Attempt to fix flaky integration ApproveForMyOrg

The lifecycle tests use the PeersWithChannel function to get a list of peers to invoke ApproveChaincodeForMyOrg on. That function then picks the first peer from each org in this list. Consequently, a random subset of peers from orgs are picked. Then sometimes, the peers will have the implicit collection private data in their transient store, and sometimes, they will not. By making this function deterministic, we should ensure that these tests consistently get Peer1 for their org, which is almost always the peer which is selected for individual approvals. This is untested (except soon to be in CI) but will hopefully reduce the flakiness of these tests. Signed-off-by: Jason Yellick <jyellick@us.ibm.com>
hyperledger · Dec 10, 2019 · b1d37c6 · b1d37c6
1 parent 206795d
commit b1d37c6
Showing 1 changed file with 12 additions and 0 deletions.
diff --git a/integration/nwo/network.go b/integration/nwo/network.go
@@ -14,6 +14,7 @@ import (
 	"os"
 	"os/exec"
 	"path/filepath"
+	"sort"
 	"strconv"
 	"strings"
 	"syscall"
@@ -1321,6 +1322,17 @@ func (n *Network) PeersWithChannel(chanName string) []*Peer {
 			}
 		}
 	}
+
+	// This is a bit of a hack to make the output of this function deterministic.
+	// When this function's output is supplied as input to functions such as ApproveChaincodeForMyOrg
+	// it causes a different subset of peers to be picked, which can create flakiness in tests.
+	sort.Slice(peers, func(i, j int) bool {
+		if peers[i].Organization < peers[j].Organization {
+			return true
+		}
+
+		return peers[i].Organization == peers[j].Organization && peers[i].Name < peers[j].Name
+	})
 	return peers
 }