[FAB-10008] Fine tune gossip pull timeouts

The gossip request wait time needs to be slightly above the digest wait time
in order to give the initiating peer some time to:
- process the digests received
- send back the requests

Otherwise, the responding peer(s) might abandon the session
because they prune the session after requestWaitTime passes since
the time they send a digest.

Change-Id: Ied95076e067fa9fac04513b53da0536f975aa9ca
Signed-off-by: yacovm <yacovm@il.ibm.com>

Author: yacovm

examples/cluster/config/core.yaml

@@ -119,6 +119,7 @@ peer:
         # Number of peers selected to push messages to
         propagatePeerNum: 3
         # Determines frequency of pull phases(unit: second)
+        # Must be greater than digestWaitTime + responseWaitTime
         pullInterval: 4s
         # Number of peers to pull from
         pullPeerNum: 3
@@ -141,9 +142,11 @@ peer:
         # Buffer size of sending messages
         sendBuffSize: 20
         # Time to wait before pull engine processes incoming digests (unit: second)
+        # Should be slightly smaller than requestWaitTime
         digestWaitTime: 1s
-        # Time to wait before pull engine removes incoming nonce (unit: second)
-        requestWaitTime: 1s
+        # Time to wait before pull engine removes incoming nonce (unit: milliseconds)
+        # Should be slightly bigger than digestWaitTime
+        requestWaitTime: 1500ms
         # Time to wait before pull engine ends pull (unit: second)
         responseWaitTime: 2s
         # Alive check interval(unit: second)

gossip/gossip/algo/pull.go

@@ -34,9 +34,9 @@ import (
 */
 
 const (
-	defDigestWaitTime   = time.Duration(1) * time.Second
-	defRequestWaitTime  = time.Duration(1) * time.Second
-	defResponseWaitTime = time.Duration(2) * time.Second
+	defDigestWaitTime   = time.Duration(1000) * time.Millisecond
+	defRequestWaitTime  = time.Duration(1500) * time.Millisecond
+	defResponseWaitTime = time.Duration(2000) * time.Millisecond
 )
 
 // SetDigestWaitTime sets the digest wait time

gossip/gossip/algo/pull_test.go

@@ -553,9 +553,9 @@ func TestDefaultConfig(t *testing.T) {
 	// Check if we can read default duration when no properties are
 	// defined in config file.
 	viper.Reset()
-	assert.Equal(t, time.Duration(1)*time.Second, util.GetDurationOrDefault("peer.gossip.digestWaitTime", defDigestWaitTime))
-	assert.Equal(t, time.Duration(1)*time.Second, util.GetDurationOrDefault("peer.gossip.requestWaitTime", defRequestWaitTime))
-	assert.Equal(t, time.Duration(2)*time.Second, util.GetDurationOrDefault("peer.gossip.responseWaitTime", defResponseWaitTime))
+	assert.Equal(t, time.Duration(1000)*time.Millisecond, util.GetDurationOrDefault("peer.gossip.digestWaitTime", defDigestWaitTime))
+	assert.Equal(t, time.Duration(1500)*time.Millisecond, util.GetDurationOrDefault("peer.gossip.requestWaitTime", defRequestWaitTime))
+	assert.Equal(t, time.Duration(2000)*time.Millisecond, util.GetDurationOrDefault("peer.gossip.responseWaitTime", defResponseWaitTime))
 
 	// Check if the properties in the config file (core.yaml)
 	// are set to the desired duration.
@@ -567,9 +567,9 @@ func TestDefaultConfig(t *testing.T) {
 	viper.AutomaticEnv()
 	err := viper.ReadInConfig()
 	assert.NoError(t, err)
-	assert.Equal(t, time.Duration(1)*time.Second, util.GetDurationOrDefault("peer.gossip.digestWaitTime", defDigestWaitTime))
-	assert.Equal(t, time.Duration(1)*time.Second, util.GetDurationOrDefault("peer.gossip.requestWaitTime", defRequestWaitTime))
-	assert.Equal(t, time.Duration(2)*time.Second, util.GetDurationOrDefault("peer.gossip.responseWaitTime", defResponseWaitTime))
+	assert.Equal(t, time.Duration(1000)*time.Millisecond, util.GetDurationOrDefault("peer.gossip.digestWaitTime", defDigestWaitTime))
+	assert.Equal(t, time.Duration(1500)*time.Millisecond, util.GetDurationOrDefault("peer.gossip.requestWaitTime", defRequestWaitTime))
+	assert.Equal(t, time.Duration(2000)*time.Millisecond, util.GetDurationOrDefault("peer.gossip.responseWaitTime", defResponseWaitTime))
 }
 
 func Strcmp(a interface{}, b interface{}) bool {

sampleconfig/core.yaml

@@ -168,6 +168,7 @@ peer:
         # Number of peers selected to push messages to
         propagatePeerNum: 3
         # Determines frequency of pull phases(unit: second)
+        # Must be greater than digestWaitTime + responseWaitTime
         pullInterval: 4s
         # Number of peers to pull from
         pullPeerNum: 3
@@ -190,9 +191,11 @@ peer:
         # Buffer size of sending messages
         sendBuffSize: 200
         # Time to wait before pull engine processes incoming digests (unit: second)
+        # Should be slightly smaller than requestWaitTime
         digestWaitTime: 1s
-        # Time to wait before pull engine removes incoming nonce (unit: second)
-        requestWaitTime: 1s
+        # Time to wait before pull engine removes incoming nonce (unit: milliseconds)
+        # Should be slightly bigger than digestWaitTime
+        requestWaitTime: 1500ms
         # Time to wait before pull engine ends pull (unit: second)
         responseWaitTime: 2s
         # Alive check interval(unit: second)