From db32a6589720e5b15931cef82e477118dfd92669 Mon Sep 17 00:00:00 2001 From: Suhas Daftuar Date: Mon, 23 Oct 2017 13:59:07 -0400 Subject: Track tip update time and last new block announcement from each peer --- src/net_processing.cpp | 37 +++++++++++++++++++++++++++++++++++-- 1 file changed, 35 insertions(+), 2 deletions(-) (limited to 'src/net_processing.cpp') diff --git a/src/net_processing.cpp b/src/net_processing.cpp index 4c3aacaf5..59df5d650 100644 --- a/src/net_processing.cpp +++ b/src/net_processing.cpp @@ -127,6 +127,10 @@ namespace { /** Number of outbound peers with m_chain_sync.m_protect. */ int g_outbound_peers_with_protect_from_disconnect = 0; + + /** When our tip was last updated. */ + int64_t g_last_tip_update = 0; + /** Relay map, protected by cs_main. */ typedef std::map MapRelay; MapRelay mapRelay; @@ -231,6 +235,9 @@ struct CNodeState { ChainSyncTimeoutState m_chain_sync; + //! Time of last new block announcement + int64_t m_last_block_announcement; + CNodeState(CAddress addrIn, std::string addrNameIn) : address(addrIn), name(addrNameIn) { fCurrentlyConnected = false; nMisbehavior = 0; @@ -254,6 +261,7 @@ struct CNodeState { fWantsCmpctWitness = false; fSupportsDesiredCmpctVersion = false; m_chain_sync = { 0, nullptr, false, false }; + m_last_block_announcement = 0; } }; @@ -797,6 +805,8 @@ void PeerLogicValidation::BlockConnected(const std::shared_ptr& pb } LogPrint(BCLog::MEMPOOL, "Erased %d orphan tx included or conflicted by block\n", nErased); } + + g_last_tip_update = GetTime(); } // All of the following cache a recent block, and are protected by cs_most_recent_block @@ -1215,6 +1225,7 @@ bool static ProcessHeadersMessage(CNode *pfrom, CConnman *connman, const std::ve return true; } + bool received_new_header = false; const CBlockIndex *pindexLast = nullptr; { LOCK(cs_main); @@ -1255,6 +1266,12 @@ bool static ProcessHeadersMessage(CNode *pfrom, CConnman *connman, const std::ve } hashLastBlock = header.GetHash(); } + + // If we don't have the last header, then they'll have given us + // something new (if these headers are valid). + if (mapBlockIndex.find(hashLastBlock) == mapBlockIndex.end()) { + received_new_header = true; + } } CValidationState state; @@ -1319,6 +1336,10 @@ bool static ProcessHeadersMessage(CNode *pfrom, CConnman *connman, const std::ve // because it is set in UpdateBlockAvailability. Some nullptr checks // are still present, however, as belt-and-suspenders. + if (received_new_header && pindexLast->nChainWork > chainActive.Tip()->nChainWork) { + nodestate->m_last_block_announcement = GetTime(); + } + if (nCount == MAX_HEADERS_RESULTS) { // Headers message had its maximum size; the peer may have more headers. // TODO: optimize: if pindexLast is an ancestor of chainActive.Tip or pindexBestHeader, continue @@ -2219,6 +2240,8 @@ bool static ProcessMessage(CNode* pfrom, const std::string& strCommand, CDataStr CBlockHeaderAndShortTxIDs cmpctblock; vRecv >> cmpctblock; + bool received_new_header = false; + { LOCK(cs_main); @@ -2228,6 +2251,10 @@ bool static ProcessMessage(CNode* pfrom, const std::string& strCommand, CDataStr connman->PushMessage(pfrom, msgMaker.Make(NetMsgType::GETHEADERS, chainActive.GetLocator(pindexBestHeader), uint256())); return true; } + + if (mapBlockIndex.find(cmpctblock.header.GetHash()) == mapBlockIndex.end()) { + received_new_header = true; + } } const CBlockIndex *pindex = nullptr; @@ -2266,6 +2293,14 @@ bool static ProcessMessage(CNode* pfrom, const std::string& strCommand, CDataStr assert(pindex); UpdateBlockAvailability(pfrom->GetId(), pindex->GetBlockHash()); + CNodeState *nodestate = State(pfrom->GetId()); + + // If this was a new header with more work than our tip, update the + // peer's last block announcement time + if (received_new_header && pindex->nChainWork > chainActive.Tip()->nChainWork) { + nodestate->m_last_block_announcement = GetTime(); + } + std::map::iterator> >::iterator blockInFlightIt = mapBlocksInFlight.find(pindex->GetBlockHash()); bool fAlreadyInFlight = blockInFlightIt != mapBlocksInFlight.end(); @@ -2288,8 +2323,6 @@ bool static ProcessMessage(CNode* pfrom, const std::string& strCommand, CDataStr