kshinn
commented
5 years ago The ethereumjs code seems to have a problem parsing the body and assumes the protocol is out of spec or an unknown variant.
Open kshinn opened 5 years ago
kshinn
commented
5 years ago The ethereumjs code seems to have a problem parsing the body and assumes the protocol is out of spec or an unknown variant.
egalano
commented
5 years ago It'd be interesting to see as much information as possible from the client connection that is reporting the subprotocol error. Strange custom clients, older versions maybe? The DISCONNECT_REQUESTED could also be coming from a peer that does not have available peering slots
kshinn
commented
5 years ago From what I can tell the disconnect request are part of the normal protocol. I've been seeing them after a successful status call. There is an explicit peering table full error I see as well. There may be value in tracking these in the db as well.
On Fri, Dec 28, 2018, 11:21 PM E.G. Galano <notifications@github.com wrote:
It'd be interesting to see as much information as possible from the client connection that is reporting the subprotocol error. Strange custom clients, older versions maybe? The DISCONNECT_REQUESTED could also be coming from a peer that does not have available peering slots
— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub https://github.com/INFURA/devp2p-network/issues/6#issuecomment-450472185, or mute the thread https://github.com/notifications/unsubscribe-auth/AACEUXoU_QiNvsAbGX29xhicb9xKSgIEks5u9xgMgaJpZM4ZkcVu .
kshinn
commented
5 years ago Got a little more information. The reason we are getting a protocol mismatch looks like a network ID mismatch between the Peer status message and the devp2p status message. It ends up throwing an assertion message.
Devp2p client message:
infura:main 52.3.158.184: eth.62,eth.63 +216ms
infura:p2p [ <Buffer 3f>,
infura:p2p <Buffer 01>, <-- Mainnet
infura:p2p <Buffer 04 00 00 00 00>,
infura:p2p <Buffer d4 e5 67 40 f8 76 ae f8 c0 10 b8 6a 40 d5 f5 67 45 a1 18 d0 90 6a 34 e6 9a ec 8c 0d b1 cb 8f a3>,
infura:p2p <Buffer d4 e5 67 40 f8 76 ae f8 c0 10 b8 6a 40 d5 f5 67 45 a1 18 d0 90 6a 34 e6 9a ec 8c 0d b1 cb 8f a3> ] +195msLooking at the wire protocol the first byte is the protocol version (63)
The second is the NetworkID. The 01 above is Mainnet, what the devp2p client is expecting.
Peer status
infura:p2p [ <Buffer 3f>,
infura:p2p <Buffer 04>, <-- Rinkby
infura:p2p <Buffer 65 51 35>,
infura:p2p <Buffer de ed 9b 12 f5 b3 bd bb 51 a9 e6 65 04 00 37 6d 11 0a 90 76 09 96 5e cc d6 b8 61 f0 b7 ea 53 03>,
infura:p2p <Buffer 63 41 fd 3d af 94 b7 48 c7 2c ed 5a 5b 26 02 8f 24 74 f5 f0 0d 82 45 04 e4 fa 37 a7 57 67 e1 77> ] +2ms
infura:main 52.3.158.184:30303 (Peer Error) AssertionError [ERR_ASSERTION]: NetworkId mismatch: 01 / 04 +1ms
infura:main 52.3.158.184 SUBPROTOCOL_ERROR +2msPosted for reference
| No. | hex | Network | Notes |
|---|---|---|---|
| 0 | 0x00 | Olympic | Ethereum public pre-release testnet |
| 1 | 0x01 | Frontier | Homestead, Metropolis, the Ethereum public main network |
| 1 | 0x01 | Classic | the (un)forked public Ethereum Classic main network, chain ID 61 |
| 1 | 0x01 | Expanse | an alternative Ethereum implementation, chain ID 2 |
| 2 | 0x02 | Morden | the public Ethereum testnet, now Ethereum Classic testnet |
| 3 | 0x03 | Ropsten | the public cross-client Ethereum testnet |
| 4 | 0x04 | Rinkeby | the public Geth PoA testnet |
| 8 | 0x08 | Ubiq | the public Gubiq main network with flux difficulty chain ID 8 |
| 42 | 0x2A | Kovan | the public Parity PoA testnet |
| 77 | 0x4D | Sokol | the public POA Network testnet |
| 99 | 0x63 | Core | the public POA Network main network |
| 100 | 0x64 | xDai | the public MakerDAO/POA Network main network |
| 401697 | 0x62121 | Tobalaba | the public Energy Web Foundation testnet |
| 7762959 | 0x7F740F | Musicoin | the music blockchain |
| 61717561 | 0x3ADBC39 | Aquachain | ASIC resistant chain |
I'm seeing a lot of peers getting removed because of a subprotocol error.
I've checked the handshake and it looks like the library is doing that properly on the peer connect. I suspect we are seeing a number of missing nodes because of these protocol mismatches.