指标

Flink exposes a metric system that allows gathering and exposing metrics to external systems.

Registering metrics

You can access the metric system from any user function that extends [RichFunction]({{< ref "docs/dev/datastream/user_defined_functions" >}}#rich-functions) by calling getRuntimeContext().getMetricGroup(). This method returns a MetricGroup object on which you can create and register new metrics.

Metric types

Flink supports Counters, Gauges, Histograms and Meters.

Counter

A Counter is used to count something. The current value can be in- or decremented using inc()/inc(long n) or dec()/dec(long n). You can create and register a Counter by calling counter(String name) on a MetricGroup.

java


public class MyMapper extends RichMapFunction<String, String> {
  private transient Counter counter;

  @Override
  public void open(OpenContext ctx) {
    this.counter = getRuntimeContext()
      .getMetricGroup()
      .counter("myCounter");
  }

  @Override
  public String map(String value) throws Exception {
    this.counter.inc();
    return value;
  }
}

python


class MyMapper(MapFunction):
    def __init__(self):
        self.counter = None

    def open(self, runtime_context: RuntimeContext):
        self.counter = runtime_context \
            .get_metrics_group() \
            .counter("my_counter")

    def map(self, value: str):
        self.counter.inc()
        return value

Alternatively you can also use your own Counter implementation:

java


public class MyMapper extends RichMapFunction<String, String> {
  private transient Counter counter;

  @Override
  public void open(OpenContext ctx) {
    this.counter = getRuntimeContext()
      .getMetricGroup()
      .counter("myCustomCounter", new CustomCounter());
  }

  @Override
  public String map(String value) throws Exception {
    this.counter.inc();
    return value;
  }
}

python

Still not supported in Python API.

Gauge

A Gauge provides a value of any type on demand. In order to use a Gauge you must first create a class that implements the org.apache.flink.metrics.Gauge interface. There is no restriction for the type of the returned value. You can register a gauge by calling gauge(String name, Gauge gauge) on a MetricGroup.

java


public class MyMapper extends RichMapFunction<String, String> {
  private transient int valueToExpose = 0;

  @Override
  public void open(OpenContext ctx) {
    getRuntimeContext()
      .getMetricGroup()
      .gauge("MyGauge", new Gauge<Integer>() {
        @Override
        public Integer getValue() {
          return valueToExpose;
        }
      });
  }

  @Override
  public String map(String value) throws Exception {
    valueToExpose++;
    return value;
  }
}

python


class MyMapper(MapFunction):
    def __init__(self):
        self.value_to_expose = 0

    def open(self, runtime_context: RuntimeContext):
        runtime_context \
            .get_metrics_group() \
            .gauge("my_gauge", lambda: self.value_to_expose)

    def map(self, value: str):
        self.value_to_expose += 1
        return value

Note that reporters will turn the exposed object into a String, which means that a meaningful toString() implementation is required.

Histogram

A Histogram measures the distribution of long values. You can register one by calling histogram(String name, Histogram histogram) on a MetricGroup.

java

public class MyMapper extends RichMapFunction<Long, Long> {
  private transient Histogram histogram;

  @Override
  public void open(OpenContext ctx) {
    this.histogram = getRuntimeContext()
      .getMetricGroup()
      .histogram("myHistogram", new MyHistogram());
  }

  @Override
  public Long map(Long value) throws Exception {
    this.histogram.update(value);
    return value;
  }
}

python

Still not supported in Python API.

Flink does not provide a default implementation for Histogram, but offers a {{< gh_link file="flink-metrics/flink-metrics-dropwizard/src/main/java/org/apache/flink/dropwizard/metrics/DropwizardHistogramWrapper.java" name="Wrapper" >}} that allows usage of Codahale/DropWizard histograms. To use this wrapper add the following dependency in your pom.xml:

xml

<dependency>
      <groupId>org.apache.flink</groupId>
      <artifactId>flink-metrics-dropwizard</artifactId>
      <version>{{< version >}}</version>
</dependency>

You can then register a Codahale/DropWizard histogram like this:

java

public class MyMapper extends RichMapFunction<Long, Long> {
  private transient Histogram histogram;

  @Override
  public void open(OpenContext ctx) {
    com.codahale.metrics.Histogram dropwizardHistogram =
      new com.codahale.metrics.Histogram(new SlidingWindowReservoir(500));

    this.histogram = getRuntimeContext()
      .getMetricGroup()
      .histogram("myHistogram", new DropwizardHistogramWrapper(dropwizardHistogram));
  }
  
  @Override
  public Long map(Long value) throws Exception {
    this.histogram.update(value);
    return value;
  }
}

python

Still not supported in Python API.

