public class Metrics extends Object
The metrics exposed by this class use the Metrics library and you should refer its documentation for details on how to handle the exposed metric objects.
By default, metrics are exposed through JMX, which is very useful for development and browsing, but for production environments you may want to have a look at the reporters provided by the Metrics library which could be more efficient/adapted.
Modifier and Type | Class and Description |
---|---|
class |
Metrics.Errors
Metrics on errors encountered.
|
Modifier and Type | Method and Description |
---|---|
com.codahale.metrics.Gauge<Integer> |
getBlockingExecutorQueueDepth()
Returns the number of queued up tasks in the
blocking executor . |
com.codahale.metrics.Meter |
getBytesReceived()
Returns the number of bytes received so far.
|
com.codahale.metrics.Meter |
getBytesSent()
Returns the number of bytes sent so far.
|
com.codahale.metrics.Gauge<Integer> |
getConnectedToHosts()
Returns the number of Cassandra hosts the driver is currently connected to (that is have at
least one connection opened to).
|
Metrics.Errors |
getErrorMetrics()
Returns an object grouping metrics related to the errors encountered.
|
com.codahale.metrics.Gauge<Integer> |
getExecutorQueueDepth()
Returns the number of queued up tasks in the
main internal executor . |
com.codahale.metrics.Gauge<Integer> |
getInFlightRequests()
Returns the total number of in flight requests to Cassandra hosts.
|
com.codahale.metrics.Gauge<Integer> |
getKnownHosts()
Returns the number of Cassandra hosts currently known by the driver (that is whether they are
currently considered up or down).
|
com.codahale.metrics.Gauge<Integer> |
getOpenConnections()
Returns the total number of currently opened connections to Cassandra hosts.
|
com.codahale.metrics.Gauge<Integer> |
getReconnectionSchedulerQueueSize()
Returns the number of queued up tasks in the
reconnection executor . |
com.codahale.metrics.MetricRegistry |
getRegistry()
Returns the registry containing all metrics.
|
com.codahale.metrics.Gauge<Integer> |
getRequestQueueDepth()
Returns the total number of enqueued requests on all Cassandra hosts.
|
com.codahale.metrics.Timer |
getRequestsTimer()
Returns metrics on the user requests performed on the Cluster.
|
com.codahale.metrics.Gauge<Integer> |
getTaskSchedulerQueueSize()
Returns the number of queued up tasks in the
scheduled tasks executor . |
com.codahale.metrics.Gauge<Integer> |
getTrashedConnections()
Returns the total number of currently "trashed" connections to Cassandra hosts.
|
public com.codahale.metrics.MetricRegistry getRegistry()
The metrics registry allows you to easily use the reporters that ship with Metrics or a custom written one.
For instance, if metrics
is this
object, you could export the metrics to csv
files using:
com.codahale.metrics.CsvReporter.forRegistry(metrics.getRegistry()).build(new File("measurements/")).start(1, TimeUnit.SECONDS);
If you already have a MetricRegistry
in your application and wish to add the
driver's metrics to it, the recommended approach is to use a listener:
// Your existing registry: final com.codahale.metrics.MetricRegistry myRegistry = ... cluster.getMetrics().getRegistry().addListener(new com.codahale.metrics.MetricRegistryListener() { @Override public void onGaugeAdded(String name, Gauge<?> gauge) { if (myRegistry.getNames().contains(name)) { // name is already taken, maybe prefix with a namespace ... } else { myRegistry.register(name, gauge); } } ... // Implement other methods in a similar fashion });Since reporting is handled by your registry, you'll probably also want to disable JMX reporting with
Cluster.Builder.withoutJMXReporting()
.public com.codahale.metrics.Timer getRequestsTimer()
This metric exposes
Timer
metric object exposing the rate and latency for user requests.public Metrics.Errors getErrorMetrics()
public com.codahale.metrics.Gauge<Integer> getKnownHosts()
public com.codahale.metrics.Gauge<Integer> getConnectedToHosts()
public com.codahale.metrics.Gauge<Integer> getOpenConnections()
public com.codahale.metrics.Gauge<Integer> getTrashedConnections()
When the load to a host decreases, the driver will reclaim some connections in order to save
resources. No requests are sent to these connections anymore, but they are kept open for an
additional amount of time (PoolingOptions.getIdleTimeoutSeconds()
), in case the load
goes up again. This metric counts connections in that state.
public com.codahale.metrics.Gauge<Integer> getInFlightRequests()
public com.codahale.metrics.Gauge<Integer> getRequestQueueDepth()
Session.State#getRequestQueueDepth(Host)
public com.codahale.metrics.Gauge<Integer> getExecutorQueueDepth()
main internal executor
.
If the executor's task queue is not accessible – which happens when the executor is not an
instance of ThreadPoolExecutor
– then this gauge returns -1.
public com.codahale.metrics.Gauge<Integer> getBlockingExecutorQueueDepth()
blocking executor
.
If the executor's task queue is not accessible – which happens when the executor is not an
instance of ThreadPoolExecutor
– then this gauge returns -1.
public com.codahale.metrics.Gauge<Integer> getReconnectionSchedulerQueueSize()
reconnection executor
.
A queue size > 0 does not necessarily indicate a backlog as some tasks may not have been scheduled to execute yet.
If the executor's task queue is not accessible – which happens when the executor is not an
instance of ThreadPoolExecutor
– then this gauge returns -1.
public com.codahale.metrics.Gauge<Integer> getTaskSchedulerQueueSize()
scheduled tasks executor
.
A queue size > 0 does not necessarily indicate a backlog as some tasks may not have been scheduled to execute yet.
If the executor's task queue is not accessible – which happens when the executor is not an
instance of ThreadPoolExecutor
– then this gauge returns -1.
public com.codahale.metrics.Meter getBytesSent()
Note that this measures unencrypted traffic, even if SSL is enabled (the probe is inserted before SSL handlers in the Netty pipeline). In practice, SSL overhead should be negligible after the initial handshake.
public com.codahale.metrics.Meter getBytesReceived()
Note that this measures unencrypted traffic, even if SSL is enabled (the probe is inserted before SSL handlers in the Netty pipeline). In practice, SSL overhead should be negligible after the initial handshake.
Copyright © 2012–2019. All rights reserved.