public void Should_RemoveNodeMetricsAndDisposeMetricsContext_When_HostIsRemoved()
{
var metricsRoot = new App.Metrics.MetricsBuilder().Build();
try
{
var cluster = GetNewCluster(b => b.WithMetrics(metricsRoot.CreateDriverMetricsProvider()));
var session = cluster.Connect();
var metrics = session.GetMetrics();
Assert.AreEqual(3, cluster.Metadata.Hosts.Count);
Assert.AreEqual(3, metrics.NodeMetrics.Count);
// get address for host that will be removed from the cluster in this test
var address = TestCluster.ClusterIpPrefix + "2";
var hostToBeRemoved = cluster.Metadata.Hosts.First(h => h.Address.Address.Equals(IPAddress.Parse(address)));
// check node metrics are valid
var gauge = metrics.GetNodeGauge(hostToBeRemoved, NodeMetric.Gauges.OpenConnections);
var appMetricsGaugeValue = metricsRoot.Snapshot.GetGaugeValue(gauge.Context, gauge.Name);
Assert.Greater(gauge.GetValue().Value, 0);
Assert.AreEqual(gauge.GetValue().Value, appMetricsGaugeValue);
// check node metrics context in app metrics is valid
var context = metricsRoot.Snapshot.GetForContext(gauge.Context);
Assert.True(context.IsNotEmpty());
Assert.AreEqual(2, context.Gauges.Count());
// remove host from cluster
TestCluster.DecommissionNode(2);
TestCluster.Stop(2);
try
{
TestHelper.RetryAssert(() => { Assert.AreEqual(2, cluster.Metadata.Hosts.Count, "metadata hosts count failed"); }, 200, 50);
TestHelper.RetryAssert(() => { Assert.AreEqual(2, metrics.NodeMetrics.Count, "Node metrics count failed"); }, 10, 500);
}
catch
{
TestCluster.Start(2, "--jvm_arg=\"-Dcassandra.override_decommission=true\"");
throw;
}
// Check node's metrics were removed from app metrics registry
context = metricsRoot.Snapshot.GetForContext(gauge.Context);
Assert.False(context.IsNotEmpty());
Assert.AreEqual(0, context.Gauges.Count());
TestCluster.Start(2, "--jvm_arg=\"-Dcassandra.override_decommission=true\"");
TestHelper.RetryAssert(() => { Assert.AreEqual(3, cluster.Metadata.Hosts.Count, "metadata hosts count after bootstrap failed"); },
200, 50);
// when new host is chosen by LBP, connection pool is created
foreach (var _ in Enumerable.Range(0, 5))
{
session.Execute("SELECT * FROM system.local");
}
TestHelper.RetryAssert(() => { Assert.AreEqual(3, metrics.NodeMetrics.Count, "Node metrics count after bootstrap failed"); }, 10,
500);
// Check node's metrics were added again
context = metricsRoot.Snapshot.GetForContext(gauge.Context);
Assert.True(context.IsNotEmpty());
Assert.AreEqual(2, context.Gauges.Count());
}
finally
{
metricsRoot.Manage.Disable();
}
}
public async Task Should_ContinueQuerying_When_ANodeGoesDown()
{
var session = await CreateSessionAsync(act : builder =>
builder.WithPoolingOptions(
new PoolingOptions().SetHeartBeatInterval(50))
.WithReconnectionPolicy(new ConstantReconnectionPolicy(40))
.WithQueryOptions(new QueryOptions().SetDefaultIdempotence(true))).ConfigureAwait(false);
Assert.IsTrue(session.Cluster.AllHosts().All(h => h.IsUp));
var restarted = true;
var t = Task.Run(async() =>
{
TestCluster.Stop(1);
