-
Notifications
You must be signed in to change notification settings - Fork 15
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
test: add separate-thread concurrency test #305
Conversation
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
self review
@@ -48,7 +48,6 @@ async function writeFile (file: string, contents: Uint8Array): Promise<void> { | |||
// attempts to write the same block by two different function calls | |||
return | |||
} |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
should add the newline back here.
@@ -37,7 +37,7 @@ async function writeFile (file: string, contents: Uint8Array): Promise<void> { | |||
try { | |||
await writeAtomic(file, contents) | |||
} catch (err: any) { | |||
if (err.code === 'EPERM' && err.syscall === 'rename') { | |||
if (err.syscall === 'rename' && ['ENOENT', 'EPERM'].includes(err.code)) { |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
FYI that on my mac, an ENOENT
Error was firing on the rename
call.
example:
blockstore-fs: Error putting block [Error: ENOENT: no such file or directory, rename '/var/folders/bl/_gl5_59s11v7qz5ysd6bfgb00000gn/T/test-0.8216405349064326/LY/.63718.93' -> '/var/folders/bl/_gl5_59s11v7qz5ysd6bfgb00000gn/T/test-0.8216405349064326/LY/BCIQPGZJ6QLZOFG3OP45NLMSJUWGJCO72QQKHLDTB6FXIB6BDSLRQYLY.data'] {
blockstore-fs: errno: -2,
blockstore-fs: code: 'ERR_PUT_FAILED',
blockstore-fs: syscall: 'rename',
blockstore-fs: path: '/var/folders/bl/_gl5_59s11v7qz5ysd6bfgb00000gn/T/test-0.8216405349064326/LY/.63718.93',
blockstore-fs: dest: '/var/folders/bl/_gl5_59s11v7qz5ysd6bfgb00000gn/T/test-0.8216405349064326/LY/BCIQPGZJ6QLZOFG3OP45NLMSJUWGJCO72QQKHLDTB6FXIB6BDSLRQYLY.data'
blockstore-fs: }
I think if we validate that the final file exists and is F_OK, W_OK as we were doing for EPERM
, this is safe.
I also thought about checking that error.dest
matches, but it seems like that is effectively what we're doing with fs.access
, so this should work
return true | ||
} catch (err) { | ||
// eslint-disable-next-line no-console | ||
console.error('Error opening blockstore', err) |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
so we can get some error information in testing environment without dealing with logger? i can add in a logger if desired.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
rethrow the error here. done.
@@ -37,7 +37,7 @@ async function writeFile (path: string, contents: Uint8Array): Promise<void> { | |||
try { | |||
await writeAtomic(path, contents) | |||
} catch (err: any) { | |||
if (err.code === 'EPERM' && err.syscall === 'rename') { | |||
if (err.syscall === 'rename' && ['ENOENT', 'EPERM'].includes(err.code)) { |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
see comment from same changes in blockstore-fs: #305 (comment)
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
looks like we're also getting an EPERM on 'open' with fast-atomic-write on windows:
Error: EPERM: operation not permitted, open 'C:\Users\RUNNER~1\AppData\Local\Temp\test-0.9268640174546598\LY.6280.3'
https://github.com/ipfs/js-stores/actions/runs/8757918340/job/24047367236?pr=305#step:5:687
This is not failing on windows with steno: #285
Note that https://github.com/ipfs/js-stores/actions/runs/8757918340/job/24037687392?pr=305#step:5:645 is failing here for windows, but passing in #285. going to retry the job. new job at https://github.com/ipfs/js-stores/actions/runs/8757918340?pr=305 |
findings in #305 (comment).. seem's like fast-write-atomic steps on itself with temp files when ran in a separate concurrent thread. |
fixes #284 fast-write-atomic hasn't been updated in 5 years, is CJS, and is slower than steno (updated 2 months ago). Benchmarks for various content-types & libraries (though we only use Uint8Arrays) can be found at https://github.com/SgtPooki/fast-write-atomic#benchmarks However, there may be further room for improvement by moving to [fs.createWriteStream](https://nodejs.org/api/fs.html#fscreatewritestreampath-options) ``` ╰─ ✔ ❯ hyperfine --parameter-list branch 284-chore-replace-fast-write-atomic-with-steno,main --setup "git switch {branch} && npm run reset && npm i && npm run build" --runs 20 -w 1 "npm run test:node" Benchmark 1: npm run test:node (branch = 284-chore-replace-fast-write-atomic-with-steno) Time (mean ± σ): 27.212 s ± 0.832 s [User: 34.810 s, System: 6.051 s] Range (min … max): 25.927 s … 29.324 s 20 runs Benchmark 2: npm run test:node (branch = main) Time (mean ± σ): 42.971 s ± 0.637 s [User: 35.297 s, System: 7.534 s] Range (min … max): 42.178 s … 44.796 s 20 runs Summary npm run test:node (branch = 284-chore-replace-fast-write-atomic-with-steno) ran 1.58 ± 0.05 times faster than npm run test:node (branch = main) ``` --- ### Updated benchmarks of `npm run test` as of 2024-04-19 ``` ╭─ ~/code/work/protocol.ai/ipfs/js-stores main ?1 ╰─ ✔ ❯ hyperfine --parameter-list branch main,test/not-same-event-loop-concurrency,284-chore-replace-fast-write-atomic-with-steno --setup "git switch {branch} && npm run reset && npm i && npm run build && cd packages/datastore-fs" "npm run test" Benchmark 1: npm run test (branch = main) Time (mean ± σ): 99.415 s ± 2.918 s [User: 69.659 s, System: 23.361 s] Range (min … max): 96.134 s … 105.200 s 10 runs Benchmark 2: npm run test (branch = test/not-same-event-loop-concurrency) Time (mean ± σ): 103.456 s ± 3.186 s [User: 74.442 s, System: 25.261 s] Range (min … max): 98.813 s … 108.429 s 10 runs Benchmark 3: npm run test (branch = 284-chore-replace-fast-write-atomic-with-steno) Time (mean ± σ): 80.308 s ± 2.107 s [User: 74.331 s, System: 22.228 s] Range (min … max): 78.219 s … 84.277 s 10 runs Summary npm run test (branch = 284-chore-replace-fast-write-atomic-with-steno) ran 1.24 ± 0.05 times faster than npm run test (branch = main) 1.29 ± 0.05 times faster than npm run test (branch = test/not-same-event-loop-concurrency) [49m1.944s] ╭─ ~/code/work/protocol.ai/ipfs/js-stores 284-chore-re…c-with-steno ?1 ╰─ ✔ ❯ hyperfine --parameter-list branch main,test/not-same-event-loop-concurrency,284-chore-replace-fast-write-atomic-with-steno --setup "git switch {branch} && npm run reset && npm i && npm run build && cd packages/blockstore-fs" "npm run test" Benchmark 1: npm run test (branch = main) Time (mean ± σ): 98.840 s ± 2.612 s [User: 68.486 s, System: 22.585 s] Range (min … max): 97.005 s … 104.396 s 10 runs Benchmark 2: npm run test (branch = test/not-same-event-loop-concurrency) Time (mean ± σ): 105.307 s ± 2.335 s [User: 72.442 s, System: 24.766 s] Range (min … max): 101.167 s … 109.007 s 10 runs Benchmark 3: npm run test (branch = 284-chore-replace-fast-write-atomic-with-steno) Time (mean ± σ): 77.012 s ± 1.829 s [User: 74.442 s, System: 21.938 s] Range (min … max): 75.258 s … 80.825 s 10 runs Summary npm run test (branch = 284-chore-replace-fast-write-atomic-with-steno) ran 1.28 ± 0.05 times faster than npm run test (branch = main) 1.37 ± 0.04 times faster than npm run test (branch = test/not-same-event-loop-concurrency) ```
@achingbrain this PR originally had only tests but since #285 was merged, the changes are larger now. Are we good with merging this one? |
ping @achingbrain |
🎉 This PR is included in version blockstore-fs-v2.0.2 🎉 The release is available on: Your semantic-release bot 📦🚀 |
🎉 This PR is included in version datastore-fs-v10.0.2 🎉 The release is available on: Your semantic-release bot 📦🚀 |
Basically spawned from #285 (comment)
This PR adds a test to both datastore-fs and blockstore-fs that validates puts are somewhat concurrent safe across multiple threads.