consensus changes: faster catch up, delay when out of sync

* Consider there is consensus when we detect that we've fallen behind (old code was waiting to reach consensus with other part of the network); this increases the chance that we catch up to the network. * If we can't catch up to the last closed ledger in SQL consider ourselves syncing
stellar-deprecated · Nov 18, 2014 · 067d715 · 067d715
1 parent a4d6790
commit 067d715
Show file tree

Hide file tree

Showing 5 changed files with 40 additions and 18 deletions.
diff --git a/src/ripple_app/consensus/LedgerConsensus.cpp b/src/ripple_app/consensus/LedgerConsensus.cpp
@@ -57,7 +57,7 @@ class LedgerConsensusImp
             (boost::posix_time::microsec_clock::universal_time ())
     {
         WriteLog (lsDEBUG, LedgerConsensus) << "Creating consensus object";
-        WriteLog (lsTRACE, LedgerConsensus) 
+        WriteLog (lsDEBUG, LedgerConsensus) 
             << "LCL:" << previousLedger->getHash () << ", ct=" << closeTime;
         mPreviousProposers = getApp().getOPs ().getPreviousProposers ();
         mPreviousMSeconds = getApp().getOPs ().getPreviousConvergeTime ();

diff --git a/src/ripple_app/ledger/LedgerMaster.cpp b/src/ripple_app/ledger/LedgerMaster.cpp
@@ -659,8 +659,11 @@ class LedgerMasterImp
 
             InboundLedger::pointer l =
                 getApp().getInboundLedgers().findCreate(hash, 0, InboundLedger::fcGENERIC);
-            if (l && l->isComplete() && !l->isFailed())
+            if (l && l->isComplete() && !l->isFailed()) {
+                WriteLog (lsDEBUG, LedgerMaster) << 
+                    "checkAccept has complete ledger " << to_string (hash);
                 ledger = l->getLedger();
+            }
             else
             {
                 WriteLog (lsDEBUG, LedgerMaster) << 

diff --git a/src/ripple_app/ledger/LedgerTiming.cpp b/src/ripple_app/ledger/LedgerTiming.cpp
@@ -115,7 +115,7 @@ bool ContinuousLedgerTiming::haveConsensus (
     bool forReal,               // deciding whether to stop consensus process
     bool& failed)               // we can't reach a consensus
 {
-    WriteLog (lsTRACE, LedgerTiming) <<
+    WriteLog (lsDEBUG, LedgerTiming) <<
         "CLC::haveConsensus: prop=" << currentProposers <<
         "/" << previousProposers <<
         " agree=" << currentAgree << " validated=" << currentFinished <<
@@ -131,39 +131,49 @@ bool ContinuousLedgerTiming::haveConsensus (
 		return true;
 	}
 
-	if(currentAgreeTime <= LEDGER_MIN_CONSENSUS_TIME)
+    // move on asap if we are behind
+    // If 80% of the nodes on your UNL have moved on, you should declare consensus
+    if (((currentFinished * 100) / (currentProposers + 1)) > 80)
+    {
+        CondLog (forReal, lsWARNING, LedgerTiming) <<
+            "We see no consensus, but 80% of nodes have moved on";
+        failed = true;
+        return true;
+    }
+
+    if (currentAgreeTime <= LEDGER_MIN_CONSENSUS_TIME)
+    {
         return false;
+    }
 
     if (currentProposers < (previousProposers * 3 / 4))
     {
+        /* SANITY: probably need to not allow for this and replace the code with :
+        CondLog (forReal, lsDEBUG, LedgerTiming) << "cannot enter consensus: too few proposers";
+        return false;
+        */
+
         // Less than 3/4 of the last ledger's proposers are present, we may need more time
-		if(currentAgreeTime < (previousAgreeTime + LEDGER_MIN_CONSENSUS_TIME))
+        if (currentAgreeTime < (previousAgreeTime + LEDGER_MIN_CONSENSUS_TIME))
         {
-			WriteLog(lsTRACE, LedgerTiming) <<
+            CondLog (forReal, lsDEBUG, LedgerTiming) <<
                 "too fast, not enough proposers";
             return false;
         }
+        CondLog (forReal, lsDEBUG, LedgerTiming) <<
+                "Ignoring lack of proposers as we're " << (currentAgreeTime - previousAgreeTime) << " ms passed last concensus";
     }
 
     // If 80% of current proposers (plus us) agree on a set, we have consensus
     if (((currentAgree * 100 + 100) / (currentProposers + 1)) > 80)
     {
-		WriteLog(lsTRACE, LedgerTiming) << "normal consensus";
+        CondLog (forReal, lsINFO, LedgerTiming) << "normal consensus";
         failed = false;
         return true;
     }
 
-    // If 80% of the nodes on your UNL have moved on, you should declare consensus
-    if (((currentFinished * 100) / (currentProposers + 1)) > 80)
-    {
-		WriteLog(lsTRACE, LedgerTiming) <<
-            "We see no consensus, but 80% of nodes have moved on";
-        failed = true;
-        return true;
-    }
-
     // no consensus yet
-	WriteLog(lsTRACE, LedgerTiming) << "no consensus";
+    CondLog (forReal, lsDEBUG, LedgerTiming) << "no consensus";
     return false;
 }
 

diff --git a/src/ripple_app/misc/NetworkOPs.cpp b/src/ripple_app/misc/NetworkOPs.cpp
@@ -972,6 +972,15 @@ int NetworkOPsImp::beginConsensus (uint256 const& networkClosed, Ledger::pointer
 
         return 3;
     }
+    else if (!stellar::gLedgerMaster->ensureSync(prevLedger, true))
+    {
+        if (mMode == omTRACKING || mMode == omFULL)
+        {
+            m_journal.warning << "Don't have LCL in database, going back to syncing";
+            setMode (omSYNCING);
+        }
+        return 3;
+    }
 
     assert (prevLedger->getHash () == closingLedger->getParentHash ());
     assert (closingLedger->getParentHash () == m_ledgerMaster.getClosedLedger ()->getHash ());

diff --git a/src/ripple_core/functional/Config.cpp b/src/ripple_core/functional/Config.cpp
@@ -126,7 +126,7 @@ Config::Config ()
 
     TRANSACTION_FEE_BASE    = DEFAULT_FEE_DEFAULT;
 
-    NETWORK_QUORUM          = 0;    // Don't need to see other nodes
+    NETWORK_QUORUM          = 1;
     VALIDATION_QUORUM       = 1;    // Only need one node to vouch
 
     FEE_ACCOUNT_RESERVE     = DEFAULT_FEE_ACCOUNT_RESERVE;