if (!fAlreadyInFlight && !CanDirectFetch(chainparams.GetConsensus())) return true; - CNodeState *nodestate = State(pfrom->GetId()); - if (IsWitnessEnabled(pindex->pprev, chainparams.GetConsensus()) && !nodestate->fSupportsDesiredCmpctVersion) { // Don't bother trying to process compact blocks from v1 peers // after segwit activates. -- cgit v1.2.3 From ac7b37cd2bd612a64a4009ba82f1cd1d57f37434 Mon Sep 17 00:00:00 2001 From: Suhas Daftuar Date: Tue, 24 Oct 2017 16:56:07 -0400 Subject: Connect to an extra outbound peer if our tip is stale If our tip hasn't updated in a while, that may be because our peers are not relaying blocks to us that we would consider valid. Allow connection to an additional outbound peer in that circumstance. Also, periodically check to see if we are exceeding our target number of outbound peers, and disconnect the one which has least recently announced a new block to us (choosing the newest such peer in the case of tie). --- src/net_processing.cpp | 99 +++++++++++++++++++++++++++++++++++++++++++++++++- 1 file changed, 97 insertions(+), 2 deletions(-) (limited to 'src/net_processing.cpp') diff --git a/src/net_processing.cpp b/src/net_processing.cpp index 59df5d650..3aa13fff4 100644 --- a/src/net_processing.cpp +++ b/src/net_processing.cpp @@ -23,6 +23,7 @@ #include "primitives/transaction.h" #include "random.h" #include "reverse_iterator.h" +#include "scheduler.h" #include "tinyformat.h" #include "txmempool.h" #include "ui_interface.h" @@ -127,7 +128,6 @@ namespace { /** Number of outbound peers with m_chain_sync.m_protect. */ int g_outbound_peers_with_protect_from_disconnect = 0; - /** When our tip was last updated. */ int64_t g_last_tip_update = 0; @@ -435,6 +435,15 @@ void MaybeSetPeerAsAnnouncingHeaderAndIDs(NodeId nodeid, CConnman* connman) { } } +bool TipMayBeStale(const Consensus::Params &consensusParams) +{ + AssertLockHeld(cs_main); + if (g_last_tip_update == 0) { + g_last_tip_update = GetTime(); + } + return g_last_tip_update < GetTime() - consensusParams.nPowTargetSpacing * 3 && mapBlocksInFlight.empty(); +} + // Requires cs_main bool CanDirectFetch(const Consensus::Params &consensusParams) { @@ -772,9 +781,17 @@ static bool StaleBlockRequestAllowed(const CBlockIndex* pindex, const Consensus: (GetBlockProofEquivalentTime(*pindexBestHeader, *pindex, *pindexBestHeader, consensusParams) < STALE_RELAY_AGE_LIMIT); } -PeerLogicValidation::PeerLogicValidation(CConnman* connmanIn) : connman(connmanIn) { +PeerLogicValidation::PeerLogicValidation(CConnman* connmanIn, CScheduler &scheduler) : connman(connmanIn), m_stale_tip_check_time(0) { // Initialize global variables that cannot be constructed at startup. recentRejects.reset(new CRollingBloomFilter(120000, 0.000001)); + + const Consensus::Params& consensusParams = Params().GetConsensus(); + // Stale tip checking and peer eviction are on two different timers, but we + // don't want them to get out of sync due to drift in the scheduler, so we + // combine them in one function and schedule at the quicker (peer-eviction) + // timer. + static_assert(EXTRA_PEER_CHECK_INTERVAL < STALE_CHECK_INTERVAL, "peer eviction timer should be less than stale tip check timer"); + scheduler.scheduleEvery(std::bind(&PeerLogicValidation::CheckForStaleTipAndEvictPeers, this, consensusParams), EXTRA_PEER_CHECK_INTERVAL * 1000); } void PeerLogicValidation::BlockConnected(const std::shared_ptr& pblock, const CBlockIndex* pindex, const std::vector& vtxConflicted) { @@ -1424,6 +1441,7 @@ bool static ProcessHeadersMessage(CNode *pfrom, CConnman *connman, const std::ve // If this is an outbound peer, check to see if we should protect // it from the bad/lagging chain logic. if (g_outbound_peers_with_protect_from_disconnect < MAX_OUTBOUND_PEERS_TO_PROTECT_FROM_DISCONNECT && nodestate->pindexBestKnownBlock->nChainWork >= chainActive.Tip()->nChainWork && !nodestate->m_chain_sync.m_protect) { + LogPrint(BCLog::NET, "Protecting outbound peer=%d from eviction\n", pfrom->GetId()); nodestate->m_chain_sync.m_protect = true; ++g_outbound_peers_with_protect_from_disconnect; } @@ -3004,6 +3022,83 @@ void PeerLogicValidation::ConsiderEviction(CNode *pto, int64_t time_in_seconds) } } +void PeerLogicValidation::EvictExtraOutboundPeers(int64_t time_in_seconds) +{ + // Check whether we have too many outbound peers + int extra_peers = connman->GetExtraOutboundCount(); + if (extra_peers > 0) { + // If we have more outbound peers than we target, disconnect one. + // Pick the outbound peer that least recently announced + // us a new block, with ties broken by choosing the more recent + // connection (higher node id) + NodeId worst_peer = -1; + int64_t oldest_block_announcement = std::numeric_limits::max(); + + LOCK(cs_main); + + connman->ForEachNode([&](CNode* pnode) { + // Ignore non-outbound peers, or nodes marked for disconnect already + if (!IsOutboundDisconnectionCandidate(pnode) || pnode->fDisconnect) return; + CNodeState *state = State(pnode->GetId()); + if (state == nullptr) return; // shouldn't be possible, but just in case + // Don't evict our protected peers + if (state->m_chain_sync.m_protect) return; + if (state->m_last_block_announcement < oldest_block_announcement || (state->m_last_block_announcement == oldest_block_announcement && pnode->GetId() > worst_peer)) { + worst_peer = pnode->GetId(); + oldest_block_announcement = state->m_last_block_announcement; + } + }); + if (worst_peer != -1) { + bool disconnected = connman->ForNode(worst_peer, [&](CNode *pnode) { + // Only disconnect a peer that has been connected to us for + // some reasonable fraction of our check-frequency, to give + // it time for new information to have arrived. + // Also don't disconnect any peer we're trying to download a + // block from. + CNodeState &state = *State(pnode->GetId()); + if (time_in_seconds - pnode->nTimeConnected > MINIMUM_CONNECT_TIME && state.nBlocksInFlight == 0) { + LogPrint(BCLog::NET, "disconnecting extra outbound peer=%d (last block announcement received at time %d)\n", pnode->GetId(), oldest_block_announcement); + pnode->fDisconnect = true; + return true; + } else { + LogPrint(BCLog::NET, "keeping outbound peer=%d chosen for eviction (connect time: %d, blocks_in_flight: %d)\n", pnode->GetId(), pnode->nTimeConnected, state.nBlocksInFlight); + return false; + } + }); + if (disconnected) { + // If we disconnected an extra peer, that means we successfully + // connected to at least one peer after the last time we + // detected a stale tip. Don't try any more extra peers until + // we next detect a stale tip, to limit the load we put on the + // network from these extra connections. + connman->SetTryNewOutboundPeer(false); + } + } + } +} + +void PeerLogicValidation::CheckForStaleTipAndEvictPeers(const Consensus::Params &consensusParams) +{ + if (connman == nullptr) return; + + int64_t time_in_seconds = GetTime(); + + EvictExtraOutboundPeers(time_in_seconds); + + if (time_in_seconds > m_stale_tip_check_time) { + LOCK(cs_main); + // Check whether our tip is stale, and if so, allow using an extra + // outbound peer + if (TipMayBeStale(consensusParams)) { + LogPrintf("Potential stale tip detected, will try using extra outbound peer (last tip update: %d seconds ago)\n", time_in_seconds - g_last_tip_update); + connman->SetTryNewOutboundPeer(true); + } else if (connman->GetTryNewOutboundPeer()) { + connman->SetTryNewOutboundPeer(false); + } + m_stale_tip_check_time = time_in_seconds + STALE_CHECK_INTERVAL; + } +} + class CompareInvMempoolOrder { CTxMemPool *mp; -- cgit v1.2.3 From 626291508c433488439b662f2e88882048fb59fb Mon Sep 17 00:00:00 2001 From: Suhas Daftuar Date: Thu, 26 Oct 2017 10:32:46 -0400 Subject: Add unit test for stale tip checking --- src/net_processing.cpp | 9 +++++++++ 1 file changed, 9 insertions(+) (limited to 'src/net_processing.cpp') diff --git a/src/net_processing.cpp b/src/net_processing.cpp index 3aa13fff4..faa0c7620 100644 --- a/src/net_processing.cpp +++ b/src/net_processing.cpp @@ -550,6 +550,15 @@ void FindNextBlocksToDownload(NodeId nodeid, unsigned int count, std::vectorm_last_block_announcement = time_in_seconds; +} + // Returns true for outbound peers, excluding manual connections, feelers, and // one-shots bool IsOutboundDisconnectionCandidate(const CNode *node) -- cgit v1.2.3