Continuous-testing failure-issue primitive + nightly fuzz job

[timsehn]

What

Sets up a reusable pattern for continuous (scheduled) test jobs that stay silent when they pass and cut a GitHub issue when they fail — no email infra required (GitHub's watcher notifications are the email path). Two pieces:

1. .github/actions/report-failure-issue (the reusable primitive)

A composite action that keeps one open tracking issue per label: it creates the issue if none is open, otherwise comments on it, so repeated failures append to a single triage thread instead of spamming. It auto-stamps the run URL, commit, ref, and workflow name into the body. It never auto-closes — a human closes after fixing.

Any future continuous job wires it up in one step:

- name: Report failure
  if: failure()          # or an explicit failure condition
  uses: ./.github/actions/report-failure-issue
  with:
    label: my-check
    title: "my-check failed"
    body-file: ${{ runner.temp }}/report.md
    token: ${{ secrets.GITHUB_TOKEN }}

This is the sqlite-upstream-drift issue handling, factored out and generalized.

2. .github/workflows/nightly-fuzz.yml (first consumer)

Runs the five on-disk-format libFuzzer harnesses — WAL replay, prolly node, refs blob, chunk index, catalog — for 10 min each nightly (07:00 UTC; duration configurable via workflow_dispatch). This is the "make fuzzing continuous" item: the harnesses already existed and ran 1 min/target per-PR in smoke.yml; this gives them sustained nightly runtime.

• Continues past a crash so one run reports every crashing target.
• On any crash: uploads reproducers as the fuzz-crashes artifact, files a fuzz-crash issue (with per-target ASan logs + base64 reproducers, and the bash test/run_fuzz.sh <target> <file> repro command), then fails the job.
• Build/run steps are identical to the proven smoke.yml fuzz job.

Testing

• Both YAML files parse.
• The novel logic was simulated locally with stub fuzzers: continue-on-crash loop produces the correct failed list, per-target reproducers land via -artifact_prefix, and the issue body composes correctly (made base64 portable via stdin redirect so it works on any runner).
• libFuzzer itself couldn't link locally (this machine's CLT clang lacks the fuzzer runtime), but the build/run steps are copied verbatim from smoke.yml, which passes on the Linux CI runners.

Note: the schedule only starts running once this is on the default branch; use the Run workflow button (workflow_dispatch) to trigger it on demand.

🤖 Generated with Claude Code

[github-actions]

Version-Control Performance Ceilings

Runs: median of 3 executions per benchmark, excluding fixture setup. The
working-set fixtures use 800 committed tables with 125 rows each
(100000 rows total); data-dirty cases update 1 row in 40
tables, and schema-dirty cases update 1 row plus add a column in 20 tables.
Branch tests use 300 branches, checkout uses a clean 400-table
branch switch over 100000 rows, and merge
tests use 100000-row tables with 2000 changed or conflicting
rows per side.

Benchmark	Median ms	Ceiling ms	Result
status_clean_many_tables	59	200	PASS
status_dirty_many_tables	62	200	PASS
diff_regular_working_one_table	55	150	PASS
diff_regular_working_many_tables	64	200	PASS
diff_stat_working_many_tables	65	200	PASS
diff_schema_working_many_tables	64	200	PASS
branch_list_many_branches	18	100	PASS
branch_create_delete	30	100	PASS
checkout_branch_clean	107	200	PASS
merge_data_no_conflicts	38	150	PASS
merge_schema_no_conflicts	19	100	PASS
merge_data_conflicts	68	250	PASS
merge_data_conflicts_with_resolve	68	250	PASS

[github-actions]

Sysbench-Style Benchmark (composite PK): Doltlite vs SQLite

Companion to the classic Sysbench-Style Benchmark. Every workload here
runs against tables with a 2-column INTEGER PRIMARY KEY(a, b) WITHOUT ROWID.
Individual ratios gated at 2.5×; section averages gated at 2×. Autocommit writes use 10× / 5×.

In-Memory

Reads

Test	SQLite (us)	Doltlite (us)	Multiplier
oltp_point_select	34,835	42,343	1.22
oltp_range_select	20,463	22,634	1.11
oltp_sum_range	18,959	21,922	1.16
oltp_order_range	3,665	3,986	1.09
oltp_distinct_range	4,720	5,240	1.11
oltp_index_scan	4,866	6,775	1.39
select_random_points	29,430	34,234	1.16
select_random_ranges	8,181	9,390	1.15
covering_index_scan	4,370	4,822	1.10
groupby_scan	37,224	41,749	1.12
index_join	8,447	11,698	1.38
index_join_scan	4,149	5,843	1.41
types_table_scan	1,171,025	1,385,367	1.18
table_scan	1,413,949	1,496,318	1.06
oltp_read_only	160,041	179,721	1.12
Average			1.18

Writes

Test	SQLite (us)	Doltlite (us)	Multiplier
oltp_bulk_insert	244,099	363,260	1.49
oltp_insert	19,375	37,842	1.95
oltp_update_index	72,552	128,777	1.77
oltp_update_non_index	54,018	87,060	1.61
oltp_delete_insert	52,066	102,909	1.98
oltp_write_only	28,847	60,683	2.10
types_delete_insert	34,902	58,425	1.67
oltp_read_write	115,312	168,606	1.46
Average			1.76

File-Backed

Reads

Test	SQLite (us)	Doltlite (us)	Multiplier
oltp_point_select	105,192	66,934	0.64
oltp_range_select	28,266	25,151	0.89
oltp_sum_range	26,810	24,673	0.92
oltp_order_range	4,463	4,299	0.96
oltp_distinct_range	5,637	5,613	1.00
oltp_index_scan	12,125	9,492	0.78
select_random_points	40,397	38,548	0.95
select_random_ranges	15,420	11,793	0.76
covering_index_scan	11,642	7,209	0.62
groupby_scan	37,423	42,123	1.13
index_join	12,611	13,512	1.07
index_join_scan	5,231	6,465	1.24
types_table_scan	1,161,563	1,378,050	1.19
table_scan	1,426,961	1,491,140	1.04
oltp_read_only	265,883	223,085	0.84
Average			0.94

Writes

Test	SQLite (us)	Doltlite (us)	Multiplier
oltp_bulk_insert	265,216	388,766	1.47
oltp_insert	27,064	47,979	1.77
oltp_update_index	104,790	157,413	1.50
oltp_update_non_index	78,988	109,567	1.39
oltp_delete_insert	83,301	124,963	1.50
oltp_write_only	55,088	81,877	1.49
types_delete_insert	52,352	73,563	1.41
oltp_read_write	144,365	189,415	1.31
Average			1.48

File-Backed (autocommit)

Each statement runs as its own transaction — exposes per-commit
fixed costs that the wrapped-in-BEGIN/COMMIT tests amortize away.
SQLite uses WAL mode with synchronous=FULL in this section so
the comparison uses SQLite's durable WAL autocommit path.

Reads

Reads have no commit cost; these are the same SQL files as the
File-Backed Reads section, included here for symmetry and to
catch any per-statement overhead doltlite pays on the read path.

Test	SQLite (us)	Doltlite (us)	Multiplier
oltp_point_select	59,465	67,686	1.14
oltp_range_select	23,580	25,645	1.09
oltp_sum_range	22,049	24,859	1.13
oltp_order_range	4,065	4,391	1.08
oltp_distinct_range	5,103	5,521	1.08
oltp_index_scan	7,575	9,478	1.25
select_random_points	33,199	38,561	1.16
select_random_ranges	10,815	11,876	1.10
covering_index_scan	7,094	7,404	1.04
groupby_scan	36,720	42,480	1.16
index_join	10,216	13,694	1.34
index_join_scan	4,836	6,507	1.35
types_table_scan	1,160,346	1,376,178	1.19
table_scan	1,442,768	1,493,710	1.04
oltp_read_only	195,797	221,008	1.13
Average			1.15