await Task.Delay(2000).ConfigureAwait(false);
TestCluster.Start(1, "--jvm_arg \"-Ddse.product_type=DATASTAX_APOLLO\"");
await Task.Delay(500).ConfigureAwait(false);
try
{
TestHelper.RetryAssert(
() =>
{
var dict = Session.Cluster.Metadata.TokenToReplicasMap.GetByKeyspace("system_distributed");
Assert.AreEqual(3, dict.First().Value.Count);
Assert.AreEqual(3, Session.Cluster.AllHosts().Count);
Assert.IsTrue(Session.Cluster.AllHosts().All(h => h.IsUp));
},
20,
500);
}
finally
{
Volatile.Write(ref restarted, true);
}
});
var t2 = Task.Run(async() =>
{
while (true)
{
if (Volatile.Read(ref restarted))
{
return;
}
var tasks = new List <Task>();
long counter = 0;
foreach (var _ in Enumerable.Range(0, 32))
{
tasks.Add(Task.Run(async() =>
{
while (true)
{
var c = Interlocked.Increment(ref counter);
if (c > 1000)
{
return;
}
await session.ExecuteAsync(new SimpleStatement("SELECT key FROM system.local")).ConfigureAwait(false);
}
}));
}
await Task.WhenAll(tasks).ConfigureAwait(false);
}
});
await Task.WhenAll(t, t2).ConfigureAwait(false);
}
public void Should_AllMetricsHaveValidValues_When_NodeIsDown()
{
_metricsRoot = new MetricsBuilder().Build();
TestCluster.Stop(2);
try
{
var cluster = GetNewTemporaryCluster(b => b.WithMetrics(
_metricsRoot.CreateDriverMetricsProvider(),
new DriverMetricsOptions()
.SetEnabledSessionMetrics(SessionMetric.AllSessionMetrics)
.SetEnabledNodeMetrics(NodeMetric.AllNodeMetrics)));
var session = cluster.Connect();
foreach (var i in Enumerable.Range(0, 1000))
{
session.Execute("SELECT * FROM system.local");
}
var metrics = session.GetMetrics();
var downNode = session.Cluster.AllHosts().Single(h => h.Address.Address.ToString() == (TestCluster.ClusterIpPrefix + "2"));
foreach (var h in cluster.AllHosts())
{
foreach (var c in MetricsTests.Counters)
{
if (h.Address.Equals(downNode.Address))
{
Assert.GreaterOrEqual(metrics.GetNodeCounter(h, c).GetValue(), 0);
}
else
{
Assert.AreEqual(0, metrics.GetNodeCounter(h, c).GetValue());
}
}
Assert.AreEqual(2, MetricsTests.Gauges.Length);
Assert.AreEqual(0, metrics.GetNodeGauge(h, NodeMetric.Gauges.InFlight).GetValue());
if (h.Address.Equals(downNode.Address))
{
Assert.AreEqual(0, metrics.GetNodeGauge(h, NodeMetric.Gauges.OpenConnections).GetValue());
Assert.AreEqual(0, metrics.GetNodeTimer(h, NodeMetric.Timers.CqlMessages).GetValue().Histogram.Max);
Assert.AreEqual(0, metrics.GetNodeTimer(h, NodeMetric.Timers.CqlMessages).GetValue().Histogram.Median);
Assert.AreEqual(0, metrics.GetNodeTimer(h, NodeMetric.Timers.CqlMessages).GetValue().Histogram.Min);
Assert.AreEqual(0, metrics.GetNodeTimer(h, NodeMetric.Timers.CqlMessages).GetValue().Histogram.Mean);
Assert.AreEqual(0, metrics.GetNodeTimer(h, NodeMetric.Timers.CqlMessages).GetValue().Histogram.StdDev);
Assert.AreEqual(0, metrics.GetNodeTimer(h, NodeMetric.Timers.CqlMessages).GetValue().Histogram.Sum);
Assert.Greater(metrics.GetNodeTimer(h, NodeMetric.Timers.CqlMessages).GetValue().Histogram.SampleSize, 0);
