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
{{ message }}
This repository has been archived by the owner on Nov 15, 2023. It is now read-only.
While testing #14619, I noticed that full sync can sometimes get stuck at 0.0 bps which happens because ChainSync has MAX_DOWNLOAD_AHEAD blocks queued and it's waiting for one peer to respond to a request so it can start sending more block requests/flush its queue. For some reason the timeout mechanism is not working as it should and a request can stay unanswered for more than the 20 second limit it's currently configured at, causing syncing to stall.
FWIW, I've seen this happen on master as well. Timeout seems to be working just fine in tests.
The text was updated successfully, but these errors were encountered:
altonen
added
I3-bug
The node fails to follow expected behavior.
U1-asap
No need to stop dead in your tracks, however issue should be addressed as soon as possible.
labels
Jul 30, 2023
These nodes are parachain collators, meaning they support the /block-announces/1 substream but not the block request protocol while sc-network-sync expects any node to support the request protocol that supports the block announcement protocol.
We may have to introduce a new node subclass to SyncingEngine for collators. But the core issue right now is why is the multistream-select not concluding as failed for the request protocol.
While testing #14619, I noticed that full sync can sometimes get stuck at 0.0 bps which happens because
ChainSync
hasMAX_DOWNLOAD_AHEAD
blocks queued and it's waiting for one peer to respond to a request so it can start sending more block requests/flush its queue. For some reason the timeout mechanism is not working as it should and a request can stay unanswered for more than the 20 second limit it's currently configured at, causing syncing to stall.FWIW, I've seen this happen on master as well. Timeout seems to be working just fine in tests.
The text was updated successfully, but these errors were encountered: