Chrome 文件系统 API 挂起

Question

disclaimer, self-answered post to hopefully save others time.

Setup :

I've been using chrome's implementation of the file systems API, [1] [2] [3] .

This requires enabling the flag chrome://flags/#native-file-system-api .

For starters I want to recursively read a directory and obtain a list of files. This is simple enough:

paths = [];
let recursiveRead = async (path, handle) => {
    let reads = [];
    // window.handle = handle;
    for await (let entry of await handle.getEntries()) { // <<< HANGING
        if (entry.isFile)
            paths.push(path.concat(entry.name));
        else if (/* check some whitelist criteria to restrict which dirs are read*/)
            reads.push(recursiveRead(path.concat(entry.name), entry));
    }
    await Promise.all(reads);
    console.log('done', path, paths.length);
};

chooseFileSystemEntries({type: 'openDirectory'}).then(handle => {
    recursiveRead([], handle).then(() => {
        console.log('COMPLETELY DONE', paths.length);
    });
});

I've also implemented a non-recursive while-loop-queue version. And lastly, I've implemented a node fs.readdir version. All 3 solutions work fine for small directories.

The problem:

But then I tried running it on some sub-directories of the chromium source code ('base', 'components', and 'chrome'); together the 3 sub-dirs consist of ~63,000 files. While the node implementation worked fine (and surprisingly it utilized cached results between runs, resulting in instantaneous runs after the first), both browser implementations hung.

Attempted debugging:

Sometimes, they would return the full 63k files and print 'COMPLETLEY DONE' as expected. But most often (90% of the time) they would read 10k-40k files before hanging.

I dug deeper into the hanging, and apparently the for await line was hanging. So I added the line window.handle = handle immediately before the for loop; when the function hung, I ran the for loop directly in the browser console, and it worked correctly. So now I'm stuck. I have seemingly working code that randomly hangs.

Answer 1

Solution:

I tried skipping over directories that would hang:

let whitelistDirs = {src: ['base', 'chrome', 'components', /*'ui'*/]}; // 63800

let readDirEntry = (handle, timeout = 500) => {
    return new Promise(async (resolve, reject) => {
        setTimeout(() => reject('timeout'), timeout);
        let entries = [];
        for await (const entry of await handle.getEntries())
            entries.push(entry);
        resolve(entries);
    });
};

let readWhile = async entryHandle => {
    let paths = [];
    let pending = [{path: [], handle: entryHandle}];
    while (pending.length) {
        let {path, handle} = pending.pop();
        await readDirEntry(handle)
            .then(entries =>
                entries.forEach(entry => {
                    if (entry.isFile)
                        paths.push({path: path.concat(entry.name), handle: entry});
                    else if (path.length || !whitelistDirs[handle.name] || whitelistDirs[handle.name].includes(entry.name))
                        pending.push({path: path.concat(entry.name), handle: entry});
                }))
            .catch(() => console.log('skipped', handle.name));
        console.log('paths read:', paths.length, 'pending remaining:', pending.length, path);
    }
    console.log('read complete, paths.length');
    return paths;
};

chooseFileSystemEntries({type: 'openDirectory'}).then(handle => {
    readWhile(handle).then(() => {
        console.log('COMPLETELY DONE', paths.length);
    });
});

And the results showed a pattern. Once a directory read hung and was skipped, the subsequent ~10 dir reads would likewise hang and be skipped. Then the following reads would resume functioning properly until the next similar incident.

// begins skipping
paths read: 45232 pending remaining: 49 (3) ["chrome", "browser", "favicon"]
VM60:25 skipped extensions
VM60:26 paths read: 45239 pending remaining: 47 (3) ["chrome", "browser", "extensions"]
VM60:25 skipped enterprise_reporting
VM60:26 paths read: 45239 pending remaining: 46 (3) ["chrome", "browser", "enterprise_reporting"]
VM60:25 skipped engagement
VM60:26 paths read: 45266 pending remaining: 45 (3) ["chrome", "browser", "engagement"]
VM60:25 skipped drive
VM60:26 paths read: 45271 pending remaining: 44 (3) ["chrome", "browser", "drive"]
// begins working properly again

So the issue seemed temporal. I added a simple retry wrapper with a 500ms wait between retries, and the reads began working fine.

readDirEntryRetry = async (handle, timeout = 500, tries = 5, waitBetweenTries = 500) => {
    while (tries--) {
        try {
            return await readWhile(handle, timeout);
        } catch (e) {
            console.log('readDirEntry failed, tries remaining:', tries, handle.name);
            await sleep(waitBetweenTries);
            if (!tries)
                return e;
        }
    }
};

Conclusion:

The non-standard Native File System API hangs when reading large directories. Simply retrying after waiting resolves the issue. Took me a good week to arrive at this solution, so thought it'd be worth sharing.