Writes

Test	SQLite (us)	Doltlite (us)	Multiplier
oltp_bulk_insert_ac	26,029	85,581	3.29
oltp_insert_ac	26,972	104,556	3.88
oltp_update_index_ac	29,284	119,570	4.08
oltp_update_non_index_ac	25,538	97,920	3.83
oltp_delete_insert_ac	25,929	111,139	4.29
oltp_write_only_ac	27,098	108,269	4.00
types_delete_insert_ac	22,061	96,100	4.36
oltp_read_write_ac	33,610	114,340	3.40
Average			3.89

100000 rows, median of 5 invocations per test; autocommit writes use 9, workload-only timing via host monotonic clock when available.

Performance Ceiling Check (2.5x individual, 2x average; autocommit writes: 10x / 5x)

All tests within ceilings.

[github-actions]

Sysbench-Style Benchmark: Doltlite vs SQLite

In-Memory

Reads

Test	SQLite (us)	Doltlite (us)	Multiplier
oltp_point_select	23,818	27,931	1.17
oltp_range_select	10,347	12,896	1.25
oltp_sum_range	9,136	12,325	1.35
oltp_order_range	2,615	3,016	1.15
oltp_distinct_range	3,625	4,183	1.15
oltp_index_scan	3,863	4,948	1.28
select_random_points	10,329	11,026	1.07
select_random_ranges	2,975	4,077	1.37
covering_index_scan	4,235	4,155	0.98
groupby_scan	31,496	35,922	1.14
index_join	5,813	7,783	1.34
index_join_scan	3,326	4,670	1.40
types_table_scan	1,107,182	1,326,303	1.20
table_scan	1,258,704	1,442,890	1.15
oltp_read_only	103,395	120,891	1.17
Average			1.21

Writes

Test	SQLite (us)	Doltlite (us)	Multiplier
oltp_bulk_insert	184,238	248,208	1.35
oltp_insert	15,784	28,149	1.78
oltp_update_index	51,085	92,117	1.80
oltp_update_non_index	35,792	59,254	1.66
oltp_delete_insert	45,327	71,278	1.57
oltp_write_only	21,793	44,695	2.05
types_delete_insert	24,864	39,565	1.59
oltp_read_write	63,384	105,869	1.67
Average			1.68

File-Backed

Reads

Test	SQLite (us)	Doltlite (us)	Multiplier
oltp_point_select	107,550	56,636	0.53
oltp_range_select	19,229	15,756	0.82
oltp_sum_range	18,175	15,485	0.85
oltp_order_range	3,585	3,375	0.94
oltp_distinct_range	4,631	4,584	0.99
oltp_index_scan	12,593	8,290	0.66
select_random_points	19,106	14,788	0.77
select_random_ranges	11,430	7,083	0.62
covering_index_scan	13,039	7,522	0.58
groupby_scan	32,537	36,641	1.13
index_join	10,479	9,810	0.94
index_join_scan	4,295	5,126	1.19
types_table_scan	1,106,261	1,332,089	1.20
table_scan	1,269,313	1,442,881	1.14
oltp_read_only	220,229	163,543	0.74
Average			0.87

Writes

Test	SQLite (us)	Doltlite (us)	Multiplier
oltp_bulk_insert	200,412	266,376	1.33
oltp_insert	22,176	35,254	1.59
oltp_update_index	75,617	113,911	1.51
oltp_update_non_index	56,927	79,688	1.40
oltp_delete_insert	64,188	93,099	1.45
oltp_write_only	42,393	64,246	1.52
types_delete_insert	39,772	52,827	1.33
oltp_read_write	84,792	125,746	1.48
Average			1.45

File-Backed (autocommit)

Each statement runs as its own transaction — exposes per-commit
fixed costs that the wrapped-in-BEGIN/COMMIT tests amortize away.
SQLite uses WAL mode with synchronous=FULL in this section so
the comparison uses SQLite's durable WAL autocommit path.

Reads

Reads have no commit cost; these are the same SQL files as the
File-Backed Reads section, included here for symmetry and to
catch any per-statement overhead doltlite pays on the read path.

