fix(ccusage): parallelise all-agent loading - #1066

@coderabbitai please review this PR. The key change is that all-agent source loaders now run concurrently, while final row ordering remains deterministic.

Deploying with    Cloudflare Workers

The latest updates on your project. Learn more about integrating Git with Workers.

#!/bin/bash
# Fetch PR details
gh pr view 1066 --repo ryoppippi/ccusage --json title,body,files,additions,deletions,changedFiles

# Get the diff
gh pr diff 1066 --repo ryoppippi/ccusage

@ryoppippi: This is a clean, well-structured parallelisation. The use of thread::scope + mpsc::channel is idiomatic Rust for bounded concurrent work, and the regression test with AtomicUsize + spin-wait is an excellent way to assert that loaders actually overlap rather than serialise. A few observations below.

🟡 load_kind is an identity copy of kind

let load_kind = if kind == AgentReportKind::Session {
    AgentReportKind::Session
} else {
    AgentReportKind::Daily
};

This is just kind — the if-else returns the same variant you started with. If AgentReportKind: Copy, this can be replaced with:

let load_kind = kind;

or, if the only reason for the binding is to allow multiple closures to capture it without moving kind (which is also used later in if kind == AgentReportKind::Session and aggregate_rows(rows, kind)), the naming is fine but the if-else branch is redundant.

🟡 Test reuses UsageLoadAgent::Claude for both specs

AgentLoadSpec {
    index,
    agent,
    progress_agent: crate::progress::UsageLoadAgent::Claude, // ← both specs
    ...
}

load_agent_rows_parallel calls progress.start(spec.progress_agent) for each spec, so start(Claude) is called twice. Depending on the internals of UsageLoadProgress, this could silently swallow a double-start, mask a bug, or produce misleading terminal output during the test. Using distinct variants (e.g. Claude and Codex) would make the test more faithful.

🟢 Order preservation is correct

loaded.sort_by_key(|loaded| loaded.index) inside thread::scope after the channel drains guarantees output order matches the original specs slice regardless of which thread finishes first. ✓

🟢 Panic safety is correct

Because drop(sender) is called before the receiver loop, the receiver ends exactly when all spawned senders are dropped (either sent or panicked). The subsequent join loop catches panics via handle.join().is_err() and calls progress.fail() — each agent gets exactly one progress update in all paths. ✓

🟡 Error determinism: first-by-arrival, not first-by-index

When multiple agents fail, first_error captures whichever error arrives first on the channel (non-deterministic). If users expect the error message to consistently name the "first" agent (by declaration order), you might want to sort the errors by index before selecting. Low priority since one error is usually enough context, but worth noting.

Overall the change is a solid improvement — eliminates ~100 lines of duplicated sequential code, reduces wall-clock load time proportionally to the number of agents, and keeps output ordering deterministic. The two minor points above are easy to address but not blocking.

sequenceDiagram
  participant Main
  participant LoaderThread
  participant Channel
  participant Progress as ProgressAgent
  Main->>LoaderThread: spawn scoped thread with AgentLoadSpec.loader()
  LoaderThread->>Channel: send (index, agent, progress_agent, result)
  Main->>Channel: receive messages
  Main->>Progress: mark success/failure per agent
  Main->>Main: join threads, detect panics
  Main->>Main: sort by spec index and return results or first error

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

_{Comment @coderabbitai help to get the list of available commands and usage tips.}

@coderabbitai addressed the review notes in d0596c5.

• Rewrote the all-report load-kind mapping as an explicit match so weekly/monthly daily-row aggregation is clear.
• Updated the concurrency test to use distinct progress agents.
• Made multi-loader error selection deterministic by declaration index.

Validation after the fix:

• cargo test --manifest-path rust/Cargo.toml loads_agent_rows_concurrently
• pnpm typecheck
• pnpm run test
• pre-push clippy

Please review again.

#!/bin/bash
# Fetch the diff for the updated commit
gh pr diff 1066 --repo ryoppippi/ccusage

@ryoppippi: All three fixes look correct — verified against the diff for d0596c5.

1. Load-kind mapping — the explicit match now clearly documents that Weekly and Monthly share the Daily row-loading path. ✓

2. Distinct progress agents in test — Claude and Codex are now used as separate entries, making the test properly faithful to production behaviour. ✓

3. Deterministic error selection — errors.sort_by_key(|(index, _)| *index) before .into_iter().next() guarantees the first-by-declaration-order error is always surfaced. ✓

The implementation is clean and solid. Happy to approve.

(´｡• ᵕ •｡`) ♡

[approve]

@coderabbitai merged latest origin/main and resolved the conflict by keeping all-agent loading parallel while adding the new Kilo source to the concurrent loader set.

Validation after the merge:

• pnpm run format
• pnpm typecheck
• pnpm run test
• pre-push clippy

Please review the updated branch again.

#!/bin/bash
# Fetch the current diff for the updated PR branch
gh pr diff 1066 --repo ryoppippi/ccusage

@ryoppippi: The merge conflict resolution is clean. Here's what I verified:

1. Kilo correctly wired in — load_kilo_rows(load_kind, shared, &pricing) at index 5 is consistent with the other pricing-aware loaders (codex, amp, copilot, gemini). The load_kind mapping correctly passes AgentReportKind::Session for session reports and AgentReportKind::Daily for everything else, matching the semantics of the old sequential code. ✓

2. Agent ordering preserved — Indices 0–7 (claude → codex → opencode → amp → pi → kilo → copilot → gemini) are contiguous and consistent; the post-join sort-by-index will produce deterministic output order. ✓

3. Previously approved fixes intact — explicit match for load_kind, distinct Claude/Codex progress agents in the concurrency test, and deterministic error selection by index are all still in place. ✓

No issues introduced by the merge.

(´｡• ᵕ •｡`) ♡

[approve]

Open in StackBlitz

npx https://pkg.pr.new/ryoppippi/ccusage@1066

npx https://pkg.pr.new/ryoppippi/ccusage/@ccusage/ccusage-darwin-arm64@1066

npx https://pkg.pr.new/ryoppippi/ccusage/@ccusage/ccusage-darwin-x64@1066

npx https://pkg.pr.new/ryoppippi/ccusage/@ccusage/ccusage-linux-arm64@1066

npx https://pkg.pr.new/ryoppippi/ccusage/@ccusage/ccusage-linux-x64@1066

npx https://pkg.pr.new/ryoppippi/ccusage/@ccusage/ccusage-win32-arm64@1066

npx https://pkg.pr.new/ryoppippi/ccusage/@ccusage/ccusage-win32-x64@1066

commit: 9f7d1e7

ccusage performance comparison

PR SHA: a6a7e1c66d46
Base SHA: 35cb05e38459

This compares the Rust PR release binary against the configured base package on the same CI runner.

Package runner startup

Execution setup measures any pre-benchmark package materialization used by the execution benchmark. Bunx temp cache measures one bunx -p <url> ccusage --version run with an empty Bun install cache. Warm reuses that cache and reports the median of repeated runs.

Cached bunx execution performance

Runs the same large fixture through bunx -p <pkg.pr.new URL> ccusage after the Bun install cache has already been populated by the startup measurement. This separates cached package-runner execution from first-fetch package materialization.

Fixtures: Claude /home/runner/work/_temp/ccusage-large-fixture (1.01 GiB, 2,597 files), Codex /home/runner/work/_temp/ccusage-large-codex-fixture (1.01 GiB, 2,597 files)
Base package: 35cb05e38459; PR package: a6a7e1c66d46. Both run through bunx -p <pkg.pr.new URL> ccusage using the warmed Bun install cache from package runner startup, measured by hyperfine with 0 warmups and 1 runs.

Package runtime diagnostics

Compares the PR package wrapper, the installed native optional dependency binary, and the workspace release binary on the same large fixture. This identifies whether slow package results come from JavaScript wrapper overhead, the published native binary build, or the Rust core itself.

Fixtures: Claude /home/runner/work/_temp/ccusage-large-fixture (1.01 GiB, 2,597 files), Codex /home/runner/work/_temp/ccusage-large-codex-fixture (1.01 GiB, 2,597 files)
All rows run --offline --json, measured by hyperfine with 0 warmups and 1 runs. This isolates wrapper overhead from the installed native optional dependency and the workspace release binary built on the runner.

Committed fixture performance

Committed small fixtures for stable PR-to-PR feedback and explicit Claude/Codex command coverage.

Fixtures: Claude apps/ccusage/test/fixtures/claude (0.00 MiB, 2 files), Codex apps/ccusage/test/fixtures/codex (0.00 MiB, 1 files)
Base runs the published ccusage package from pkg.pr.new, installed before measurement; PR runs rust/target/release/ccusage directly. Both run --offline --json, measured by hyperfine with 2 warmups and 7 runs.

Large real-world-shaped fixture performance

Generated fixtures shaped from aggregate local log statistics: thousands of JSONL files, many small sessions, and a long tail of larger sessions. No real prompts, paths, or outputs are stored in the fixtures.

Fixtures: Claude /home/runner/work/_temp/ccusage-large-fixture (1.01 GiB, 2,597 files), Codex /home/runner/work/_temp/ccusage-large-codex-fixture (1.01 GiB, 2,597 files)
Base runs the published ccusage package from pkg.pr.new, installed before measurement; PR runs rust/target/release/ccusage directly. Both run --offline --json, measured by hyperfine with 0 warmups and 1 runs.

Artifact size

Lower medians and smaller artifacts are better. CI runner noise still applies; use same-run ratios as directional PR feedback, not release guarantees.

ccusage performance comparison

PR SHA: a6a7e1c66d46
Base SHA: 35cb05e38459

This compares the PR package against the configured base package on the same CI runner.

Package runner startup

Execution setup measures any pre-benchmark package materialization used by the execution benchmark. Bunx temp cache measures one bunx -p <url> ccusage --version run with an empty Bun install cache. Warm reuses that cache and reports the median of repeated runs.

Cached bunx execution performance

Runs the same large fixture through bunx -p <pkg.pr.new URL> ccusage after the Bun install cache has already been populated by the startup measurement. This separates cached package-runner execution from first-fetch package materialization.

Fixtures: Claude /home/runner/work/_temp/ccusage-large-fixture (1.01 GiB, 2,597 files), Codex /home/runner/work/_temp/ccusage-large-codex-fixture (1.01 GiB, 2,597 files)
Base package: 35cb05e38459; PR package: a6a7e1c66d46. Both run through bunx -p <pkg.pr.new URL> ccusage using the warmed Bun install cache from package runner startup, measured by hyperfine with 0 warmups and 1 runs.

Package runtime diagnostics

Compares the PR package wrapper, the installed native optional dependency binary, and the workspace release binary on the same large fixture. This identifies whether slow package results come from JavaScript wrapper overhead, the published native binary build, or the Rust core itself.

Fixtures: Claude /home/runner/work/_temp/ccusage-large-fixture (1.01 GiB, 2,597 files), Codex /home/runner/work/_temp/ccusage-large-codex-fixture (1.01 GiB, 2,597 files)
All rows run --offline --json, measured by hyperfine with 0 warmups and 1 runs. This isolates wrapper overhead from the installed native optional dependency and the workspace release binary built on the runner.

Committed fixture performance

Committed small fixtures for stable PR-to-PR feedback and explicit Claude/Codex command coverage.

Fixtures: Claude apps/ccusage/test/fixtures/claude (0.00 MiB, 2 files), Codex apps/ccusage/test/fixtures/codex (0.00 MiB, 1 files)
Base runs the published ccusage package from pkg.pr.new, installed before measurement; PR runs the published ccusage package from pkg.pr.new, installed before measurement. Both run --offline --json, measured by hyperfine with 2 warmups and 7 runs.

