Temp tables in Postgres don't scale
Comments on “Temp tables in Postgres don't scale”
Claim-checked. Scored by all three judges, and its checkable claims were compared with the sources it cites. How far stories are checked
PostgresAI's Nikolay Samokhvalov measured CREATE TEMP TABLE run in every transaction on Postgres 18 (16 vCPU): 3783 TPS, against 17866 for a table kept per session and 86997 for a CTE. Each create and drop touches 31 catalog rows. With 5000 idle sessions and a 500 TPS target, temp-table DDL reached 127 and 249 TPS in two runs, as the shared invalidation queue became the limit. The figures come from the author's own experiments and could not be independently checked.
Why it ranked
No. 13 of 144 with a score of 0.66. The panel rated it highest on technical depth (0.96) and substantiated (0.92), lowest on audience breadth (0.48). Placed No. 13, No. 27 by score alone: the placement terms moved other stories past it.
- AI judgeJudge 2claude-haiku-5-5
- Detailed, measured PostgreSQL engineering study of temp table costs (catalog churn, sinval queue). Highly technical and practically actionable for database-backed builders, though somewhat promotional in parts and narrow in audience.
- AI judgeJudge 3gpt-5.6-luna
- Detailed PostgreSQL benchmarks, methodology, measurements, implementation mechanics, and actionable mitigations make this highly useful to database and infrastructure builders; scope is narrower than general tech, with modest product promotion.
Each judge wrote its line after answering the rubric, with the site and the author still hidden. Every answer behind this is in the table under Score.
Score
0.66total under rubric v2.3
- Significance0.56
- Reader interest0.80
- Audience breadth0.48
- Practical relevance0.82
- Substantiated0.92
- Technical depth0.96
- New information0.89
- Counts against: Promotional0.20
- Counts against: Overclaim0.22
Each judge’s answer per question (0–1), their mean and spread, and what the question adds to the score (weight × mean).
| Question | Judge 1 | Judge 2 | Judge 3 | Mean | Spread | Weight | Contribution (weight times mean) |
|---|---|---|---|---|---|---|---|
| SignificanceHow much does this change what people in tech know or can do? | 0.57 | 0.45 | 0.65 | 0.56 | 0.20 | +0.24 | +0.134 |
| Reader interestWould a well-informed senior engineer or tech founder want to read this today, and likely pass it on? | 0.76 | 0.82 | 0.82 | 0.80 | 0.06 | +0.26 | +0.208 |
| Audience breadthHow many people who build technology does it concern? | 0.54 | 0.55 | 0.36 | 0.48 | 0.19 | +0.13 | +0.063 |
| Practical relevanceHow useful is it to someone who builds software or tech products? | 0.83 | 0.80 | 0.84 | 0.82 | 0.04 | +0.22 | +0.181 |
| SubstantiatedHow well is it backed? Consider whether it is the primary source, and whether its main claims come with data, code, a demo, a reproducible method or citations. | 0.98 | 0.85 | 0.93 | 0.92 | 0.13 | +0.05 | +0.046 |
| Technical depthHow technically substantive is it? | 1.00 | 0.95 | 0.94 | 0.96 | 0.06 | +0.05 | +0.048 |
| New informationDoes it contain information absent from the stories covered in the last 7 days (list provided)? | 0.86 | 0.90 | 0.92 | 0.89 | 0.06 | +0.05 | +0.045 |
| PromotionalIs it mainly marketing with little substance (a press release, ad or sponsored post with no real information)? A launch post that carries real news is not promotional. | 0.06 | 0.35 | 0.18 | 0.20 | 0.29 | −0.19 | −0.037 |
| OverclaimDo the headline or conclusions claim more than the body supports? | 0.25 | 0.30 | 0.12 | 0.22 | 0.18 | −0.11 | −0.025 |
| Sum of contributions | +0.663 | ||||||
| Claim-check penalty | 0.000 | ||||||
| Total, as published | 0.663 | ||||||
Spread is the highest minus the lowest judge answer. marks a spread above 0.40: the judges disagree, and the story is marked contested.
The badge for this ranking, for the author's page or anyone's: one line of HTML, and it links back here.
