Added performance-related metrics for the Topology and Intents:
===
* "Topology.EventNotification.LastEventTimestamp"
Timestamp of the last Topology event (system nanoseconds)
* "Topology.EventNotification.ListenerEventRate"
Rate of the Topology events published to the Topology listeners
===
* "Intents.AddOperation.BeginOperationTimestamp"
Timestamp of the incoming Add Intent API operation (system nanoseconds)
* "Intents.AddOperation.EndOperationTimestamp"
Timestamp of the Add Intent operation completion (system nanoseconds)
* "Intents.AddOperation.IncomingRate"
Rate of the incoming Add Intent API operations
* "Intents.AddOperation.ProcessingRate"
Rate of processing the Add Intent operations
===
* "Intents.RemoveOperation.BeginOperationTimestamp"
Timestamp of the incoming Remove Intent API operation (system nanoseconds)
* "Intents.RemoveOperation.EndOperationTimestamp"
Timestamp of the Remove Intent operation completion (system nanoseconds)
* "Intents.RemoveOperation.IncomingRate"
Rate of the incoming Remove Intent API operations
* "Intents.RemoveOperation.ProcessingRate"
Rate of processing the Remove Intent operations
===
All performance metrics are exposed via the Metrics REST API:
* GET all metrics:
url = "http://%s:%s/wm/onos/metrics" % (self.onos_ip, self.onos_port)
* GET a specific metric:
url = "http://%s:%s/wm/onos/metrics?ids=%s" % (self.onos_ip, self.onos_port, args.metric_id)
where "metric_id" is the name of the Metric. E.g.:
ids=Topology.EventNotification.LastEventTimestamp
* GET multiple metrics:
url = "http://%s:%s/wm/onos/metrics?ids=%s" % (self.onos_ip, self.onos_port, args.metric_id)
where "metric_id" is comma-separated list of Metric names. E.g:
ids=Topology.EventNotification.LastEventTimestamp,Topology.EventNotification.ListenerEventRate
===
The JSON format of the output is the following:
{
"meters": [
{
"name": "Intents.AddOperation.IncomingRate",
"meter": {
"count": 2,
"mean_rate": 0.007488255279864508,
"m5_rate": 0.006082798637345469,
"m15_rate": 0.021083988195124116,
"units": "events/second",
"m1_rate": 0.002155350653737004
}
},
...
],
"histograms": [],
"timers": [],
"gauges": [
{
"gauge": {
"value": 179956769775795
},
"name": "Intents.AddOperation.BeginOperationTimestamp"
},
...
],
"counters": []
}
where
- "meter.count" is the number of events
- "meter.mean_rate" is the mean rate of the events
- "meter.m5_rate" is the rate of the events over the last 5-minute interval
- "meter.m15_rate" is the rate of the events over the last 15-minute interval
- "meter.m1_rate" is the rate of the events over the last 1-minute interval
- "meter.units" is the units of the rate (should be "events/second")
- "gauge.value" is the value of the particular metric.
In case of the "*Timestamp" metrics, it is the system nanoseconds
timestamp for the particular event or operation.
Also, fixed some of the unit tests.
NOTE: Currently, for some of the tests we have to explicitly
remove all metrics when tearing-down a test.
In the future we should have a single base unit test class,
and only its tearDown() method should do such cleanup.
Change-Id: Iad5b47b908a29dcfd9fb08e7a010ddf2627fd808
diff --git a/src/main/java/net/onrc/onos/core/topology/TopologyManager.java b/src/main/java/net/onrc/onos/core/topology/TopologyManager.java
index 7e7f2b8..aeb06c2 100644
--- a/src/main/java/net/onrc/onos/core/topology/TopologyManager.java
+++ b/src/main/java/net/onrc/onos/core/topology/TopologyManager.java
@@ -26,6 +26,9 @@
import net.onrc.onos.core.datastore.topology.KVLink;
import net.onrc.onos.core.datastore.topology.KVPort;
import net.onrc.onos.core.datastore.topology.KVSwitch;
+import net.onrc.onos.core.metrics.OnosMetrics;
+import net.onrc.onos.core.metrics.OnosMetrics.MetricsComponent;
+import net.onrc.onos.core.metrics.OnosMetrics.MetricsFeature;
import net.onrc.onos.core.registry.IControllerRegistryService;
import net.onrc.onos.core.util.Dpid;
import net.onrc.onos.core.util.EventEntry;
@@ -36,6 +39,8 @@
import org.slf4j.Logger;
import org.slf4j.LoggerFactory;
+import com.codahale.metrics.Gauge;
+import com.codahale.metrics.Meter;
import com.esotericsoftware.kryo.Kryo;
/**
@@ -69,6 +74,32 @@
private Kryo kryo = KryoFactory.newKryoObject();
//
+ // Metrics
+ //
+ private static final MetricsComponent METRICS_COMPONENT =
+ OnosMetrics.registerComponent("Topology");
+ private static final MetricsFeature METRICS_FEATURE_EVENT_NOTIFICATION =
+ METRICS_COMPONENT.registerFeature("EventNotification");
+ //
+ // Timestamp of the last Topology event (system nanoseconds)
+ private volatile long lastEventTimestamp = 0;
+ private final Gauge<Long> gaugeLastEventTimestamp =
+ OnosMetrics.registerMetric(METRICS_COMPONENT,
+ METRICS_FEATURE_EVENT_NOTIFICATION,
+ "LastEventTimestamp",
+ new Gauge<Long>() {
+ @Override
+ public Long getValue() {
+ return lastEventTimestamp;
+ }
+ });
+ // Rate of the Topology events published to the Topology listeners
+ private final Meter listenerEventRate =
+ OnosMetrics.createMeter(METRICS_COMPONENT,
+ METRICS_FEATURE_EVENT_NOTIFICATION,
+ "ListenerEventRate");
+
+ //
// Local state for keeping track of reordered events.
// NOTE: Switch Events are not affected by the event reordering.
//
@@ -466,11 +497,23 @@
}
}
+ //
+ // Update the metrics
+ //
+ long totalEvents =
+ apiAddedSwitchEvents.size() + apiRemovedSwitchEvents.size() +
+ apiAddedPortEvents.size() + apiRemovedPortEvents.size() +
+ apiAddedLinkEvents.size() + apiRemovedLinkEvents.size() +
+ apiAddedHostEvents.size() + apiRemovedHostEvents.size();
+ this.listenerEventRate.mark(totalEvents);
+ this.lastEventTimestamp = System.nanoTime();
+
+ //
// Deliver the events
- long timestamp = System.nanoTime();
+ //
for (ITopologyListener listener : this.topologyListeners) {
TopologyEvents events =
- new TopologyEvents(timestamp,
+ new TopologyEvents(lastEventTimestamp,
kryo.copy(apiAddedSwitchEvents),
kryo.copy(apiRemovedSwitchEvents),
kryo.copy(apiAddedPortEvents),