Meter

A Meter measures an average throughput. An occurrence of an event can be registered with the markEvent() method. Occurrence of multiple events at the same time can be registered with markEvent(long n) method. You can register a meter by calling meter(String name, Meter meter) on a MetricGroup.

java

public class MyMapper extends RichMapFunction<Long, Long> {
  private transient Meter meter;

  @Override
  public void open(OpenContext ctx) {
    this.meter = getRuntimeContext()
      .getMetricGroup()
      .meter("myMeter", new MyMeter());
  }

  @Override
  public Long map(Long value) throws Exception {
    this.meter.markEvent();
    return value;
  }
}

python


class MyMapperMeter(MapFunction):
    def __init__(self):
        self.meter = None

    def open(self, runtime_context: RuntimeContext):
        # an average rate of events per second over 120s, default is 60s.
        self.meter = runtime_context \
            .get_metrics_group() \
            .meter("my_meter", time_span_in_seconds=120)

    def map(self, value: str):
        self.meter.mark_event()
        return value

Flink offers a {{< gh_link file="flink-metrics/flink-metrics-dropwizard/src/main/java/org/apache/flink/dropwizard/metrics/DropwizardMeterWrapper.java" name="Wrapper" >}} that allows usage of Codahale/DropWizard meters. To use this wrapper add the following dependency in your pom.xml:

xml

<dependency>
      <groupId>org.apache.flink</groupId>
      <artifactId>flink-metrics-dropwizard</artifactId>
      <version>{{< version >}}</version>
</dependency>

You can then register a Codahale/DropWizard meter like this:

java

public class MyMapper extends RichMapFunction<Long, Long> {
  private transient Meter meter;

  @Override
  public void open(OpenContext ctx) {
    com.codahale.metrics.Meter dropwizardMeter = new com.codahale.metrics.Meter();

    this.meter = getRuntimeContext()
      .getMetricGroup()
      .meter("myMeter", new DropwizardMeterWrapper(dropwizardMeter));
  }

  @Override
  public Long map(Long value) throws Exception {
    this.meter.markEvent();
    return value;
  }
}

python

Still not supported in Python API.

Scope

Every metric is assigned an identifier and a set of key-value pairs under which the metric will be reported.

The identifier is based on 3 components: a user-defined name when registering the metric, an optional user-defined scope and a system-provided scope. For example, if A.B is the system scope, C.D the user scope and E the name, then the identifier for the metric will be A.B.C.D.E.

You can configure which delimiter to use for the identifier (default: .) by setting the metrics.scope.delimiter key in [Flink configuration file]({{< ref "docs/deployment/config#flink-配置文件" >}}).

User Scope

You can define a user scope by calling MetricGroup#addGroup(String name), MetricGroup#addGroup(int name) or MetricGroup#addGroup(String key, String value). These methods affect what MetricGroup#getMetricIdentifier and MetricGroup#getScopeComponents return.

java


counter = getRuntimeContext()
  .getMetricGroup()
  .addGroup("MyMetrics")
  .counter("myCounter");

counter = getRuntimeContext()
  .getMetricGroup()
  .addGroup("MyMetricsKey", "MyMetricsValue")
  .counter("myCounter");

python


counter = runtime_context \
    .get_metric_group() \
    .add_group("my_metrics") \
    .counter("my_counter")

counter = runtime_context \
    .get_metric_group() \
    .add_group("my_metrics_key", "my_metrics_value") \
    .counter("my_counter")

System Scope

The system scope contains context information about the metric, for example in which task it was registered or what job that task belongs to.

