Patch CI statistics

Pipeline reads CI wall-clock time. Corpus reads today's patchsets laid out over PostgreSQL's history.

range24h 7d 30d 90d all

granularityhour day week

Runs

10,919

terminal, in range

Success rate

84.7%

9,247 of 10,919

Median build

2m 52s

successful and failed alike

Median test

5m 18s

successful and failed alike

Median queue wait

0m 00s

queued to started

Infra errors

0.1%

7 runs

GitHub re-attempts

attempt > 1 on one run id

Runs by result

terminal runs only; timeouts fold into their failure band

Show data

When	Result	Runs
10 Aug 00:00	build failed	327
10 Aug 00:00	cancelled	6
10 Aug 00:00	infra error	7
10 Aug 00:00	success	6,465
10 Aug 00:00	tests failed	843
17 Aug 00:00	build failed	141
17 Aug 00:00	success	2,774
17 Aug 00:00	tests failed	347

Success rate

hollow points sit below 5 terminal runs in the bucket

Show data

When	Success rate	Terminal runs
10 Aug 00:00	84.5%	7,648
17 Aug 00:00	85.0%	3,262

Build and test duration by era

median seconds; successful runs only

Show data

When	Era	Build median	Test median	Build p25-p75
17 Aug 00:00	bookworm	2m 55s	3m 39s	2m 48s - 3m 08s
10 Aug 00:00	trixie	2m 53s	5m 21s	2m 47s - 2m 55s
17 Aug 00:00	trixie	2m 52s	5m 19s	2m 46s - 2m 55s

ccache hit rate

median per bucket; the image's warm cache was baked against one reference commit, so a patch on a different base misses most of it

Show data

When	Hit rate	Runs
10 Aug 00:00	2.5%	6,465
17 Aug 00:00	1.5%	2,774

Build time against ccache hit rate

one point per successful run; table shows the first 50

Show data

Hit rate	Build
1.5%	2m 47s
1.5%	2m 48s
4.7%	2m 51s
4.7%	2m 50s
1.5%	2m 55s
4.7%	2m 54s
4.7%	2m 51s
4.7%	3m 04s
1.5%	2m 41s
4.7%	2m 52s
4.7%	2m 52s
1.5%	2m 48s
4.7%	1m 53s
10.5%	2m 48s
1.5%	2m 56s
1.5%	1m 58s
34.7%	0m 43s
1.7%	3m 08s
17.2%	2m 22s
1.5%	2m 55s
1.6%	2m 20s
1.6%	2m 55s
1.6%	3m 11s
1.6%	3m 09s
1.6%	3m 10s
1.6%	3m 07s
1.6%	2m 59s
1.5%	2m 25s
1.5%	2m 53s
1.6%	3m 07s
1.6%	2m 54s
1.6%	3m 09s
1.5%	3m 02s
1.2%	2m 39s
1.2%	2m 40s
1.5%	2m 48s
1.2%	2m 37s
1.5%	2m 21s
1.2%	2m 48s
1.5%	2m 54s
1.3%	2m 49s
1.3%	2m 49s
1.3%	2m 50s
1.3%	1m 45s
1.3%	1m 49s
1.5%	2m 55s
1.3%	2m 37s
1.3%	2m 38s
1.5%	2m 56s
1.5%	2m 46s

Queue latency

median seconds

Show data

When	Push to queued	Queued to started
10 Aug 00:00	-5767m 28s	0m 00s
17 Aug 00:00	0m 01s	0m 00s

Average concurrency

time-weighted average of runs in flight per bucket, against the 30-slot budget

Show data

When	Avg concurrency
10 Aug 00:00	13.2
17 Aug 00:00	4.1

Push lag

time from the patch being posted to CI queueing it

Show data

When	Lag	Runs
10 Aug 00:00	same day	124
10 Aug 00:00	days	378
10 Aug 00:00	weeks	782
10 Aug 00:00	months	2,898
10 Aug 00:00	older	3,466
17 Aug 00:00	same day	60
17 Aug 00:00	days	132
17 Aug 00:00	weeks	334
17 Aug 00:00	months	1,265
17 Aug 00:00	older	1,480

