2.2%)	3,502 (+1.9% ± 0.9%)	2,328 (-4.9% ± 1.6%)
lb	17,073 (-0.1% ± 2.0%)	3,586 (+0.6% ± 0.7%)	2,520 (-3.6% ± 2.3%)
sharding-w2	42,192 (+5.7%)	5,665 (+0.9%)	6,103 (-3.5%)
sharding-w8	39,750 (-4.5%)	5,252 (+0.7%)	9,405 (+6.1%)

Microbench: all 135 measures within bounds.

• extended_cached: 15 of 15 benches faster, median -3.1% ± 0.3%
• extended_cold: 15 of 15 benches faster, median -2.3% ± 0.3%
• simple_cold: 15 of 15 benches faster, median -2.8% ± 0.4%

lb flamegraph vs 1ab9634

Open interactive for hover, zoom and search.

Open interactive.

microbench flamegraph vs 1ab9634

Open interactive for hover, zoom and search.

Open interactive.

pooler flamegraph vs 1ab9634

Open interactive for hover, zoom and search.

Open interactive.

sharding-w2 flamegraph vs 1ab9634

Open interactive for hover, zoom and search.

Open interactive.

sharding-w8 flamegraph vs 1ab9634

Open interactive for hover, zoom and search.

Open interactive.

Logs

lb, microbench, pooler, sharding-w2, sharding-w8, verdict.json

Compared 22e8c67 with 1ab9634.