Which context information should be included can be configured by setting the following keys in [Flink configuration file]({{< ref "docs/deployment/config#flink-配置文件" >}}). Each of these keys expect a format string that may contain constants (e.g. "taskmanager") and variables (e.g. "<task_id>") which will be replaced at runtime.

metrics.scope.jm
- Default: <host>.jobmanager
- Applied to all metrics that were scoped to a job manager.
metrics.scope.jm-job
- Default: <host>.jobmanager.<job_name>
- Applied to all metrics that were scoped to a job manager and job.
metrics.scope.tm
- Default: <host>.taskmanager.<tm_id>
- Applied to all metrics that were scoped to a task manager.
metrics.scope.tm-job
- Default: <host>.taskmanager.<tm_id>.<job_name>
- Applied to all metrics that were scoped to a task manager and job.
metrics.scope.task
- Default: <host>.taskmanager.<tm_id>.<job_name>.<task_name>.<subtask_index>
- Applied to all metrics that were scoped to a task.
metrics.scope.operator
- Default: <host>.taskmanager.<tm_id>.<job_name>.<operator_name>.<subtask_index>
- Applied to all metrics that were scoped to an operator.

There are no restrictions on the number or order of variables. Variables are case sensitive.

The default scope for operator metrics will result in an identifier akin to localhost.taskmanager.1234.MyJob.MyOperator.0.MyMetric

If you also want to include the task name but omit the task manager information you can specify the following format:

metrics.scope.operator: <host>.<job_name>.<task_name>.<operator_name>.<subtask_index>

This could create the identifier localhost.MyJob.MySource_->_MyOperator.MyOperator.0.MyMetric.

Note that for this format string an identifier clash can occur should the same job be run multiple times concurrently, which can lead to inconsistent metric data. As such it is advised to either use format strings that provide a certain degree of uniqueness by including IDs (e.g <job_id>) or by assigning unique names to jobs and operators.

List of all Variables

JobManager: <host>
TaskManager: <host>, <tm_id>
Job: <job_id>, <job_name>
Task: <task_id>, <task_name>, <task_attempt_id>, <task_attempt_num>, <subtask_index>
Operator: <operator_id>,<operator_name>, <subtask_index>

Important: For the Batch API, <operator_id> is always equal to <task_id>.

User Variables

You can define a user variable by calling MetricGroup#addGroup(String key, String value). This method affects what MetricGroup#getMetricIdentifier, MetricGroup#getScopeComponents and MetricGroup#getAllVariables() returns.

Important: User variables cannot be used in scope formats.

java


counter = getRuntimeContext()
  .getMetricGroup()
  .addGroup("MyMetricsKey", "MyMetricsValue")
  .counter("myCounter");

python

counter = runtime_context
    .get_metric_group() \
    .add_group("my_metrics_key", "my_metrics_value") \
    .counter("my_counter")

Additional Variables for operators

You can define custom variables that will be assigned to all metrics reported by a given operator using Transformation.addMetricVariable. For example:

java


fooSource =
  execEnv.fromSource(
    kafkaSource,
    getWatermarkStrategy(),
    "KafkaSource-Foo")
      .addMetricVariable("table_name", "Foo");

barSource =
  execEnv.fromSource(
    kafkaSource,
    getWatermarkStrategy(),
    "KafkaSource-Bar")
      .addMetricVariable("table_name", "Bar");

Will assign table_name variable with respective values Foo and Bar to all metrics reported by the KafkaSource, like numRecordsOut or currentOutputWatermark. If supported by your chosen metric reporter, those additional variables will be then converted to labels or tags.

Reporter

For information on how to set up Flink's metric reporters please take a look at the [metric reporters documentation]({{< ref "docs/deployment/metric_reporters" >}}).

System metrics

By default Flink gathers several metrics that provide deep insights on the current state. This section is a reference of all these metrics.

The tables below generally feature 5 columns:

The "Scope" column describes which scope format is used to generate the system scope. For example, if the cell contains "Operator" then the scope format for "metrics.scope.operator" is used. If the cell contains multiple values, separated by a slash, then the metrics are reported multiple times for different entities, like for both job- and taskmanagers.
The (optional)"Infix" column describes which infix is appended to the system scope.
The "Metrics" column lists the names of all metrics that are registered for the given scope and infix.
The "Description" column provides information as to what a given metric is measuring.
The "Type" column describes which metric type is used for the measurement.

