Error: ENOENT: no such file or directory

[haadcode]

As of 0.14.1 I'm seeing this fairly constantly:

Error: ENOENT: no such file or directory, open '/tmp/orbit-tests/blocks/1220a892/1220a89239577c1a6763066dd2ff7ed57078efcb951ad5f94fc7262ea673d3e938da.data'

This happens when I do ipfs.object.put and then try to get it immediately after. Adding a timeout before .get seems to get rid of the problem, so my suspicion is that we're returning from put before the data has been written to "disk" (disk being wherever we write the data blocks).

[haadcode]

To emphasise the urgency: this breaks pretty much all tests for Orbit and down the line in orbit-db.

[dignifiedquire]

And another bug due to node streams 😢

[daviddias]

@dignifiedquire there are some layers of indirection, because of the streams to {indexedDB, fs}, but IIRC, we fixed that in IndexedDB, but I guess that fs-blob-store never got the love required since we never noticed a bug there (which probably was just because fs was way faster).

Do you see this problem in orbit in the browser or in a Node.js process?

We have locks for writing https://github.com/ipfs/js-ipfs-repo/blob/master/src/stores/datastore.js#L34 so that we don't step into concurrent processes toes, we might want to add them to reads (wait for the write to finish). Also, seems like it is the time to add LRU caching to blocks, it will improve perf and would mitigate this issue.

@haadcode do you feel like tackling these?

[dignifiedquire]

I am not sure the fix will actually work on fs-blob-store, all my tries have failed so far.

[daviddias]

This should be fixed once #403 is done

[dignifiedquire]

#403 was merged and released, so this should be fixed now. Please reopen if this still happens.

[daviddias]

@haadcode please, verify :)

[SignpostMarv]

seeing this under { version: '0.41.2', repo: 7, commit: '' } on node 13.9.0 on win10.