Large real-world-shaped fixture performance

Generated fixtures shaped from aggregate local log statistics: thousands of JSONL files, many small sessions, and a long tail of larger sessions. No real prompts, paths, or outputs are stored in the fixtures.

Fixtures: Claude /home/runner/work/_temp/ccusage-large-fixture (1.01 GiB, 2,597 files), Codex /home/runner/work/_temp/ccusage-large-codex-fixture (1.01 GiB, 2,597 files)
Base runs the published ccusage package from pkg.pr.new, installed before measurement; PR runs the published ccusage package from pkg.pr.new, installed before measurement. Both run --offline --json, measured by hyperfine with 0 warmups and 1 runs.

Artifact size

Lower medians and smaller artifacts are better. CI runner noise still applies; use same-run ratios as directional PR feedback, not release guarantees.

@coderabbitai merged latest origin/main again and resolved the new Hermes conflict by keeping the all-agent loader parallel and adding Hermes to the concurrent load set.

Validation after this merge:

• pnpm run format
• pnpm typecheck
• pnpm run test
• pre-push clippy

Please review the updated branch again.

#!/bin/bash
gh pr diff 1066 --repo ryoppippi/ccusage

@ryoppippi: The Hermes merge resolution is clean. Here's what I verified:

1. Hermes correctly wired in — load_hermes_rows(load_kind, shared, &pricing) at index 4 is consistent with the other pricing-aware loaders (amp, kilo, copilot, gemini). ✓

2. Agent ordering preserved — Indices 0–8 (claude → codex → opencode → amp → hermes → pi → kilo → copilot → gemini) are contiguous with no gaps; the post-join sort-by-index produces deterministic output order. ✓

3. Previously approved fixes intact — explicit match for load_kind, distinct Claude/Codex progress agents in the concurrency test, and deterministic error selection by index are all still in place. ✓

No issues introduced by this merge.

(´｡• ᵕ •｡`) ♡

[approve]

ccusage performance comparison

PR SHA: 9f7d1e71ea97
Base SHA: 34f9c68d7c61

This compares the Rust PR release binary against the configured base package on the same CI runner.

Package runner startup

Execution setup measures any pre-benchmark package materialization used by the execution benchmark. Bunx temp cache measures one bunx -p <url> ccusage --version run with an empty Bun install cache. Warm reuses that cache and reports the median of repeated runs.

Cached bunx execution performance

Runs the same large fixture through bunx -p <pkg.pr.new URL> ccusage after the Bun install cache has already been populated by the startup measurement. This separates cached package-runner execution from first-fetch package materialization.

Fixtures: Claude /home/runner/work/_temp/ccusage-large-fixture (1.01 GiB, 2,597 files), Codex /home/runner/work/_temp/ccusage-large-codex-fixture (1.01 GiB, 2,597 files)
Base package: 34f9c68d7c61; PR package: 9f7d1e71ea97. Both run through bunx -p <pkg.pr.new URL> ccusage using the warmed Bun install cache from package runner startup, measured by hyperfine with 0 warmups and 1 runs.

Package runtime diagnostics

Compares the PR package wrapper, the installed native optional dependency binary, and the workspace release binary on the same large fixture. This identifies whether slow package results come from JavaScript wrapper overhead, the published native binary build, or the Rust core itself.

Fixtures: Claude /home/runner/work/_temp/ccusage-large-fixture (1.01 GiB, 2,597 files), Codex /home/runner/work/_temp/ccusage-large-codex-fixture (1.01 GiB, 2,597 files)
All rows run --offline --json, measured by hyperfine with 0 warmups and 1 runs. This isolates wrapper overhead from the installed native optional dependency and the workspace release binary built on the runner.

Committed fixture performance

Committed small fixtures for stable PR-to-PR feedback and explicit Claude/Codex command coverage.

Fixtures: Claude apps/ccusage/test/fixtures/claude (0.00 MiB, 2 files), Codex apps/ccusage/test/fixtures/codex (0.00 MiB, 1 files)
Base runs the published ccusage package from pkg.pr.new, installed before measurement; PR runs rust/target/release/ccusage directly. Both run --offline --json, measured by hyperfine with 2 warmups and 7 runs.

