Hacktoberfest 2026: the issues maintainers tagged for October, open and beginner-friendly. Browse Hacktoberfest issues

Hold mutation leases across operator appends and cell writes (residual #2037 per-row lease cost)

Closed
  • Is something wrong?
#2,072 1 comment 0 reactions 0 assignees View on GitHub

Maintainers usually reply within 1 day

Nobody has claimed this yet.

Assessment

Difficulty
4/5
Estimated time
3-5 days
Newbie friendliness
35/100
Issue type
Refactor
Clarity
Mostly clear
Activity status
Active
Tech stack
c
Domain
backend, performance, testing

Research direction

Start by tracing the mutation-lease implementation from relation.c, then inspect the listed operator call sites in join.c, ops.c, inline.c, eval.c, kfusion.c, and session.c; compare with the held-lease batch appender used by filter.c. Run the sanitizer debug session test pinned to 2 CPUs and inspect mutation-set acquisition counts using wl_columnar_relation_test_mutation_nonce_peek. Done means converted operator tests reject per-row or per-cell acquisitions, outputs remain bit-identical, and the specified denial and renewal-failure cases cleanly destroy their outputs.

Written by the indexing model from the issue text.

Description

Discovery

Found while tracing the main sanitizer CI timeouts (Sanitizers / ubuntu-latest / gcc: tdd_recursive 30 s and session 180 s on 2-vCPU hosted runners) to the #2037 mutation-lease series. Since that series, the public single-relation mutators acquire, validate and finish a full mutation set on every call:

  • col_rel_append_row (relation.c): one set per appended row.
  • col_rel_set (relation.c, from df9c7be8 "Guard single-relation payload mutation with leases"): one set per cell.

The fix for the filter operator adds a held-lease batch appender (col_rel_append_batch_begin/row/end) and moves every filter.c fill onto it. Pinned to 2 CPUs, sanitizer debug:

test a8b5b5dd (pre-#2037) main 04638f77 with the filter fix
tdd_recursive 8.8 s 26.0 s 10.9 s
session 61.6 s 187.0 s 104.7 s

session is still about 1.7x its baseline. With the filter fix applied, perf on session still shows col_rel_mutation_set_acquire at 12.7 % and col_rel_mutation_lease_validate at 13.9 %. 12.5 % is reached through wl_columnar_join_diff_op, whose col_join_append_pair writes each output cell with col_rel_set.

Remaining per-row and per-cell lease callers on operator paths

At origin/main 04638f77:

  • col_rel_set, once per cell: join.c col_join_append_pair (1133, 1141, 1146); ops.c aggregation (796–873); inline.c 136, 362.
  • col_rel_append_row, once per row: join.c 1166 (col_join_append_pair growth fallback), 2478 (antijoin), 2932 (semijoin); ops.c 443, 902, 1096; eval.c 819, 2468; kfusion.c 167; session.c 3998, 5386.

Scope

Move the operator paths among these onto a held lease. Use the batch appender for appends. For per-cell writes, add a leased equivalent of col_rel_set, or write each row under the batch. Keep the lease contract intact, keep results bit-identical, and leave memory admission unchanged. Treat a hosted CI session time above about 150 s as the trigger to do this sooner.

Acceptance criteria

  • session pinned to 2 CPUs in the sanitizer debug build is back near its pre-#2037 time, measured against a8b5b5dd.
  • Each converted operator has a test that counts mutation-set acquisitions per call (wl_columnar_relation_test_mutation_nonce_peek) and fails on per-row or per-cell acquisition.
  • A mid-batch denial test, raised in review of the filter fix: a governor budget that refuses the second growth inside wl_columnar_filter_op (timestamped and slow paths) and fill_filtered_rel. It asserts ENOSPC, sess->memory_budget_denied, col_rel_destroy_checked(out) == 0 and no leak under ASan. Also one lease-renewal failure (nonce seeded near exhaustion) leaving the batch inactive with a clean destroy.
Dominant language
C
Stars
18
Forks
2
Avg merge
10h 9m
Merged PRs (30d)
173

Getting set up

First steps

  1. Read the whole issue, then the project's contributing guide.
  2. Comment on the issue to say you are picking it up — it saves two people doing the same work.
  3. Fork the repository and make your change on a branch.
  4. Open a pull request that references the issue number.

More from semantic-reasoning/wirelog

All issues in semantic-reasoning/wirelog

Similar issues

More C issues

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.