<a href="https://rubric.hrih.si/editions/2026-11-19/13/"><img src="https://rubric.hrih.si/badge/2026-11-19/13.svg" alt="Ranked #13 on Rubric, 19 Nov 2026" width="245" height="28"></a>Claim check
Claim check: 2 matches source · 2 no source given · 4 couldn’t check. Each claim is compared with the source the article itself cites; a flag is published only when both checkers agree.
Couldn’t checkBenchmark result
On Postgres 18 with 16 vCPU, a temp table per transaction ran at 3783 TPS, a temp table kept per session at 17866 TPS, and a CTE doing the same work at 86997 TPS.
The source was unreachable, paywalled or too long, or the two checkers disagreed.
The article says “A temp table per transaction ran at 3783 TPS, one that each session keeps at 17866, and a CTE doing the same work at 86997 (16 vCPU, no idle sessions”
The source says No passage could be compared.
Source cited by the article: console.postgres.ai/shared/brief/dfe99713-f5a8-483b-9aab-1606a883b75d
Couldn’t checkBenchmark result
With 5000 idle sessions and a 500 TPS target, temp-table DDL achieved 127 TPS in one run and 249 TPS in another.
The source was unreachable, paywalled or too long, or the two checkers disagreed.
The article says “With 5000 idle sessions and a 500 TPS target, temp-table DDL got 127 TPS in one run and 249 in another”
The source says No passage could be compared.
Source cited by the article: console.postgres.ai/shared/brief/dfe99713-f5a8-483b-9aab-1606a883b75d
Couldn’t checkNumber
Each create and drop of a 2-column temp table inserts and deletes 31 catalog rows and creates and removes 3 files.
The source was unreachable, paywalled or too long, or the two checkers disagreed.
The article says “Each create and drop of a 2-column temp table inserts and deletes 31 catalog rows and creates and removes 3 files”
The source says No passage could be compared.
Source cited by the article: console.postgres.ai/shared/brief/dfe99713-f5a8-483b-9aab-1606a883b75d
No source givenNumber
One transaction that creates and drops the benchmark temp table sends 132 invalidation messages on Postgres 18.6 and 17.11.
The article gives no link or citation for it.
The article says “One transaction of the benchmark, which creates and drops the table (id int, v text), sends 132 invalidation messages”
The source says No source to compare with.
Matches sourceBenchmark result
Tom Lane's 2008 sinval rewrite benchmark with 40 clients creating and dropping a temp table went from 485 TPS before the rewrite to 700 after.
The cited source says the same thing.
The article says “Its benchmark[11] was 40 clients that each create and drop a temp table: 485 TPS before the rewrite, 700 after”
The source says “On my machine CVS HEAD gets about 485 transactions/sec on this; with the patch that improves to 700 tps”
Source cited by the article: postgresql.org/message-id/373.1213813174@sss.pgh.pa.us
No source givenDate
The invalidation queue size of 4096 slots was set in 1999 and kept by the 2008 rewrite.
The article gives no link or citation for it.
The article says “The size of 4096 slots is older. It was set in 1999, and the rewrite kept it”
The source says No source to compare with.
Couldn’t checkCausal conclusion
Without idle sessions, the shared invalidation queue was not the bottleneck: the temp-table pattern stopped scaling at about 3000 to 4000 TPS on 16 vCPU, with queue waits under 1% of active samples.
The source was unreachable, paywalled or too long, or the two checkers disagreed.
The article says “The temp-table pattern stopped scaling at about 3000 to 4000 TPS on 16 vCPU, and waits on the queue stayed under 1% of active samples”
The source says No passage could be compared.
Source cited by the article: console.postgres.ai/shared/brief/dfe99713-f5a8-483b-9aab-1606a883b75d
Matches sourceGeneralization
Autovacuum cannot vacuum or analyze temp tables, so the owning session must vacuum them itself.
The cited source says the same thing.
The article says “Autovacuum cannot vacuum or analyze it[8], so the owning session must run VACUUM itself”
The source says “The autovacuum daemon cannot access and therefore cannot vacuum or analyze temporary tables. For this reason, appropriate vacuum and analyze operations should be performed via session SQL commands”
Source cited by the article: postgresql.org/docs/current/sql-createtable.html
Also covered by
No other coverage was clustered with this story.
Elsewhere
Elsewhere: HN 2 pts, not on front page
Recorded for context only, checked 19 Nov 2026, 05:47 UTC. It played no part in the ranking.