Test	SQLite (us)	Doltlite (us)	Multiplier
oltp_point_select	51,047	56,529	1.11
oltp_range_select	13,357	15,745	1.18
oltp_sum_range	12,111	15,372	1.27
oltp_order_range	3,070	3,383	1.10
oltp_distinct_range	4,102	4,599	1.12
oltp_index_scan	6,865	8,429	1.23
select_random_points	13,562	14,690	1.08
select_random_ranges	5,784	7,109	1.23
covering_index_scan	7,146	7,442	1.04
groupby_scan	31,865	36,536	1.15
index_join	7,635	10,239	1.34
index_join_scan	3,824	5,285	1.38
types_table_scan	1,109,402	1,320,992	1.19
table_scan	1,267,171	1,439,982	1.14
oltp_read_only	141,420	163,542	1.16
Average			1.18

Writes

Test	SQLite (us)	Doltlite (us)	Multiplier
oltp_bulk_insert_ac	16,425	64,984	3.96
oltp_insert_ac	18,244	84,068	4.61
oltp_update_index_ac	21,624	100,197	4.63
oltp_update_non_index_ac	17,017	76,812	4.51
oltp_delete_insert_ac	18,926	87,650	4.63
oltp_write_only_ac	18,800	85,961	4.57
types_delete_insert_ac	16,168	76,234	4.72
oltp_read_write_ac	22,511	94,270	4.19
Average			4.48

100000 rows, median of 5 invocations per test; autocommit writes use 9, workload-only timing via host monotonic clock when available.

Performance Ceiling Check (2.5x individual, 2x average; autocommit writes: 10x / 5x)

All tests within ceilings.

[github-actions]

Sysbench-Style Benchmark (BLOB PK): Doltlite vs SQLite

Companion to the classic Sysbench-Style Benchmark. Every workload here
runs against tables with a 16-byte big-endian BLOB PRIMARY KEY.
Individual ratios gated at 2.5×; section averages gated at 2×. Autocommit writes use 10× / 5×.

In-Memory

Reads

Test	SQLite (us)	Doltlite (us)	Multiplier
oltp_point_select	30,597	37,431	1.22
oltp_range_select	12,555	14,936	1.19
oltp_sum_range	11,532	14,432	1.25
oltp_order_range	2,840	3,176	1.12
oltp_distinct_range	3,991	4,407	1.10
oltp_index_scan	4,540	6,451	1.42
select_random_points	18,004	21,173	1.18
select_random_ranges	4,148	5,440	1.31
covering_index_scan	4,393	4,764	1.08
groupby_scan	31,767	35,148	1.11
index_join	6,789	9,775	1.44
index_join_scan	4,244	5,851	1.38
types_table_scan	1,151,966	1,379,238	1.20
table_scan	1,376,470	1,477,919	1.07
oltp_read_only	129,622	139,808	1.08
Average			1.21

Writes

Test	SQLite (us)	Doltlite (us)	Multiplier
oltp_bulk_insert	244,545	362,091	1.48
oltp_insert	19,885	40,506	2.04
oltp_update_index	73,097	132,426	1.81
oltp_update_non_index	49,730	86,309	1.74
oltp_delete_insert	50,521	106,258	2.10
oltp_write_only	29,210	64,956	2.22
types_delete_insert	33,439	57,351	1.72
oltp_read_write	91,528	145,246	1.59
Average			1.84

File-Backed

Reads

Test	SQLite (us)	Doltlite (us)	Multiplier
oltp_point_select	102,082	62,709	0.61
oltp_range_select	21,637	17,895	0.83
oltp_sum_range	20,482	17,446	0.85
oltp_order_range	3,893	3,595	0.92
oltp_distinct_range	4,959	4,798	0.97
oltp_index_scan	12,167	9,231	0.76
select_random_points	27,637	24,629	0.89
select_random_ranges	11,430	7,832	0.69
covering_index_scan	12,268	7,371	0.60
groupby_scan	32,998	35,926	1.09
index_join	11,008	11,765	1.07
index_join_scan	5,211	6,090	1.17
types_table_scan	1,142,959	1,362,680	1.19
table_scan	1,431,406	1,503,959	1.05
oltp_read_only	231,966	181,848	0.78
Average			0.90