Note that all dots in the infix/metric name columns are still subject to the "metrics.delimiter" setting.

Thus, in order to infer the metric identifier:

Take the scope-format based on the "Scope" column
Append the value in the "Infix" column if present, and account for the "metrics.delimiter" setting
Append metric name.

CPU

Memory

The memory-related metrics require Oracle's memory management (also included in OpenJDK's Hotspot implementation) to be in place. Some metrics might not be exposed when using other JVM implementations (e.g. IBM's J9).

  This value might not be necessarily equal to the maximum value specified through -Xmx or 
  the equivalent Flink configuration parameter. Some GC algorithms allocate heap memory that won't 
  be available to the user code and, therefore, not being exposed through the heap metrics.</td>
  <td>Gauge</td>
</tr>
<tr>
  <td>NonHeap.Used</td>
  <td>The amount of non-heap memory currently used (in bytes).</td>
  <td>Gauge</td>
</tr>
<tr>
  <td>NonHeap.Committed</td>
  <td>The amount of non-heap memory guaranteed to be available to the JVM (in bytes).</td>
  <td>Gauge</td>
</tr>
<tr>
  <td>NonHeap.Max</td>
  <td>The maximum amount of non-heap memory that can be used for memory management (in bytes).</td>
  <td>Gauge</td>
</tr>
<tr>
  <td>Metaspace.Used</td>
  <td>The amount of memory currently used in the Metaspace memory pool (in bytes).</td>
  <td>Gauge</td>
</tr>
<tr>
  <td>Metaspace.Committed</td>
  <td>The amount of memory guaranteed to be available to the JVM in the Metaspace memory pool (in bytes).</td>
  <td>Gauge</td>
</tr>
<tr>
  <td>Metaspace.Max</td>
  <td>The maximum amount of memory that can be used in the Metaspace memory pool (in bytes).</td>
  <td>Gauge</td>
</tr>
<tr>
  <td>Direct.Count</td>
  <td>The number of buffers in the direct buffer pool.</td>
  <td>Gauge</td>
</tr>
<tr>
  <td>Direct.MemoryUsed</td>
  <td>The amount of memory used by the JVM for the direct buffer pool (in bytes).</td>
  <td>Gauge</td>
</tr>
<tr>
  <td>Direct.TotalCapacity</td>
  <td>The total capacity of all buffers in the direct buffer pool (in bytes).</td>
  <td>Gauge</td>
</tr>
<tr>
  <td>Mapped.Count</td>
  <td>The number of buffers in the mapped buffer pool.</td>
  <td>Gauge</td>
</tr>
<tr>
  <td>Mapped.MemoryUsed</td>
  <td>The amount of memory used by the JVM for the mapped buffer pool (in bytes).</td>
  <td>Gauge</td>
</tr>
<tr>
  <td>Mapped.TotalCapacity</td>
  <td>The number of buffers in the mapped buffer pool (in bytes).</td>
  <td>Gauge</td>
</tr>
<tr>
  <td rowspan="2">Status.Flink.Memory</td>
  <td>Managed.Used</td>
  <td>The amount of managed memory currently used.</td>
  <td>Gauge</td>
</tr>
<tr>
  <td>Managed.Total</td>
  <td>The total amount of managed memory.</td>
  <td>Gauge</td>
</tr>

</tbody> </table>

File Descriptors

Threads

GarbageCollection

ClassLoader

Network

{{< hint warning >}} Deprecated: use Default shuffle service metrics {{< /hint >}}

    <strong>(only available if <tt>taskmanager.network.detailed-metrics</tt> config option is set)</strong></td>
  <td>totalQueueLen</td>
  <td>Total number of queued buffers in all input/output channels.</td>
  <td>Gauge</td>
</tr>
<tr>
  <td>minQueueLen</td>
  <td>Minimum number of queued buffers in all input/output channels.</td>
  <td>Gauge</td>
</tr>
<tr>
  <td>maxQueueLen</td>
  <td>Maximum number of queued buffers in all input/output channels.</td>
  <td>Gauge</td>
</tr>
<tr>
  <td>avgQueueLen</td>
  <td>Average number of queued buffers in all input/output channels.</td>
  <td>Gauge</td>
</tr>

</tbody> </table>

Default shuffle service

Metrics related to data exchange between task executors using netty network communication.