Large real-world-shaped fixture performance

Generated fixtures shaped from aggregate local log statistics: thousands of JSONL files, many small sessions, and a long tail of larger sessions. No real prompts, paths, or outputs are stored in the fixtures.

Fixtures: Claude /home/runner/work/_temp/ccusage-large-fixture (1.01 GiB, 2,597 files), Codex /home/runner/work/_temp/ccusage-large-codex-fixture (1.01 GiB, 2,597 files)
Base runs the published ccusage package from pkg.pr.new, installed before measurement; PR runs rust/target/release/ccusage directly. Both run --offline --json, measured by hyperfine with 0 warmups and 1 runs.

Artifact size

Lower medians and smaller artifacts are better. CI runner noise still applies; use same-run ratios as directional PR feedback, not release guarantees.

ccusage performance comparison

PR SHA: 9f7d1e71ea97
Base SHA: 34f9c68d7c61

This compares the PR package against the configured base package on the same CI runner.

Package runner startup

Execution setup measures any pre-benchmark package materialization used by the execution benchmark. Bunx temp cache measures one bunx -p <url> ccusage --version run with an empty Bun install cache. Warm reuses that cache and reports the median of repeated runs.

Cached bunx execution performance

Runs the same large fixture through bunx -p <pkg.pr.new URL> ccusage after the Bun install cache has already been populated by the startup measurement. This separates cached package-runner execution from first-fetch package materialization.

Fixtures: Claude /home/runner/work/_temp/ccusage-large-fixture (1.01 GiB, 2,597 files), Codex /home/runner/work/_temp/ccusage-large-codex-fixture (1.01 GiB, 2,597 files)
Base package: 34f9c68d7c61; PR package: 9f7d1e71ea97. Both run through bunx -p <pkg.pr.new URL> ccusage using the warmed Bun install cache from package runner startup, measured by hyperfine with 0 warmups and 1 runs.

Package runtime diagnostics

Compares the PR package wrapper, the installed native optional dependency binary, and the workspace release binary on the same large fixture. This identifies whether slow package results come from JavaScript wrapper overhead, the published native binary build, or the Rust core itself.

Fixtures: Claude /home/runner/work/_temp/ccusage-large-fixture (1.01 GiB, 2,597 files), Codex /home/runner/work/_temp/ccusage-large-codex-fixture (1.01 GiB, 2,597 files)
All rows run --offline --json, measured by hyperfine with 0 warmups and 1 runs. This isolates wrapper overhead from the installed native optional dependency and the workspace release binary built on the runner.

Committed fixture performance

Committed small fixtures for stable PR-to-PR feedback and explicit Claude/Codex command coverage.

Fixtures: Claude apps/ccusage/test/fixtures/claude (0.00 MiB, 2 files), Codex apps/ccusage/test/fixtures/codex (0.00 MiB, 1 files)
Base runs the published ccusage package from pkg.pr.new, installed before measurement; PR runs the published ccusage package from pkg.pr.new, installed before measurement. Both run --offline --json, measured by hyperfine with 2 warmups and 7 runs.

Large real-world-shaped fixture performance

Generated fixtures shaped from aggregate local log statistics: thousands of JSONL files, many small sessions, and a long tail of larger sessions. No real prompts, paths, or outputs are stored in the fixtures.

Fixtures: Claude /home/runner/work/_temp/ccusage-large-fixture (1.01 GiB, 2,597 files), Codex /home/runner/work/_temp/ccusage-large-codex-fixture (1.01 GiB, 2,597 files)
Base runs the published ccusage package from pkg.pr.new, installed before measurement; PR runs the published ccusage package from pkg.pr.new, installed before measurement. Both run --offline --json, measured by hyperfine with 0 warmups and 1 runs.

Artifact size

Lower medians and smaller artifacts are better. CI runner noise still applies; use same-run ratios as directional PR feedback, not release guarantees.