Peering - disconnect peers that have multiple discovery ports #7089
Conversation
Signed-off-by: stefan.pingel@consensys.net <stefan.pingel@consensys.net>
Signed-off-by: stefan.pingel@consensys.net <stefan.pingel@consensys.net>
Signed-off-by: stefan.pingel@consensys.net <stefan.pingel@consensys.net>
Signed-off-by: stefan.pingel@consensys.net <stefan.pingel@consensys.net>
|
very cool find @pinges - are there more tests you wanted to add? |
|
11 holesky nodes - 10 of them get to 100% peers within 2h the 11th one eventually gets to 100% peers (although it does take ~7h) Current behaviour on holesky is > 90% of nodes get stuck at 0 peers indefinitely (require setting static peers to sync) See #6805 |
|
16 mainnet nodes - all of them get to 100% peers within 1h. Current behaviour on mainnet is a small percentage - maybe 5% - of nodes get stuck on 0 peers and require a restart. |
macfarla
changed the title
| private final LHMWithMaxSize<String, Integer> ipAddressCheckMap = | ||
| new LHMWithMaxSize<>(DEFAULT_BUCKET_SIZE * N_BUCKETS); | ||
| private final CircularFifoQueue<String> invalidIPs = | ||
| new CircularFifoQueue<>(DEFAULT_BUCKET_SIZE * N_BUCKETS); |
There was a problem hiding this comment.
curious - why these specific collection types?
There was a problem hiding this comment.
CircularFifo is a simple queue with a maximum size.
LHMWithMaxSize is a private class, it is a LinkedHashMap that I have extended to have a maximum size.
Lukas complained about the name of the class and I will rename it :-)
Signed-off-by: Sally MacFarlane <macfarla.github@gmail.com>
| @@ -51,6 +54,10 @@ public class PeerTable { | |||
| private final Map<Bytes, Integer> distanceCache; | |||
| private BloomFilter<Bytes> idBloom; | |||
| private int evictionCnt = 0; | |||
| private final LHMWithMaxSize<String, Integer> ipAddressCheckMap = | |||
There was a problem hiding this comment.
I think we should rename LHMWithMaxSize as LinkedHashMapWithMaxSize to be clearer.
Signed-off-by: stefan.pingel@consensys.net <stefan.pingel@consensys.net>
PR description
In discovery we seemed to see a lot of "misconfigured" nodes that share the same IP address and TCP port, but use different discovery ports. This PR checks for these nodes and removes them from the peer table.
Fixes #6805