<table class="table table-bordered"> <thead> <tr> <th class="text-left" style="width: 18%">Scope</th> <th class="text-left" style="width: 22%">Infix</th> <th class="text-left" style="width: 22%">Metrics</th> <th class="text-left" style="width: 30%">Description</th> <th class="text-left" style="width: 8%">Type</th> </tr> </thead> <tbody> <tr> <th rowspan="7"><strong>TaskManager</strong></th> <td rowspan="7">Status.Shuffle.Netty</td> <td>AvailableMemorySegments</td> <td>The number of unused memory segments.</td> <td>Gauge</td> </tr> <tr> <td>UsedMemorySegments</td> <td>The number of used memory segments.</td> <td>Gauge</td> </tr> <tr> <td>TotalMemorySegments</td> <td>The number of allocated memory segments.</td> <td>Gauge</td> </tr> <tr> <td>AvailableMemory</td> <td>The amount of unused memory in bytes.</td> <td>Gauge</td> </tr> <tr> <td>UsedMemory</td> <td>The amount of used memory in bytes.</td> <td>Gauge</td> </tr> <tr> <td>TotalMemory</td> <td>The amount of allocated memory in bytes.</td> <td>Gauge</td> </tr> <tr> <td>RequestedMemoryUsage</td> <td>Experimental: The usage of the network memory. Shows (as percentage) the total amount of requested memory from all of the subtasks. It can exceed 100% as not all requested memory is required for subtask to make progress. However if usage exceeds 100% throughput can suffer greatly and please consider increasing available network memory, or decreasing configured size of network buffer pools.</td> <td>Gauge</td> </tr> <tr> <th rowspan="20">Task</th> <td rowspan="5">Shuffle.Netty.Input.Buffers</td> <td>inputQueueLength</td> <td>The number of queued input buffers.</td> <td>Gauge</td> </tr> <tr> <td>inputQueueSize</td> <td>The real size of queued input buffers in bytes. The size for local input channels is always `0` since the local channel takes records directly from the output queue.</td> <td>Gauge</td> </tr> <tr> <td>inPoolUsage</td> <td>An estimate of the input buffers usage. (ignores LocalInputChannels)</td> <td>Gauge</td> </tr> <tr> <td>inputFloatingBuffersUsage</td> <td>An estimate of the floating input buffers usage. (ignores LocalInputChannels)</td> <td>Gauge</td> </tr> <tr> <td>inputExclusiveBuffersUsage</td> <td>An estimate of the exclusive input buffers usage. (ignores LocalInputChannels)</td> <td>Gauge</td> </tr> <tr> <td rowspan="3">Shuffle.Netty.Output.Buffers</td> <td>outputQueueLength</td> <td>The number of queued output buffers.</td> <td>Gauge</td> </tr> <tr> <td>outputQueueSize</td> <td>The real size of queued output buffers in bytes.</td> <td>Gauge</td> </tr> <tr> <td>outPoolUsage</td> <td>An estimate of the output buffers usage. The pool usage can be > 100% if overdraft buffers are being used.</td> <td>Gauge</td> </tr> <tr> <td rowspan="4">Shuffle.Netty.<Input|Output>.<gate|partition>

