Merge pull request #6756 from brianjohnson5972/6755-fork-fix

Fix net_plugin catchup reported blocks

plugins/net_plugin/net_plugin.cpp

@@ -867,13 +867,7 @@ namespace eosio {
             fc_dlog(logger, "maybe truncating branch at  = ${h}:${id}",("h",remote_head_num)("id",remote_head_id));
          }
 
-         // base our branch off of the last handshake we sent the peer instead of our current
-         // LIB which could have moved forward in time as packets were in flight.
-         if (last_handshake_sent.generation >= 1) {
-            lib_id = last_handshake_sent.last_irreversible_block_id;
-         } else {
-            lib_id = cc.last_irreversible_block_id();
-         }
+         lib_id = last_handshake_recv.last_irreversible_block_id;
          head_id = cc.fork_db_head_block_id();
       }
       catch (const assert_exception& ex) {
@@ -1583,6 +1577,10 @@ namespace eosio {
                set_state(head_catchup);
             }
          }
+
+         if (state == in_sync) {
+            send_handshakes();
+         }
       }
       else if (state == lib_catchup) {
          if( blk_num == sync_known_lib_num ) {

tests/Node.py

@@ -997,6 +997,19 @@ def delegatebw(self, fromAccount, netQuantity, cpuQuantity, toAccount=None, tran
 
         return self.waitForTransBlockIfNeeded(trans, waitForTransBlock, exitOnError=exitOnError)
 
+    def undelegatebw(self, fromAccount, netQuantity, cpuQuantity, toAccount=None, waitForTransBlock=False, exitOnError=False):
+        if toAccount is None:
+            toAccount=fromAccount
+
+        cmdDesc="system undelegatebw"
+        cmd="%s -j %s %s \"%s %s\" \"%s %s\"" % (
+            cmdDesc, fromAccount.name, toAccount.name, netQuantity, CORE_SYMBOL, cpuQuantity, CORE_SYMBOL)
+        msg="fromAccount=%s, toAccount=%s" % (fromAccount.name, toAccount.name);
+        trans=self.processCleosCmd(cmd, cmdDesc, exitOnError=exitOnError, exitMsg=msg)
+        self.trackCmdTransaction(trans)
+
+        return self.waitForTransBlockIfNeeded(trans, waitForTransBlock, exitOnError=exitOnError)
+
     def regproducer(self, producer, url, location, waitForTransBlock=False, exitOnError=False):
         cmdDesc="system regproducer"
         cmd="%s -j %s %s %s %s" % (
@@ -1220,6 +1233,8 @@ def interruptAndVerifyExitStatus(self):
     def verifyAlive(self, silent=False):
         if not silent and Utils.Debug: Utils.Print("Checking if node(pid=%s) is alive(killed=%s): %s" % (self.pid, self.killed, self.cmd))
         if self.killed or self.pid is None:
+            self.killed=True
+            self.pid=None
             return False
 
         try:
@@ -1231,8 +1246,8 @@ def verifyAlive(self, silent=False):
             return False
         except PermissionError as ex:
             return True
-        else:
-            return True
+
+        return True
 
     def getBlockProducerByNum(self, blockNum, timeout=None, waitForBlock=True, exitOnError=True):
         if waitForBlock:

tests/nodeos_forked_chain_test.py

@@ -321,6 +321,8 @@ def getMinHeadAndLib(prodNodes):
     #identify the earliest LIB to start identify the earliest block to check if divergent branches eventually reach concensus
     (headBlockNum, libNumAroundDivergence)=getMinHeadAndLib(prodNodes)
     Print("Tracking block producers from %d till divergence or %d. Head block is %d and lowest LIB is %d" % (preKillBlockNum, lastBlockNum, headBlockNum, libNumAroundDivergence))
+    transitionCount=0
+    missedTransitionBlock=None
     for blockNum in range(preKillBlockNum,lastBlockNum):
         #avoiding getting LIB until my current block passes the head from the last time I checked
         if blockNum>headBlockNum:
@@ -342,12 +344,24 @@ def getMinHeadAndLib(prodNodes):
         if not nextProdChange and prodChanged and blockProducer1==killAtProducer:
             nextProdChange=True
         elif nextProdChange and blockProducer1!=killAtProducer:
-            actualLastBlockNum=blockNum
-            break
+            nextProdChange=False
+            if blockProducer0!=blockProducer1:
+                Print("Divergence identified at block %s, node_00 producer: %s, node_01 producer: %s" % (blockNum, blockProducer0, blockProducer1))
+                actualLastBlockNum=blockNum
+                break
+            else:
+                missedTransitionBlock=blockNum
+                transitionCount+=1
+                # allow this to transition twice, in case the script was identifying an earlier transition than the bridge node received the kill command
+                if transitionCount>1:
+                    Print("At block %d and have passed producer: %s %d times and we have not diverged, stopping looking and letting errors report" % (blockNum, killAtProducer, transitionCount))
+                    actualLastBlockNum=blockNum
+                    break
 
         #if we diverge before identifying the actualLastBlockNum, then there is an ERROR
         if blockProducer0!=blockProducer1:
-            Utils.errorExit("Groups reported different block producers for block number %d. %s != %s." % (blockNum,blockProducer0,blockProducer1))
+            extra="" if transitionCount==0 else " Diverged after expected killAtProducer transition at block %d." % (missedTransitionBlock)
+            Utils.errorExit("Groups reported different block producers for block number %d.%s %s != %s." % (blockNum,extra,blockProducer0,blockProducer1))
 
     #verify that the non producing node is not alive (and populate the producer nodes with current getInfo data to report if
     #an error occurs)
@@ -405,6 +419,23 @@ def getMinHeadAndLib(prodNodes):
 
     #ensure that the nodes have enough time to get in concensus, so wait for 3 producers to produce their complete round
     time.sleep(inRowCountPerProducer * 3 / 2)
+    remainingChecks=20
+    match=False
+    checkHead=False
+    while remainingChecks>0:
+        checkMatchBlock=killBlockNum if not checkHead else prodNodes[0].getBlockNum()
+        blockProducer0=prodNodes[0].getBlockProducerByNum(checkMatchBlock)
+        blockProducer1=prodNodes[1].getBlockProducerByNum(checkMatchBlock)
+        match=blockProducer0==blockProducer1
+        if match:
+            if checkHead:
+                break
+            else:
+                checkHead=True
+                continue
+        Print("Fork has not resolved yet, wait a little more. Block %s has producer %s for node_00 and %s for node_01.  Original divergence was at block %s. Wait time remaining: %d" % (checkMatchBlock, blockProducer0, blockProducer1, killBlockNum, remainingChecks))
+        time.sleep(1)
+        remainingChecks-=1
 
     for prodNode in prodNodes:
         info=prodNode.getInfo()
@@ -426,12 +457,20 @@ def getMinHeadAndLib(prodNodes):
 
     analyzeBPs(blockProducers0, blockProducers1, expectDivergence=False)
 
+    resolvedKillBlockProducer=None
+    for prod in blockProducers0:
+        if prod["blockNum"]==killBlockNum:
+            resolvedKillBlockProducer = prod["prod"]
+    if resolvedKillBlockProducer is None:
+        Utils.errorExit("Did not find find block %s (the original divergent block) in blockProducers0, test setup is wrong.  blockProducers0: %s" % (killBlockNum, ", ".join(blockProducers)))
+    Print("Fork resolved and determined producer %s for block %s" % (resolvedKillBlockProducer, killBlockNum))
+
     blockProducers0=[]
     blockProducers1=[]
 
     testSuccessful=True
 finally:
-    TestHelper.shutdown(cluster, walletMgr, testSuccessful, killEosInstances, killWallet, keepLogs, killAll, dumpErrorDetails)
+    TestHelper.shutdown(cluster, walletMgr, testSuccessful=testSuccessful, killEosInstances=killEosInstances, killWallet=killWallet, keepLogs=keepLogs, cleanRun=killAll, dumpErrorDetails=dumpErrorDetails)
 
     if not testSuccessful:
         Print(Utils.FileDivider)