WARNING: Version 5.x has passed its EOL date.

This documentation is no longer being maintained and may be removed. If you are running this version, we strongly advise you to upgrade. For the latest information, see the current release documentation.

« Sniffing periodically Pinging behaviour »
Elastic Docs ›Elasticsearch.Net and NEST: the .NET clients [5.x] ›API Conventions ›Sniffing behaviour

Sniffing role detection

edit

Sniffing role detection

edit

When we sniff the cluster state, we detect the role of each node, for example, whether it’s master eligible, a node that holds data, etc. We can then use this information when selecting a node to perform an API call on.

var audit = new Auditor(() => Framework.Cluster
    .Nodes(10)
    .Sniff(s => s.Fails(Always))
    .Sniff(s => s.OnPort(9202)
        .Succeeds(Always, Framework.Cluster.Nodes(8).MasterEligible(9200, 9201, 9202))
    )
    .SniffingConnectionPool()
    .AllDefaults()
)
{
    AssertPoolBeforeStartup = (pool) =>
    {
        pool.Should().NotBeNull();
        pool.Nodes.Should().HaveCount(10);
        pool.Nodes.Where(n => n.MasterEligible).Should().HaveCount(10);
    },
    AssertPoolAfterStartup = (pool) =>
    {
        pool.Should().NotBeNull();
        pool.Nodes.Should().HaveCount(8);
        pool.Nodes.Where(n => n.MasterEligible).Should().HaveCount(3);
    }
};

await audit.TraceStartup();

var audit = new Auditor(() => Framework.Cluster
    .Nodes(10)
    .Sniff(s => s.Fails(Always))
    .Sniff(s => s.OnPort(9202)
        .Succeeds(Always, Framework.Cluster.Nodes(8).StoresNoData(9200, 9201, 9202))
    )
    .SniffingConnectionPool()
    .AllDefaults()
)
{
    AssertPoolBeforeStartup = (pool) =>
    {
        pool.Should().NotBeNull();
        pool.Nodes.Should().HaveCount(10);
        pool.Nodes.Where(n => n.HoldsData).Should().HaveCount(10);
    },

    AssertPoolAfterStartup = (pool) =>
    {
        pool.Should().NotBeNull();
        pool.Nodes.Should().HaveCount(8);
        pool.Nodes.Where(n => n.HoldsData).Should().HaveCount(5);
    }
};
await audit.TraceStartup();

var audit = new Auditor(() => Framework.Cluster
    .Nodes(10)
    .Sniff(s => s.SucceedAlways()
        .Succeeds(Always, Framework.Cluster.Nodes(8).StoresNoData(9200, 9201, 9202).HttpDisabled(9201))
    )
    .SniffingConnectionPool()
    .AllDefaults()
)
{
    AssertPoolBeforeStartup = (pool) =>
    {
        pool.Should().NotBeNull();
        pool.Nodes.Should().HaveCount(10);
        pool.Nodes.Where(n => n.HoldsData).Should().HaveCount(10);
        pool.Nodes.Where(n => n.HttpEnabled).Should().HaveCount(10);
        pool.Nodes.Should().OnlyContain(n => n.Uri.Host == "localhost");
    },

    AssertPoolAfterStartup = (pool) =>
    {
        pool.Should().NotBeNull();
        pool.Nodes.Should().HaveCount(7, "we filtered the node that has no http enabled");
        pool.Nodes.Should().NotContain(n=>n.Uri.Port == 9201);
        pool.Nodes.Where(n => n.HoldsData).Should().HaveCount(5);
    }
};
await audit.TraceStartup();

In this example, We create a Virtual cluster with a Sniffing connection pool that seeds all the known master nodes. When the client sniffs on startup, we see that the cluster is 20 nodes in total, with the master eligible nodes storing no data.

var masterNodes = new[] {9200, 9201, 9202};
var totalNodesInTheCluster = 20;
//
var audit = new Auditor(() => Framework.Cluster
    .MasterOnlyNodes(masterNodes.Length)
    .Sniff(s => s.SucceedAlways()
        .Succeeds(Always, Framework.Cluster
            .Nodes(totalNodesInTheCluster)
            .StoresNoData(masterNodes)
            .MasterEligible(masterNodes))
    )
    .SniffingConnectionPool()
    .Settings(s=>s.DisablePing())
)
{
    AssertPoolBeforeStartup = pool => 
    {
        pool.Should().NotBeNull();
        pool.Nodes.Should().HaveCount(3, "we seeded 3 master only nodes at the start of the application");
        pool.Nodes.Where(n => n.HoldsData).Should().HaveCount(0, "none of which hold data");
    },
    AssertPoolAfterStartup = (pool) => 
    {
        pool.Should().NotBeNull();
        var nodes = pool.CreateView().ToList();
        nodes.Count().Should().Be(20, "Master nodes are included in the registration of nodes since we still favor sniffing on them");
    }
};

Before the sniff, assert we only see three master only nodes

After the sniff, assert we now know about the existence of 20 nodes.

