DataStax driver metrics
DataStax drivers expose metrics through language-specific libraries and APIs. To troubleshoot performance issues, DataStax recommends collecting most driver metrics in Prometheus.
Why do I need driver metrics?
Performance monitoring is an important aspect of the software development lifecycle. Metrics-based alerts and application monitoring frameworks help operators identify, track, and resolve performance issues quickly.
For example, if you manage a DataStax Enterprise (DSE) deployment with DSE OpsCenter, your operators can receive alerts when there are server-side latency spikes. If you also implement application monitoring, your operators can track latency spikes to individual components, such as single driver instances. Then, they can evaluate how to fix the latency problem in that component without wasting time investigating other parts of the application.
Enable metrics collection in DataStax drivers
C/C++ driver metrics
The C/C++ driver tracks its metrics through the internal CassMetrics object.
This object contains information about requests (latency and throughput), stats (connections), and errors (timeouts).
For more information, see C/C++ driver metrics.
For speculative execution, use the CassSpeculativeExecutionMetrics object.
C# driver metrics
C# driver metrics are disabled by default.
To enable metrics, use the Builder.WithMetrics() method when creating the session.
To export metrics to a monitoring tool, use the App.Metrics Provider extension package.
For speculative execution, use the node-level metric speculative-executions.
For more information, including a list of metrics, see C# driver metrics.
GoCQL driver metrics
Java driver metrics
Java driver metrics are disabled by default.
To enable metrics, use advanced.metrics to select your metrics library and the metrics that you want to monitor.
Depending on your driver version, the Java driver can report metrics to various metrics libraries, including built-in support for popular metrics libraries.
// create any type of registry depending on used library
Object registry = new SimpleMeterRegistry();
CqlSession session = CqlSession.builder()
// ...
.withMetricRegistry(registry)
.build()You must specify each metric that you want to capture.
For a full list of available metrics, see reference.conf.
datastax-java-driver {
advanced.metrics {
session {
enabled = [
bytes-sent,
bytes-received,
connected-nodes,
cql-requests,
cql-client-timeouts,
]
cql-requests {
highest-latency = 3 seconds
lowest-latency = 1 millisecond
significant-digits = 3
refresh-interval = 5 minutes
}
}
node {
enabled = [
pool.open-connections,
pool.available-streams,
pool.in-flight,
cql-messages,
errors.request.unsent,
errors.request.aborted,
errors.request.write-timeouts,
errors.request.read-timeouts,
errors.request.unavailables,
errors.request.others,
retries.total,
retries.aborted,
retries.read-timeout,
retries.write-timeout,
retries.unavailable,
retries.other,
ignores.total,
ignores.aborted,
ignores.read-timeout,
ignores.write-timeout,
ignores.unavailable,
ignores.other,
errors.connection.init
]
cql-messages {
highest-latency = 3 seconds
lowest-latency = 1 millisecond
significant-digits = 3
refresh-interval = 5 minutes
}
expire-after = 1 hour
}
}
}All Java driver versions support using an external metric registry.
For speculative execution, use the speculative-executions metric.
For more information, see Java driver metrics and the Prometheus with logback example project.
Node.js driver metrics
The Node.js driver’s metrics module exposes internal driver metrics in the form of counters:
-
ClientMetrics: An interface that can be used by metrics libraries, service providers, and the community to implement support for existing toolkits like DataDog, Prometheus, and Measured. -
DefaultMetrics: A default implementation ofClientMetricsthat exposes driver metrics as Node.js events, and then push them to your application metrics toolkit.
For speculative execution, use speculativeExecutions.
Python driver metrics
Python driver metrics are disabled by default.
The Python driver offers optional metrics support that requires the scales library.
To enable metrics support, install the scales dependency, and then create the Cluster object with metrics_enabled set to True.
To access the reported statistics, use a simple HTTP server to spot check the metrics.
Alternatively, you can use GraphitePusher to collect and report metrics to Graphite.