Writes

Test	SQLite (us)	Doltlite (us)	Multiplier
oltp_bulk_insert	258,946	381,161	1.47
oltp_insert	32,587	53,300	1.64
oltp_update_index	111,637	171,984	1.54
oltp_update_non_index	81,189	111,987	1.38
oltp_delete_insert	85,493	135,650	1.59
oltp_write_only	57,503	87,285	1.52
types_delete_insert	54,058	75,575	1.40
oltp_read_write	125,359	169,790	1.35
Average			1.49

File-Backed (autocommit)

Each statement runs as its own transaction — exposes per-commit
fixed costs that the wrapped-in-BEGIN/COMMIT tests amortize away.
SQLite uses WAL mode with synchronous=FULL in this section so
the comparison uses SQLite's durable WAL autocommit path.

Reads

Reads have no commit cost; these are the same SQL files as the
File-Backed Reads section, included here for symmetry and to
catch any per-statement overhead doltlite pays on the read path.

Test	SQLite (us)	Doltlite (us)	Multiplier
oltp_point_select	56,683	63,523	1.12
oltp_range_select	16,102	17,972	1.12
oltp_sum_range	15,658	17,520	1.12
oltp_order_range	3,366	3,524	1.05
oltp_distinct_range	4,456	4,765	1.07
oltp_index_scan	7,405	9,132	1.23
select_random_points	22,600	24,504	1.08
select_random_ranges	6,809	7,784	1.14
covering_index_scan	7,778	7,520	0.97
groupby_scan	32,225	35,906	1.11
index_join	9,292	11,601	1.25
index_join_scan	4,932	6,276	1.27
types_table_scan	1,155,950	1,377,707	1.19
table_scan	1,386,541	1,500,668	1.08
oltp_read_only	166,604	182,760	1.10
Average			1.13

Writes

Test	SQLite (us)	Doltlite (us)	Multiplier
oltp_bulk_insert_ac	21,736	76,591	3.52
oltp_insert_ac	24,893	99,949	4.02
oltp_update_index_ac	29,347	111,023	3.78
oltp_update_non_index_ac	23,072	90,792	3.94
oltp_delete_insert_ac	24,506	102,274	4.17
oltp_write_only_ac	25,128	99,715	3.97
types_delete_insert_ac	22,501	96,940	4.31
oltp_read_write_ac	28,787	104,544	3.63
Average			3.92

100000 rows, median of 5 invocations per test; autocommit writes use 9, workload-only timing via host monotonic clock when available.

Performance Ceiling Check (2.5x individual, 2x average; autocommit writes: 10x / 5x)

All tests within ceilings.

[github-actions]

Sysbench-Style Benchmark (TEXT PK): Doltlite vs SQLite

Companion to the classic Sysbench-Style Benchmark. Every workload here
runs against tables with a 32-char hex TEXT PRIMARY KEY (UUID-shaped).
Individual ratios gated at 2.5×; section averages gated at 2×. Autocommit writes use 10× / 5×.

In-Memory

Reads

Test	SQLite (us)	Doltlite (us)	Multiplier
oltp_point_select	29,675	38,478	1.30
oltp_range_select	13,525	14,992	1.11
oltp_sum_range	12,125	14,585	1.20
oltp_order_range	3,035	3,202	1.06
oltp_distinct_range	4,108	4,359	1.06
oltp_index_scan	4,624	6,405	1.39
select_random_points	18,122	21,385	1.18
select_random_ranges	4,074	5,403	1.33
covering_index_scan	4,630	4,667	1.01
groupby_scan	31,897	35,672	1.12
index_join	7,168	9,681	1.35
index_join_scan	4,580	5,595	1.22
types_table_scan	1,143,989	1,377,567	1.20
table_scan	1,351,029	1,457,518	1.08
oltp_read_only	123,647	141,976	1.15
Average			1.18

