MB-48157 Reduce idle time CPU utilisation of projector

Idle time CPU utilisation is high because the endpoint::run
method wakes every 1ms for flushing the bucker. The ticker
interval has been reduced to 1ms to achieve low session
consistent scan latencies with less KV ops/sec rate.

With this patch, the flushTick is moved back to 25ms. When
a mutation is to be processed, we flush it if there are more
than 100 mutations in the queue or the time since last flush
is greater than 1ms - so, scan latencies should not be impacted.

Change-Id: I67718e6f1b971e46d9137145e6fb5996ca3771d5

secondary/common/config.go

@@ -366,10 +366,10 @@ var SystemConfig = Config{
 		false, // case-insensitive
 	},
 	"projector.dataport.bufferTimeout": ConfigValue{
-		1, // 1ms
+		25, // 25ms
 		"timeout in milliseconds, to flush vbucket-mutations from, " +
 			"endpoint, does not affect existing feeds.",
-		1,     // 1ms
+		25,    // 25ms
 		false, // mutable
 		false, // case-insensitive
 	},

secondary/dataport/endpoint.go

@@ -368,7 +368,7 @@ loop:
 				})
 
 				messageCount++ // count queued up mutations.
-				if messageCount > endpoint.bufferSize {
+				if messageCount > endpoint.bufferSize || time.Since(lastActiveTime) > time.Duration(1*time.Millisecond) {
 					if err := flushBuffers(); err != nil {
 						break loop
 					}