Push kind

first push, a re-push on a newer master, or a retry of the same head

Show data

When	Kind	Runs
10 Aug 00:00	first push	126
10 Aug 00:00	rebase re-push	7,518
10 Aug 00:00	retry	4
17 Aug 00:00	first push	261
17 Aug 00:00	rebase re-push	3,010

Suites failed per failing run

one suite is usually a patch bug; dozens is usually one crash cascading

Show data

Suites failed	Runs
1	399
2	57
3	27
4	348
5	74
6	18
7	46
8	17
9	6
10	7
11	13
12	11
13	4
21	7
31	1
33	7
55	1
56	1
58	6
59	7
62	7
68	1
76	7
83	7
102	3
107	4
111	1
122	6
123	1
129	6
131	3
132	1
133	1
140	1
155	3
156	1
157	2
158	2
200	3

Top failing suites

total failures, and in brackets the number of distinct patchsets - many failures on few patchsets means one broken patch retried; roughly one per patchset means either a flaky suite or a suite that reruns others and inherits their failures

recovery/027_stream_regress

562 (91 branches)

test_plan_advice/001_replan_regress

520 (80 branches)

pg_upgrade/002_pg_upgrade

514 (85 branches)

regress/collate.icu.utf8

69 (15 branches)

regress/subselect

68 (14 branches)

regress/partition_join

62 (15 branches)

postgres_fdw/postgres_fdw

60 (9 branches)

regress/alter_table

60 (12 branches)

regress/copy2

60 (12 branches)

regress/join

58 (12 branches)

regress/partition_split

58 (10 branches)

isolation/horizons

57 (9 branches)

regress/generated_virtual

55 (11 branches)

regress/stats

54 (12 branches)

regress/plpgsql

52 (10 branches)

regress/predicate

52 (10 branches)

regress/with

52 (10 branches)

isolation/eval-plan-qual

50 (8 branches)

regress/aggregates

49 (10 branches)

regress/foreign_key

46 (10 branches)

Infra health

Everything here should be zero. Non-zero is worth a look, not a panic.

Condition	Count
infra errors	7
in flight past 48h	0
branches with a failed push	2
payloads we could not parse	0
runs with no payload	6
built but pushed no image	46

Retried heads

same head_sha pushed again, new run id

Verdict changed

100.0%

4 of 4 - meaningless until this denominator grows

discussionsall active only

Tracked patchsets

5,625

current, not superseded, all patchsets

Median base age

859 days

behind master, all patchsets

On recent master

34.2%

base 30 days or newer, all patchsets

Has a verdict

89.3%

terminal CI status, all patchsets

Has a runnable image

85.8%

counted by digest, all patchsets

recent

1,925

base age tier, all patchsets

stale

398

base age tier, all patchsets

ancient

3,224

base age tier, all patchsets

unknown

base age tier, all patchsets

Master re-check recency

when each patchset was last tried against master; overdue is past 26h with master moved since. Unlike the base age tiles above, this follows the discussions control.

current

behind

381

overdue

109

never

Base age distribution

all current patchsets against those in active discussions

Show data

Base age	All	Active discussions
<1d	1,615	318
1-7d	65	55
7-30d	245	99
30-90d	62	16
90d-1y	336	10
1-2y	363	2
2-5y	935	0
>5y	1,926	1
unknown	78	10

Outcome by base commit year

current patchsets, by the year of the master commit they apply to

Show data

Base year	Outcome	Patchsets
unknown	never applied	10
2021	success	1
2025	success	2
2025	tests failed	1
2026	success	422
2026	cancelled	1
2026	tests failed	31
2026	build failed	6
2026	infra error	1
2026	no verdict	6
2026	never applied	30

Success rate by base commit year

of patchsets carrying a verdict

Show data

Base year	Success rate	With a verdict
2021	100.0%	1
2025	66.7%	3
2026	91.6%	463

Apply failures by base commit year

patchsets that would not apply even to their own detected base - a patch-extraction / base-detection signal, not a staleness measure; the rate is in the table

Show data

Base year	Patchsets	Failed to apply	Failure rate
unknown	10	10	100.0%
2021	1	0	0.0%
2025	3	0	0.0%
2026	497	30	6.0%

Build and test time over PostgreSQL history

median seconds of successful runs, by the base commit's year

Show data

Base year	Build median	Test median
2021	2m 17s	3m 02s
2025	2m 32s	5m 43s
2026	2m 53s	5m 21s

Test suites over PostgreSQL history

median executed suites of successful runs, by the base commit's year

Show data

Base year	Suites median
2021	623
2025	905
2026	1019

Per PostgreSQL major

A major whose era image is still a disabled stub is marked - those patchsets sit in won't-retry until the image exists.

PG	Era	Patchsets	Runs	Success	Build median	Test median	Suites	Images
15	bullseye	1	1	100.0%	2m 17s	3m 02s	623	1
19	trixie	16	15	80.0%	2m 47s	4m 45s	997	14
20	trixie	483	4,430	91.0%	2m 52s	5m 21s	1019	438

Files the corpus conflicts on

from apply-stage failures; in brackets the number of distinct patchsets

src/backend/utils/adt/float.c

3 (3 branches)

src/include/utils/float.h

3 (3 branches)

doc/src/sgml/release-18.sgml

2 (2 branches)

doc/src/sgml/release-19.sgml

2 (2 branches)

src/backend/commands/publicationcmds.c

2 (2 branches)

contrib/pg_buffercache/pg_buffercache_pages.c

1 (1 branches)

doc/src/sgml/perform.sgml

1 (1 branches)

doc/src/sgml/protocol.sgml

1 (1 branches)

doc/src/sgml/ref/explain.sgml

1 (1 branches)

src/backend/access/common/reloptions.c

1 (1 branches)

src/backend/access/common/toast_internals.c

1 (1 branches)

src/backend/access/heap/heapam.c

1 (1 branches)

src/backend/access/table/toast_helper.c

1 (1 branches)

src/backend/optimizer/path/allpaths.c

1 (1 branches)

src/backend/optimizer/path/joinpath.c

1 (1 branches)

src/backend/optimizer/plan/subselect.c

1 (1 branches)

src/backend/optimizer/prep/prepjointree.c

1 (1 branches)

src/backend/optimizer/util/pathnode.c

1 (1 branches)

src/backend/postmaster/checkpointer.c

1 (1 branches)

src/backend/replication/logical/relation.c

1 (1 branches)

Verdict coverage in active discussions

patchsets in threads with a message in the last 30 days

Show data

Era	With a verdict	No verdict
bullseye	1	0
trixie	466	33
unknown	0	11

Patch size against outcome

files touched per patchset, from the recorded submission paths

Show data

Files touched	Outcome	Patchsets
unknown	success	425
unknown	cancelled	1
unknown	tests failed	32
unknown	build failed	6
unknown	infra error	1
unknown	no verdict	6
unknown	never applied	40

Patchset versions per topic

every version, superseded included; split by whether the topic was committed

Show data

Versions	Topic	Topics
1	not committed	4,980
1	committed	332
2	not committed	157
2	committed	36
3	not committed	46
3	committed	8
4	not committed	14
4	committed	5
5	committed	2
5	not committed	8
6	committed	1
6	not committed	5
7	not committed	4
8	not committed	2
10	committed	1
11	not committed	1

Why the orchestrator won't retry

The dashboard shows the top three of this same breakdown.

Reason	Patchsets
base too old	3,118
committed	118
patchset touches .github/	11

Navigation

Patch CI statistics

Runs by result

Success rate

Build and test duration by era

ccache hit rate

Build time against ccache hit rate

Queue latency

Average concurrency

Push lag

Push kind

Suites failed per failing run

Top failing suites

Infra health

Master re-check recency

Base age distribution

Outcome by base commit year

Success rate by base commit year

Apply failures by base commit year

Build and test time over PostgreSQL history

Test suites over PostgreSQL history

Per PostgreSQL major

Files the corpus conflicts on

Verdict coverage in active discussions

Patch size against outcome

Patchset versions per topic

Why the orchestrator won't retry