Failing to unregister sync peer (which not registered) #22077
Comments
|
This is funky! I restarted a totally different machine, |
|
A third machine is flipping between "is it registered or not?" |
|
The type peerSet struct {
peers map[string]*peerConnection
newPeerFeed event.Feed
peerDropFeed event.Feed
lock sync.RWMutex
}Whereas teh type peerSet struct {
ethPeers map[string]*ethPeer // Peers connected on the `eth` protocol
snapPeers map[string]*snapPeer // Peers connected on the `snap` protocolThe method in the
func (h *handler) removePeer(id string) {
// Remove the eth peer if it exists
eth := h.peers.ethPeer(id)
if eth != nil {
log.Debug("Removing Ethereum peer", "peer", id)
h.downloader.UnregisterPeer(id)
h.txFetcher.Drop(id)
if err := h.peers.unregisterEthPeer(id); err != nil {
log.Error("Peer removal failed", "peer", id, "err", err)
}
}
// Remove the snap peer if it exists
snap := h.peers.snapPeer(id)
if snap != nil {
log.Debug("Removing Snapshot peer", "peer", id)
h.downloader.SnapSyncer.Unregister(id)
if err := h.peers.unregisterSnapPeer(id); err != nil {
log.Error("Peer removal failed", "peer", id, "err", err)
}
}Something isn't quite right with the accounting of peers here, and the special |
|
Interesting.. From #22151 : What's with that |
|
related #21257 |
|
This issue happens because If two goroutines call We can fix this by refactoring the peer unregistration a bit. At the moment, we pass the |
|
@fjl want to handle this? |
|
I'm seeing Geth 1.10.0 throw a few similar errors (after upgrading from 1.9.25) ^For the same peer, a few seconds apart. For a different peer: The only slightly naughty thing I did was run the new |
|
I was synchronizing the ropsten node, and when there were dozens of blocks, I checked the synchronization log and found that it was easy to insert the error in the figure below every few logs. Can someone tell me how to deal with the change? |
|
WARN [06-03|02:38:16.714] Ancestor below allowance peer=465c4021 number=10,263,739 hash=000000..000000 allowance=10,263,739 |
|
The above error occurs when Generating state snapshot |
version
instance=Geth/v1.9.26-unstable-b9012a03-20201223/linux-amd64/go1.15.3It's running latest develop,
--syncmode=snap --snapshot. I just restarted it after having done a pruning. It keeps saying complaining about failing to unregister a peer that it for some reason thinks it's connected to, while it apparently is not.The text was updated successfully, but these errors were encountered: