perf(coverage): reduce RPC data and unnecessary serializations - #10781

Description

Adds two performance improvements I've had in mind long time:

1. Currently we are sending coverage JSON data from Node side workers to main thread, and then writing that to file system. Now we'll write the data directly in the worker and just pass the path to main thread. Initially I thought writing in workers would be worse, but it should actually be faster. RPC always requires serialization that's costly.

2. When reviewing fix(browser): disable client cdp API when allowWrite/allowExec: false #10444 I realized browser mode v8 coverage is doing unnecessary RPC hops:

• Browser command to call CDP on server side to get V8 script coverage, pass that to browser via WS RPC
• In browser we just filter out stuff from the JSON
• Send the filtered JSON again to main thread to the coverage provider.

Now we'll instead just write the results to file system in the first call, and never pass the V8 script coverage to browser.

LLM Generated analysis of perf changes

Prompted Claude to run vitest-coverage-large locally with main and this PR. There's noticeable perf change in browser + v8, as now we don't bounce the results unnecessary over the WS RPC. 🔥

Performance

Details

Measured with [`coverage-large`](https://github.com/AriPerkkio/coverage-large) using `pkg.pr.new` builds of this PR and its parent commit. The fixture was generated in a payload-heavy shape where each test file imports a 300-module graph, so that per-test coverage payloads are large (~6 MB for `v8`, ~1.9 GB total across the run):

COVERED_FILE_COUNT=300 UNCOVERED_FILE_COUNT=0 FUNCTION_COUNT=100 IMPORTS_PER_TEST=300 node ./generate-files.mjs

Durations are medians of 3 runs (2 runs for istanbul). Memory is the main process' peak RSS.

Duration

Case	Before (0f8e63f)	After (b7e2ad6)	Speed-up
node + v8	39.0 s	33.9 s	~1.15×
node + istanbul	527 s	365 s	~1.4×
browser + v8	103 s	38.2 s	~2.7×
node + v8, default fixture ¹	28.3 s	26.8 s	~1.06×

Main process peak memory

Case	Before (0f8e63f)	After (b7e2ad6)	Change
node + v8	1 322 MB	1 245 MB	−6 %
node + istanbul	4 111 MB	3 121 MB	−24 %
browser + v8	1 370 MB	1 297 MB	−5 %

Details

¹ No-regression control: the stock coverage-large shape (600 + 600 files, one covered file per test) has tiny per-test payloads, so little change is expected — or observed.

browser + istanbul is excluded from the payload-heavy comparison: at this payload size it does not complete reliably on either side of this PR (before: 190–201 of 300 test files in 26–28 min; after: 206 of 300). The istanbul coverage map only exists inside the browser page, so it must cross the WebSocket in both versions — this PR removes the return trip and main-process re-serialization, but not the dominant page-side transfer. On the stock fixture it completes fine on both sides (139–181 s before, ~136 s after).

LLM Generated sequence diagrams before-after

Output of following prompt below.

Now, draw a sequence diagram that shows how data is moved across main thread, workers and browser. Show how it was done previously and how it's now.

sequenceDiagram
    participant Worker as Test worker
    participant Main as Main process
    participant FS as File system

    Worker->>Worker: takeCoverage() - collect and filter results
    Worker->>Main: onAfterSuiteRun(results)<br/>structured clone (threads) / v8.serialize (forks)<br/>full results, MBs on large projects
    Main->>Main: JSON.stringify(results)
    Main->>FS: write .tmp/coverage-N.json
    Note over Main,FS: at report time
    FS->>Main: read + JSON.parse each file
    Main->>Main: merge, remap, report

sequenceDiagram
    participant Worker as Test worker
    participant Main as Main process
    participant FS as File system

    Worker->>Worker: takeCoverage() - collect and filter results
    Worker->>FS: JSON.stringify(results)<br/>write .tmp/coverage-uuid.json
    Worker->>Main: onAfterSuiteRun(filename)<br/>plain string, few bytes
    Note over Main,FS: at report time
    FS->>Main: read + JSON.parse each file
    Main->>Main: merge, remap, report

sequenceDiagram
    participant Page as Tester page
    participant Main as Main process
    participant FS as File system

    rect rgba(128, 128, 128, 0.1)
    Note right of Page: v8
    Page->>Main: __vitest_takeV8Coverage command
    Main->>Main: Profiler.takePreciseCoverage via CDP
    Main->>Page: unfiltered results over WebSocket<br/>MBs, includes node_modules and vite client
    Page->>Page: filter + rewrite URLs
    Page->>Main: onAfterSuiteRun(results) over WebSocket - MBs
    end

    rect rgba(128, 128, 128, 0.1)
    Note right of Page: istanbul
    Page->>Main: onAfterSuiteRun(coverage map) over WebSocket - full map
    end

    Main->>Main: JSON.stringify(results)
    Main->>FS: write .tmp/coverage-N.json

sequenceDiagram
    participant Page as Tester page
    participant Main as Main process
    participant FS as File system

    rect rgba(128, 128, 128, 0.1)
    Note right of Page: v8
    Page->>Main: __vitest_takeV8Coverage(location.href) - few bytes
    Main->>Main: Profiler.takePreciseCoverage via CDP<br/>filter + rewrite URLs
    Main->>FS: write .tmp/coverage-uuid.json
    Main->>Page: filename
    end

    rect rgba(128, 128, 128, 0.1)
    Note right of Page: istanbul
    Page->>Main: __vitest_writeCoverageFile(coverage map)<br/>one WebSocket transfer
    Main->>FS: write .tmp/coverage-uuid.json
    Main->>Page: filename
    end

    Page->>Main: onAfterSuiteRun(filename) - few bytes

Please don't delete this checklist! Before submitting the PR, please make sure you do the following:

• It's really useful if your PR references an issue where it is discussed ahead of time. If the feature is substantial or introduces breaking changes without a discussion, PR might be closed.
• Ideally, include a test that fails without this PR but passes with it.
• Please, don't make changes to pnpm-lock.yaml unless you introduce a new test example.
• Please check Allow edits by maintainers to make review process faster. Note that this option is not available for repositories that are owned by Github organizations.

Tests

• Run the tests with pnpm test:ci.

Documentation

• If you introduce new functionality, document it. You can run documentation with pnpm run docs command.

Changesets

• Changes in changelog are generated from PR name. Please, make sure that it explains your changes in an understandable manner. Please, prefix changeset messages with feat:, fix:, perf:, docs:, or chore:.