From 77615b900f94666a91e196c459753e8940bb2773 Mon Sep 17 00:00:00 2001
From: Sam Kleinman <garen@tychoish.com>
Date: Fri, 3 Sep 2021 13:03:16 -0400
Subject: [PATCH] e2e: wait for all nodes rather than just one (#6892)

---
 test/e2e/pkg/testnet.go  |  1 +
 test/e2e/runner/rpc.go   | 57 ++++++++++++++++++++++++++++++++++++----
 test/e2e/runner/start.go |  2 ++
 3 files changed, 55 insertions(+), 5 deletions(-)

diff --git a/test/e2e/pkg/testnet.go b/test/e2e/pkg/testnet.go
index 265a413a7..3195ca3ba 100644
--- a/test/e2e/pkg/testnet.go
+++ b/test/e2e/pkg/testnet.go
@@ -98,6 +98,7 @@ type Node struct {
 	LogLevel         string
 	UseLegacyP2P     bool
 	QueueType        string
+	HasStarted       bool
 }
 
 // LoadTestnet loads a testnet from a manifest file, using the filename to
diff --git a/test/e2e/runner/rpc.go b/test/e2e/runner/rpc.go
index 277962581..f7b9fd1c9 100644
--- a/test/e2e/runner/rpc.go
+++ b/test/e2e/runner/rpc.go
@@ -17,21 +17,47 @@ import (
 // progress at all.
 func waitForHeight(testnet *e2e.Testnet, height int64) (*types.Block, *types.BlockID, error) {
 	var (
-		err          error
-		maxResult    *rpctypes.ResultBlock
-		clients      = map[string]*rpchttp.HTTP{}
-		lastIncrease = time.Now()
+		err             error
+		maxResult       *rpctypes.ResultBlock
+		clients         = map[string]*rpchttp.HTTP{}
+		lastIncrease    = time.Now()
+		nodesAtHeight   = map[string]struct{}{}
+		numRunningNodes int
 	)
+	for _, node := range testnet.Nodes {
+		if node.Mode == e2e.ModeSeed {
+			continue
+		}
+
+		if node.Mode == e2e.ModeLight {
+			continue
+		}
+
+		if node.HasStarted {
+			numRunningNodes++
+		}
+	}
 
 	for {
 		for _, node := range testnet.Nodes {
+			// skip nodes that have reached the target height
+			if _, ok := nodesAtHeight[node.Name]; ok {
+				continue
+			}
+
 			if node.Mode == e2e.ModeSeed {
 				continue
 			}
+
 			if node.Mode == e2e.ModeLight {
 				continue
 			}
 
+			if !node.HasStarted {
+				continue
+			}
+
+			// cache the clients
 			client, ok := clients[node.Name]
 			if !ok {
 				client, err = node.Client()
@@ -51,7 +77,23 @@ func waitForHeight(testnet *e2e.Testnet, height int64) (*types.Block, *types.Blo
 				maxResult = result
 				lastIncrease = time.Now()
 			}
+
 			if maxResult != nil && maxResult.Block.Height >= height {
+				// the node has achieved the target height!
+
+				// add this node to the set of target
+				// height nodes
+				nodesAtHeight[node.Name] = struct{}{}
+
+				// if not all of the nodes that we
+				// have clients for have reached the
+				// target height, keep trying.
+				if numRunningNodes > len(nodesAtHeight) {
+					continue
+				}
+
+				// return once all nodes have reached
+				// the target height.
 				return maxResult.Block, &maxResult.BlockID, nil
 			}
 		}
@@ -63,7 +105,12 @@ func waitForHeight(testnet *e2e.Testnet, height int64) (*types.Block, *types.Blo
 			if maxResult == nil {
 				return nil, nil, errors.New("chain stalled at unknown height")
 			}
-			return nil, nil, fmt.Errorf("chain stalled at height %v", maxResult.Block.Height)
+
+			return nil, nil, fmt.Errorf("chain stalled at height %v [%d of %d nodes]",
+				maxResult.Block.Height,
+				len(nodesAtHeight),
+				numRunningNodes)
+
 		}
 		time.Sleep(1 * time.Second)
 	}
diff --git a/test/e2e/runner/start.go b/test/e2e/runner/start.go
index 8a241690c..1d0f02eb4 100644
--- a/test/e2e/runner/start.go
+++ b/test/e2e/runner/start.go
@@ -48,6 +48,7 @@ func Start(testnet *e2e.Testnet) error {
 		if _, err := waitForNode(node, 0, time.Minute); err != nil {
 			return err
 		}
+		node.HasStarted = true
 		logger.Info(fmt.Sprintf("Node %v up on http://127.0.0.1:%v", node.Name, node.ProxyPort))
 	}
 
@@ -96,6 +97,7 @@ func Start(testnet *e2e.Testnet) error {
 		if err != nil {
 			return err
 		}
+		node.HasStarted = true
 		logger.Info(fmt.Sprintf("Node %v up on http://127.0.0.1:%v at height %v",
 			node.Name, node.ProxyPort, status.SyncInfo.LatestBlockHeight))
 	}