    <strong>(only available if <tt>taskmanager.network.detailed-metrics</tt> config option is set)</strong></td>
  <td>totalQueueLen</td>
  <td>Total number of queued buffers in all input/output channels.</td>
  <td>Gauge</td>
</tr>
<tr>
  <td>minQueueLen</td>
  <td>Minimum number of queued buffers in all input/output channels.</td>
  <td>Gauge</td>
</tr>
<tr>
  <td>maxQueueLen</td>
  <td>Maximum number of queued buffers in all input/output channels.</td>
  <td>Gauge</td>
</tr>
<tr>
  <td>avgQueueLen</td>
  <td>Average number of queued buffers in all input/output channels.</td>
  <td>Gauge</td>
</tr>
<tr>
  <td rowspan="8">Shuffle.Netty.Input</td>
  <td>numBytesInLocal</td>
  <td>The total number of bytes this task has read from a local source.</td>
  <td>Counter</td>
</tr>
<tr>
  <td>numBytesInLocalPerSecond</td>
  <td>The number of bytes this task reads from a local source per second.</td>
  <td>Meter</td>
</tr>
<tr>
  <td>numBytesInRemote</td>
  <td>The total number of bytes this task has read from a remote source.</td>
  <td>Counter</td>
</tr>
<tr>
  <td>numBytesInRemotePerSecond</td>
  <td>The number of bytes this task reads from a remote source per second.</td>
  <td>Meter</td>
</tr>
<tr>
  <td>numBuffersInLocal</td>
  <td>The total number of network buffers this task has read from a local source.</td>
  <td>Counter</td>
</tr>
<tr>
  <td>numBuffersInLocalPerSecond</td>
  <td>The number of network buffers this task reads from a local source per second.</td>
  <td>Meter</td>
</tr>
<tr>
  <td>numBuffersInRemote</td>
  <td>The total number of network buffers this task has read from a remote source.</td>
  <td>Counter</td>
</tr>
<tr>
  <td>numBuffersInRemotePerSecond</td>
  <td>The number of network buffers this task reads from a remote source per second.</td>
  <td>Meter</td>
</tr>

</tbody> </table>

Cluster

Availability

The metrics in this table are available for each of the following job states: INITIALIZING, CREATED, RUNNING, RESTARTING, CANCELLING, FAILING. Whether these metrics are reported depends on the [metrics.job.status.enable]({{< ref "docs/deployment/config" >}}#metrics-job-status-enable) setting.

<span class="label label-info">Evolving</span> The semantics of these metrics may change in later releases.

{{< hint info >}} <span class="label label-info">Experimental</span>

While the job is in the RUNNING state the metrics in this table provide additional details on what the job is currently doing. Whether these metrics are reported depends on the [metrics.job.status.enable]({{< ref "docs/deployment/config" >}}#metrics-job-status-enable) setting.

<table class="table table-bordered table-inline"> <thead> <tr> <th class="text-left" style="width: 18%">Scope</th> <th class="text-left" style="width: 26%">Metrics</th> <th class="text-left" style="width: 48%">Description</th> <th class="text-left" style="width: 8%">Type</th> </tr> </thead> <tbody> <tr> <th rowspan="3"><strong>Job (only available on JobManager)</strong></th> <td>deployingState</td> <td>Return 1 if the job is currently deploying* tasks, otherwise return 0.</td> <td>Gauge</td> </tr> <tr> <td>deployingTime</td> <td>Return the time (in milliseconds) since the job has started deploying* tasks, otherwise return 0.</td> <td>Gauge</td> </tr> <tr> <td>deployingTimeTotal</td> <td>Return how much time (in milliseconds) the job has spent deploying* tasks in total.</td> <td>Gauge</td> </tr> </tbody> </table>

*A job is considered to be deploying tasks when:

for streaming jobs, any task is in the DEPLOYING state
for batch jobs, if at least 1 task is in the DEPLOYING state, and there are no INITIALIZING/RUNNING tasks {{< /hint >}}

{

Checkpointing

Note that for failed checkpoints, metrics are updated on a best efforts basis and may be not accurate.

State Access Latency

State Size

RocksDB

Certain RocksDB native metrics are available but disabled by default, you can find full documentation [here]({{< ref "docs/deployment/config" >}}#rocksdb-native-metrics)

ForSt

Certain ForSt native metrics are available but disabled by default, you can find full documentation [here]({{< ref "docs/deployment/config" >}}#forst-native-metrics)

Besides that, we support the following metrics:

<table class="table table-bordered"> <thead> <tr> <th class="text-left" style="width: 15%">Scope</th> <th class="text-left" style="width: 15%">Infix</th> <th class="text-left" style="width: 15%">Metrics</th> <th class="text-left" style="width: 50%">Description</th> <th class="text-left" style="width: 5%">Type</th> </tr> </thead> <tbody> <tr> <th rowspan="4"><strong>Task/Operator</strong></th> <td rowspan="4">forst.fileCache</td> <td>hit</td> <td>The hit count of ForSt state backend cache.</td> <td>Counter</td> </tr> <tr> <td>miss</td> <td>The miss count of ForSt state backend cache.</td> <td>Counter</td> </tr> <tr> <td>usedBytes</td> <td>The bytes cached in ForSt state backend cache.</td> <td>Gauge</td> </tr> <tr> <td>remainingBytes</td> <td>The remaining space in the volume for the configured cache. Only available when 'state.backend.forst.cache.reserve-size' is set above 0. </td> <td>Gauge</td> </tr> <tr> <td>lru.evict</td> <td>The number of cache files that are evicted from LRU.</td> <td>Gauge</td> </tr> <tr> <td>lru.loadback</td> <td>The number of cache files that are loaded back from remote storage into the LRU. </td> <td>Gauge</td> </tr> </tbody> </table>

State Changelog

Note that the metrics are only available via reporters.