Assert.AreEqual(0, metrics.GetNodeTimer(h, NodeMetric.Timers.CqlMessages).GetValue().Histogram.LastValue);
Assert.AreEqual(0, metrics.GetNodeTimer(h, NodeMetric.Timers.CqlMessages).GetValue().Histogram.Percentile999);
Assert.AreEqual(0, metrics.GetNodeTimer(h, NodeMetric.Timers.CqlMessages).GetValue().Histogram.Percentile75);
Assert.AreEqual(0, metrics.GetNodeTimer(h, NodeMetric.Timers.CqlMessages).GetValue().Histogram.Percentile95);
Assert.AreEqual(0, metrics.GetNodeTimer(h, NodeMetric.Timers.CqlMessages).GetValue().Histogram.Percentile98);
Assert.AreEqual(0, metrics.GetNodeTimer(h, NodeMetric.Timers.CqlMessages).GetValue().Histogram.Percentile99);
Assert.AreEqual(0, metrics.GetNodeMeter(h, NodeMetric.Meters.BytesSent).GetValue().Count);
Assert.AreEqual(0, metrics.GetNodeMeter(h, NodeMetric.Meters.BytesReceived).GetValue().Count);
}
else
{
Assert.Greater(metrics.GetNodeGauge(h, NodeMetric.Gauges.OpenConnections).GetValue(), 0);
Assert.Greater(metrics.GetNodeTimer(h, NodeMetric.Timers.CqlMessages).GetValue().Histogram.Max, 0);
Assert.Greater(metrics.GetNodeTimer(h, NodeMetric.Timers.CqlMessages).GetValue().Histogram.Median, 0);
Assert.Greater(metrics.GetNodeTimer(h, NodeMetric.Timers.CqlMessages).GetValue().Histogram.Min, 0);
Assert.Greater(metrics.GetNodeTimer(h, NodeMetric.Timers.CqlMessages).GetValue().Histogram.Mean, 0);
Assert.Greater(metrics.GetNodeTimer(h, NodeMetric.Timers.CqlMessages).GetValue().Histogram.StdDev, 0);
Assert.Greater(metrics.GetNodeTimer(h, NodeMetric.Timers.CqlMessages).GetValue().Histogram.Sum, 0);
Assert.Greater(metrics.GetNodeTimer(h, NodeMetric.Timers.CqlMessages).GetValue().Histogram.SampleSize, 0);
Assert.Greater(metrics.GetNodeTimer(h, NodeMetric.Timers.CqlMessages).GetValue().Histogram.LastValue, 0);
Assert.Greater(metrics.GetNodeTimer(h, NodeMetric.Timers.CqlMessages).GetValue().Histogram.Percentile999, 0);
Assert.Greater(metrics.GetNodeTimer(h, NodeMetric.Timers.CqlMessages).GetValue().Histogram.Percentile75, 0);
Assert.Greater(metrics.GetNodeTimer(h, NodeMetric.Timers.CqlMessages).GetValue().Histogram.Percentile95, 0);
Assert.Greater(metrics.GetNodeTimer(h, NodeMetric.Timers.CqlMessages).GetValue().Histogram.Percentile98, 0);
Assert.Greater(metrics.GetNodeTimer(h, NodeMetric.Timers.CqlMessages).GetValue().Histogram.Percentile99, 0);
Assert.Greater(metrics.GetNodeMeter(h, NodeMetric.Meters.BytesSent).GetValue().Count, 0);
Assert.Greater(metrics.GetNodeMeter(h, NodeMetric.Meters.BytesReceived).GetValue().Count, 0);
}
}
Assert.AreEqual(0, metrics.GetSessionCounter(SessionMetric.Counters.CqlClientTimeouts).GetValue());
Assert.Greater(metrics.GetSessionTimer(SessionMetric.Timers.CqlRequests).GetValue().Histogram.Max, 0);
Assert.Greater(metrics.GetSessionMeter(SessionMetric.Meters.BytesSent).GetValue().Count, 0);
Assert.Greater(metrics.GetSessionMeter(SessionMetric.Meters.BytesReceived).GetValue().Count, 0);
Assert.AreEqual(2, metrics.GetSessionGauge(SessionMetric.Gauges.ConnectedNodes).GetValue());
}
finally
{
TestCluster.Start(2);
}
}