Writes

Test	SQLite (us)	Doltlite (us)	Multiplier
oltp_bulk_insert	233,506	361,726	1.55
oltp_insert	21,838	40,215	1.84
oltp_update_index	73,114	135,639	1.86
oltp_update_non_index	49,557	88,713	1.79
oltp_delete_insert	52,149	107,463	2.06
oltp_write_only	30,196	62,444	2.07
types_delete_insert	33,439	55,932	1.67
oltp_read_write	89,018	143,883	1.62
Average			1.81

File-Backed

Reads

Test	SQLite (us)	Doltlite (us)	Multiplier
oltp_point_select	100,073	63,139	0.63
oltp_range_select	21,133	17,753	0.84
oltp_sum_range	20,184	17,258	0.86
oltp_order_range	3,932	3,514	0.89
oltp_distinct_range	5,015	4,720	0.94
oltp_index_scan	12,044	9,189	0.76
select_random_points	27,345	25,015	0.91
select_random_ranges	11,328	7,794	0.69
covering_index_scan	12,732	7,507	0.59
groupby_scan	33,370	36,145	1.08
index_join	11,724	11,581	0.99
index_join_scan	5,620	6,165	1.10
types_table_scan	1,094,478	1,343,353	1.23
table_scan	1,408,633	1,461,807	1.04
oltp_read_only	225,591	179,544	0.80
Average			0.89

Writes

Test	SQLite (us)	Doltlite (us)	Multiplier
oltp_bulk_insert	250,797	389,757	1.55
oltp_insert	42,791	52,982	1.24
oltp_update_index	115,003	169,847	1.48
oltp_update_non_index	94,858	112,744	1.19
oltp_delete_insert	91,098	136,648	1.50
oltp_write_only	87,508	86,447	0.99
types_delete_insert	56,988	75,862	1.33
oltp_read_write	139,856	165,605	1.18
Average			1.31

File-Backed (autocommit)

Each statement runs as its own transaction — exposes per-commit
fixed costs that the wrapped-in-BEGIN/COMMIT tests amortize away.
SQLite uses WAL mode with synchronous=FULL in this section so
the comparison uses SQLite's durable WAL autocommit path.

Reads

Reads have no commit cost; these are the same SQL files as the
File-Backed Reads section, included here for symmetry and to
catch any per-statement overhead doltlite pays on the read path.

Test	SQLite (us)	Doltlite (us)	Multiplier
oltp_point_select	53,657	63,086	1.18
oltp_range_select	16,437	17,724	1.08
oltp_sum_range	15,079	17,695	1.17
oltp_order_range	3,426	3,507	1.02
oltp_distinct_range	4,494	4,692	1.04
oltp_index_scan	7,333	9,035	1.23
select_random_points	21,434	24,907	1.16
select_random_ranges	6,751	7,905	1.17
covering_index_scan	8,174	7,361	0.90
groupby_scan	32,817	35,996	1.10
index_join	9,299	11,536	1.24
index_join_scan	5,082	6,059	1.19
types_table_scan	1,122,443	1,342,568	1.20
table_scan	1,396,780	1,468,787	1.05
oltp_read_only	155,490	178,868	1.15
Average			1.13

Writes

Test	SQLite (us)	Doltlite (us)	Multiplier
oltp_bulk_insert_ac	22,280	82,558	3.71
oltp_insert_ac	25,759	97,297	3.78
oltp_update_index_ac	27,509	113,580	4.13
oltp_update_non_index_ac	23,039	96,123	4.17
oltp_delete_insert_ac	26,066	105,948	4.06
oltp_write_only_ac	26,678	105,486	3.95
types_delete_insert_ac	22,450	92,232	4.11
oltp_read_write_ac	33,543	112,858	3.36
Average			3.91

100000 rows, median of 5 invocations per test; autocommit writes use 9, workload-only timing via host monotonic clock when available.

Performance Ceiling Check (2.5x individual, 2x average; autocommit writes: 10x / 5x)

All tests within ceilings.