You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Bug: attempting to fetch UnixFS files over bitswap with peers that only expose websockets will fail on large blocks. Note: this is all of Elastic IPFS/Web3.storage
Steps To Reproduce:
Build Kubo v0.18.0:
Start the daemon:
./cmd/ipfs/ipfs daemon
./cmd/ipfs/ipfs get bafybeiarntxetfbzpyp7baj7kqvqljrn5uyoobq2y4b3zegbbem5joazgy
This is a CID I stored on web3.storage. It works.
Now switch to Kubo master and do the same. It now hangs.
There also seems to be this issue that a user ran into yesterday with the limit: nhooyr/websocket#382 which shows an option to disable the read limit (but no ETA for a release)
Thanks @MarcoPolo for getting the PR out. What timeline are you thinking here for merging and getting it released? I'm asking since it's important for the Kubo 0.19 release (prevent regression) and impacting Rhea/Lassie work. Assuming @marten-seemann is out, who else would you like to pull in?
Bug: attempting to fetch UnixFS files over bitswap with peers that only expose websockets will fail on large blocks. Note: this is all of Elastic IPFS/Web3.storage
Steps To Reproduce:
Build Kubo v0.18.0:
Start the daemon:
This is a CID I stored on web3.storage. It works.
Now switch to Kubo master and do the same. It now hangs.
Heavy debugging identifies the following error generated in the websocket conn:
failed to read: read limited at 32769 bytes
-- this comes from here I believe: https://github.com/nhooyr/websocket/blob/master/read.go#L385Note that actually everything works fine until we get to the first 1MB raw block, I assume because it's over that limit.
This was originally discovered in Lassie and in debugging why it couldn't fetch while Kubo 0.18.0 could, I found that Kubo master couldn't either.
The text was updated successfully, but these errors were encountered: