Uncaught Error: no writecb in Transform class

[haadcode]

Came across the following error while running tests for orbit-db.

It happens on .files.add, using the following:

ipfs.files.add(new Buffer(JSON.stringify(defaultNetwork), 'utf-8'), (err, res) => ...)

It happens always. However, if I don't call ipfs.goOnline() at init, the error doesn't occur.

ipfs v0.13.0

Uncaught Error: no writecb in Transform class
      at afterTransform (node_modules/through2/node_modules/readable-stream/lib/_stream_transform.js:74:40)
      at TransformState.afterTransform (node_modules/through2/node_modules/readable-stream/lib/_stream_transform.js:58:12)
      at node_modules/ipfs/lib/core/ipfs/files.js:90:11
      at node_modules/ipfs-merkle-dag/lib/dag-service.js:68:16
      at done (node_modules/ipfs-bitswap/lib/index.js:198:9)
      at node_modules/ipfs-bitswap/lib/index.js:205:9
      at finish (node_modules/ipfs-bitswap/lib/index.js:263:11)
      at node_modules/ipfs-bitswap/lib/index.js:290:17
      at BufferList._callback (node_modules/ipfs-repo/lib/stores/datastore.js:60:9)
      at BufferList.end (node_modules/bl/bl.js:98:10)
      at ReadStream.onend (_stream_readable.js:498:10)
      at endReadableNT (_stream_readable.js:913:12)

[haadcode]

Also, this is pretty much a blocker as it happens every time and prevents js-ipfs to run smoothly.

[dignifiedquire]

interesting, not seen before

[haadcode]

Repro case: https://github.com/haadcode/orbit-db/tree/latest-js-ipfs (run npm i && npm test)

If you leave out the goOnline here https://github.com/haadcode/orbit-db/blob/latest-js-ipfs/test/counterdb.test.js#L46, the tests pass.

[hackergrrl]

I did some digging into this.

The symptom (writecb) happens when the callback in a Transform stream gets called multiple times. In ipfs.files.add in js-ipfs we add each file to the repo, and then ask the DAG Service to find each file that we just added. This takes us down into DAGService, then BlockService, and finally Bitswap.

It looks like there's a race condition in js-ipfs-bitswap in the case that the underlying block datastore takes a long time to do a put and there are two getBlocks pending for that same block. getBlock will call finish if its attempt to get the block succeeds or it gets a notification that the block has become available. However, if both happen, then the callback gets called twice.

I filed a PR (ipfs/js-ipfs-bitswap#20) and confirmed that it fixes your error. 🎉

[daviddias]

@haadcode can you confirm the error no longer exists? Closing, reopen if necessary