perf: add synchronous(Normal) + optimize_on_close to record store SQLite

[markovejnovic]

This PR adds two commits:

• The first commit adds benchmarking support to crates/atuin-client.
• The second commit adds a tiny performance optimization to record/sqlite_store.rs which requests sqlite pragma optimize on close as well as prefers the WAL rather than immediate fsync.

Hand-rolled PR (I hated claude's output).

Benchmarking

I used divan since this was the existing precedent. To run benchmarks:

cargo bench -p atuin-client

I think this is a little bit naive, and ideally we'd have rich benchmarking in CI, but I think that could be considered a future improvement.

If you think it's worth dropping the benchmarking addition from this PR (in lieu of a future, better, CI-integrated addition), I'm totally down to do that =)

Had the directory structure mirror the source tree

Optimization

There is a semantic change that occurred here -- in case the user's machine loses power, we risk losing the WAL on their machine, but:

• for the remote case, it's fine, since we'll pick up on those changes in the next sync with the remote.
• the local case does risk losing data, but this seems to be in-line with the history's database.

Results

Tested on an M4 mac.

At 171f7643d4337165ab96d15c010fb6638a3dfcb9 (before):

record_store         fastest       │ slowest       │ median        │ mean          │ samples │ iters
╰─ record                          │               │               │               │         │
   ╰─ sqlite_store                 │               │               │               │         │
      ╰─ push_batch                │               │               │               │         │
         ├─ 1        138.6 µs      │ 5.801 ms      │ 247.8 µs      │ 280.1 µs      │ 2504    │ 2504
         ├─ 10       177.7 µs      │ 5.204 ms      │ 200.4 µs      │ 215.3 µs      │ 3842    │ 3842
         ╰─ 100      617.7 µs      │ 8.771 ms      │ 718.1 µs      │ 779 µs        │ 2430    │ 2430

At 4f30c909f459676857e564b338cfd2622794714f (after):

record_store         fastest       │ slowest       │ median        │ mean          │ samples │ iters
╰─ record                          │               │               │               │         │
   ╰─ sqlite_store                 │               │               │               │         │
      ╰─ push_batch                │               │               │               │         │
         ├─ 1        98.83 µs      │ 5.41 ms       │ 183.7 µs      │ 189.3 µs      │ 4770    │ 4770
         ├─ 10       151.6 µs      │ 6.139 ms      │ 170.2 µs      │ 186 µs        │ 4736    │ 4736
         ╰─ 100      578.1 µs      │ 8.483 ms      │ 661 µs        │ 704.7 µs      │ 2890    │ 2890

I had claude generate a comparison table:

  ┌────────────────┬─────────────────┬────────────────┬────────┐
  │   Benchmark    │ Before (median) │ After (median) │ Change │
  ├────────────────┼─────────────────┼────────────────┼────────┤
  │ push_batch/1   │ 247.8 µs        │ 183.7 µs       │ -26%   │
  ├────────────────┼─────────────────┼────────────────┼────────┤
  │ push_batch/10  │ 200.4 µs        │ 170.2 µs       │ -15%   │
  ├────────────────┼─────────────────┼────────────────┼────────┤
  │ push_batch/100 │ 718.1 µs        │ 661.0 µs       │ -8%    │
  └────────────────┴─────────────────┴────────────────┴────────┘

Worth noting that my machine appears to be very noisy. I think less noisy benchmarks require we have a hosted bare-metal machine for this, specifically designed for benchmarking, but I think this is sufficient proof that the flags I've added are valuable.

Checks

• I am happy for maintainers to push small adjustments to this PR, to speed up the review cycle
• I have checked that there are no existing pull requests for the same thing

[github-actions]

Fossier: Manual Review Requested

@markovejnovic is a new contributor. A maintainer should review this PR before merging.

Score Breakdown

Total Score: 52.6/100 | Confidence: 100% | Outcome: REVIEW

Signal	Value	Score	Weight
account_age	3914	1.00	0.09
public_repos	85	1.00	0.05
contribution_history	96	0.48	0.05
follower_ratio	1.39	0.70	0.05
bot_signals	False	0.50	0.04
open_prs_elsewhere	37	0.00	0.09
closed_prs_elsewhere	49	0.00	0.10
merged_prs_elsewhere	298	1.00	0.08
prior_interaction	0	0.00	0.08
activity_velocity	4	0.33	0.08
pr_content	...	1.00	0.08
commit_email	marko@harmont.dev	0.80	0.04
pr_description	...	0.80	0.05
repo_stars	30469	0.30	0.04
org_membership	0	0.20	0.03
commit_verification	...	0.65	0.04
contributor_stars	25	0.50	0.04

[greptile-apps]

Greptile Summary

This PR brings record/sqlite_store.rs in line with the existing history database (database.rs) by adding optimize_on_close and synchronous=Normal — both already used there — and adds a divan benchmark suite for push_batch latency.

• The SQLite change is a safe, well-precedented two-liner: WAL + synchronous=Normal is explicitly recommended by SQLite docs, and optimize_on_close is a no-op on a cold database.
• The benchmark wires up divan correctly (dev-dependency, harness = false, temp-dir lifecycle managed via _temp_dir), but each iteration creates and tears down a fresh database, so it measures cold-start write performance rather than steady-state.

Confidence Score: 5/5

Safe to merge — the SQLite tweak mirrors an identical configuration already in production use for the history database.

Both changes are low-risk: the pragma settings are already proven in database.rs, and the benchmark only affects cargo bench, not production paths.

No files require special attention.

Important Files Changed

Filename	Overview
crates/atuin-client/src/record/sqlite_store.rs	Adds optimize_on_close and synchronous=Normal — matches the existing history DB configuration in database.rs
crates/atuin-client/benches/record/sqlite_store.rs	New divan benchmark for push_batch; db is dropped inside bench_values timing window, so pool close (including optimize_on_close) is included in the measured time
crates/atuin-client/Cargo.toml	Correctly adds divan and tempfile to dev-dependencies and declares the bench harness
crates/atuin-client/benches/record_store.rs	Minimal divan entrypoint, correct
crates/atuin-client/benches/record/mod.rs	Module re-export, no issues

_{Reviews (2): Last reviewed commit: "perf: add synchronous(Normal) + optimize..." | Re-trigger Greptile}

[ellie]

looks good, thank you!

I think definitely a nice starting place for benchmarks for us, but in the long term we need a stable CI run so that we can track things properly - as you suggested!