After the sniff has happened on 9200 before the first API call, assert that the subsequent API call hits the first non master eligible node

audit = await audit.TraceStartup(new ClientCall
{
    { SniffSuccess, 9200},
    { HealthyResponse, 9203} 
});

Healthy response from 9203, not a master eligible node

To verify that the client behaves as we expect when making requests to the virtual cluster, make 1000 different client calls and assert that each is not sent to any of the known master only nodes

var seenNodes = new HashSet<int>();
foreach (var _ in Enumerable.Range(0, 1000))
{
    audit = await audit.TraceCalls(
        new ClientCall {{HealthyResponse, (a) =>
        {
            var port = a.Node.Uri.Port;
            masterNodes.Should().NotContain(port);
            seenNodes.Add(port);
        }}}
    );
}

seenNodes.Should().HaveCount(totalNodesInTheCluster - masterNodes.Length);

seenNodes is a hash set of all the ports we hit. assert that this is equal to known total nodes - known master only nodes

Node predicates

edit

A predicate can be specified on ConnectionSettings that can be used to determine which nodes in the cluster API calls can be executed on.

As an example, Let’s create a Virtual cluster with a Sniffing connection pool that seeds all 20 nodes to begin. When the client sniffs on startup, we see the cluster is still 20 nodes in total, however we are now aware of the actual configured settings for the nodes from the cluster response.

var totalNodesInTheCluster = 20;
var setting = "node.attr.rack_id";
var value = "rack_one";
var nodesInRackOne = new[] {9204, 9210, 9213};

var audit = new Auditor(() => Framework.Cluster
    .Nodes(totalNodesInTheCluster)
    //
    .Sniff(s => s.SucceedAlways()
        .Succeeds(Always, Framework.Cluster
            .Nodes(totalNodesInTheCluster)
            .HasSetting(setting, value, nodesInRackOne))
    )
    .SniffingConnectionPool()
    .Settings(s=>s
        .DisablePing() 
        .NodePredicate(node => 
            node.Settings.ContainsKey(setting) &&
            node.Settings[setting] == value
        )
    )
)
{
    AssertPoolAfterStartup = pool => 
    {
        pool.Should().NotBeNull();
        var nodes = pool.CreateView().ToList();
        nodes.Count(n => n.Settings.ContainsKey(setting)).Should().Be(3, "only three nodes are in rack_one");
    }
};

for testing simplicity, disable pings

We only want to execute API calls to nodes in rack_one

After sniffing on startup, assert that the pool of nodes that the client will execute API calls against only contains the three nodes that are in rack_one

With the cluster set up, assert that the sniff happens on 9200 before the first API call and that API call hits the first node in rack_one

audit = await audit.TraceStartup(new ClientCall
{
    { SniffSuccess, 9200},
    { HealthyResponse, 9204}
});

To prove that the client is working as expected, do a 1000 different client calls and assert that each is sent to a node in rack_one only, respecting the node predicate on connection settings

var seenNodes = new HashSet<int>();
foreach (var _ in Enumerable.Range(0, 1000))
{
    audit = await audit.TraceCalls(
        new ClientCall {{HealthyResponse, (a) =>
        {
            var port = a.Node.Uri.Port;
            nodesInRackOne.Should().Contain(port);
            seenNodes.Add(port);
        }}}
    );
}

seenNodes.Should().HaveCount(nodesInRackOne.Length);

As another example of node predicates, let’s set up a Virtual cluster with a bad node predicate, i.e. predicate that filters out all nodes from being the targets of API calls from the client

var totalNodesInTheCluster = 20;

var audit = new Auditor(() => Framework.Cluster
    .Nodes(totalNodesInTheCluster)
    .Sniff(s => s.SucceedAlways()
        .Succeeds(Always, Framework.Cluster.Nodes(totalNodesInTheCluster))
    )
    .SniffingConnectionPool()
    .Settings(s => s
        .DisablePing()
        .NodePredicate(node => false) 
    )
);

A bad predicate that declines all nodes

Now when making the client calls, the audit trail indicates that a sniff on startup succeeds, but the subsequent API call fails because the node predicate filters out all nodes as targets on which to execute API calls

await audit.TraceUnexpectedElasticsearchException(new ClientCall
{
    { SniffOnStartup }, 
    { SniffSuccess }, 
    { NoNodesAttempted } 
}, e =>
{
    e.FailureReason.Should().Be(PipelineFailure.Unexpected);

    Func<string> debug = () => e.DebugInformation;
    debug.Invoking(s => s.Invoke()).Should().NotThrow();
});

The audit trail indicates a sniff for the very first time on startup

The sniff succeeds because the node predicate is ignored when sniffing

when trying to do an actual API call however, the predicate prevents any nodes from being attempted
« Sniffing periodically Pinging behaviour »