IO

<table class="table table-bordered"> <thead> <tr> <th class="text-left" style="width: 18%">Scope</th> <th class="text-left" style="width: 26%">Metrics</th> <th class="text-left" style="width: 48%">Description</th> <th class="text-left" style="width: 8%">Type</th> </tr> </thead> <tbody> <tr> <th rowspan="1"><strong>Job (only available on TaskManager)</strong></th> <td>[<source_id>.[<source_subtask_index>.]]<operator_id>.<operator_subtask_index>.latency</td> <td>The latency distributions from a given source (subtask) to an operator subtask (in milliseconds), depending on the <a href="{{< ref "docs/deployment/config" >}}#metrics-latency-granularity">latency granularity</a>.</td> <td>Histogram</td> </tr> <tr> <th rowspan="27"><strong>Task</strong></th> <td>numBytesInLocal</td> <td><span class="label label-danger">Attention:</span> deprecated, use <a href="{{< ref "docs/ops/metrics" >}}#default-shuffle-service">Default shuffle service metrics</a>.</td> <td>Counter</td> </tr> <tr> <td>numBytesInLocalPerSecond</td> <td><span class="label label-danger">Attention:</span> deprecated, use <a href="{{< ref "docs/ops/metrics" >}}#default-shuffle-service">Default shuffle service metrics</a>.</td> <td>Meter</td> </tr> <tr> <td>numBytesInRemote</td> <td><span class="label label-danger">Attention:</span> deprecated, use <a href="{{< ref "docs/ops/metrics" >}}#default-shuffle-service">Default shuffle service metrics</a>.</td> <td>Counter</td> </tr> <tr> <td>numBytesInRemotePerSecond</td> <td><span class="label label-danger">Attention:</span> deprecated, use <a href="{{< ref "docs/ops/metrics" >}}#default-shuffle-service">Default shuffle service metrics</a>.</td> <td>Meter</td> </tr> <tr> <td>numBuffersInLocal</td> <td><span class="label label-danger">Attention:</span> deprecated, use <a href="{{< ref "docs/ops/metrics" >}}#default-shuffle-service">Default shuffle service metrics</a>.</td> <td>Counter</td> </tr> <tr> <td>numBuffersInLocalPerSecond</td> <td><span class="label label-danger">Attention:</span> deprecated, use <a href="{{< ref "docs/ops/metrics" >}}#default-shuffle-service">Default shuffle service metrics</a>.</td> <td>Meter</td> </tr> <tr> <td>numBuffersInRemote</td> <td><span class="label label-danger">Attention:</span> deprecated, use <a href="{{< ref "docs/ops/metrics" >}}#default-shuffle-service">Default shuffle service metrics</a>.</td> <td>Counter</td> </tr> <tr> <td>numBuffersInRemotePerSecond</td> <td><span class="label label-danger">Attention:</span> deprecated, use <a href="{{< ref "docs/ops/metrics" >}}#default-shuffle-service">Default shuffle service metrics</a>.</td> <td>Meter</td> </tr> <tr> <td>numBytesOut</td> <td>The total number of bytes this task has emitted.</td> <td>Counter</td> </tr> <tr> <td>numBytesOutPerSecond</td> <td>The number of bytes this task emits per second.</td> <td>Meter</td> </tr> <tr> <td>numBuffersOut</td> <td>The total number of network buffers this task has emitted.</td> <td>Counter</td> </tr> <tr> <td>numBuffersOutPerSecond</td> <td>The number of network buffers this task emits per second.</td> <td>Meter</td> </tr> <tr> <td>numFiredTimers</td> <td>The total number of timers this task has fired.</td> <td>Counter</td> </tr> <tr> <td>numFiredTimersPerSecond</td> <td>The number of timers this task fires per second.</td> <td>Meter</td> </tr> <tr> <td>isBackPressured</td> <td>Whether the task is back-pressured.</td> <td>Gauge</td> </tr> <tr> <td>idleTimeMsPerSecond</td> <td>The time (in milliseconds) this task is idle (has no data to process) per second. Idle time excludes back pressured time, so if the task is back pressured it is not idle.</td> <td>Meter</td> </tr> <tr> <td>busyTimeMsPerSecond</td> <td>The time (in milliseconds) this task is busy (neither idle nor back pressured) per second. Can be NaN, if the value could not be calculated.</td> <td>Gauge</td> </tr> <tr> <td>backPressuredTimeMsPerSecond</td> <td>The time (in milliseconds) this task is back pressured (soft or hard) per second. It's a sum of softBackPressuredTimeMsPerSecond and hardBackPressuredTimeMsPerSecond.</td> <td>Gauge</td> </tr> <tr> <td>softBackPressuredTimeMsPerSecond</td> <td>The time (in milliseconds) this task is softly back pressured per second. Softly back pressured task will be still responsive and capable of for example triggering unaligned checkpoints.</td> <td>Gauge</td> </tr> <tr> <td>hardBackPressuredTimeMsPerSecond</td> <td>The time (in milliseconds) this task is back pressured in a hard way per second. During hard back pressured task is completely blocked and unresponsive preventing for example unaligned checkpoints from triggering.</td> <td>Gauge</td> </tr> <tr> <td>maxSoftBackPressuredTimeMs</td> <td>Maximum recorded duration of a single consecutive period of the task being softly back pressured in the last sampling period. Please check softBackPressuredTimeMsPerSecond and hardBackPressuredTimeMsPerSecond for more information.</td> <td>Gauge</td> </tr> <tr> <td>maxHardBackPressuredTimeMs</td> <td>Maximum recorded duration of a single consecutive period of the task being in the hard back pressure state in the last sampling period. Please check softBackPressuredTimeMsPerSecond and hardBackPressuredTimeMsPerSecond for more information.</td> <td>Gauge</td> </tr> <tr> <td>changelogBusyTimeMsPerSecond</td> <td>The time (in milliseconds) taken by the Changelog state backend to do IO operations, only positive when Changelog state backend is enabled. Please check 'state.changelog.dstl.dfs.upload.max-in-flight' for more information.</td> <td>Gauge</td> </tr> <tr> <td>mailboxMailsPerSecond</td> <td>The number of actions processed from the task's mailbox per second which includes all actions, e.g., checkpointing, timer, or cancellation actions.</td> <td>Meter</td> </tr> <tr> <td>mailboxLatencyMs</td> <td>The latency is the time that actions spend waiting in the task's mailbox before being processed. The metric is a statistic of the latency in milliseconds that is measured approximately once every second and includes the last 60 measurements.</td> <td>Histogram</td> </tr> <tr> <td>mailboxQueueSize</td> <td>The number of actions in the task's mailbox that are waiting to be processed.</td> <td>Gauge</td> </tr> <tr> <td>initializationTime</td> <td>The time in milliseconds that one task spends on initialization, return 0 when the task is not in initialization/running status. Most of the initialization time is usually spent in restoring from the checkpoint.</td> <td>Counter</td> </tr> <tr> <td rowspan="2"><strong>Task (only if buffer debloating is enabled and in non-source tasks)</strong></td> <td>estimatedTimeToConsumeBuffersMs</td> <td>The estimated time (in milliseconds) by the buffer debloater to consume all of the buffered data in the network exchange preceding this task. This value is calculated by approximated amount of the in-flight data and calculated throughput.</td> <td>Gauge</td> </tr> <tr> <td>debloatedBufferSize</td> <td>The desired buffer size (in bytes) calculated by the buffer debloater. Buffer debloater is trying to reduce buffer size when the amount of in-flight data (after taking into account current throughput) exceeds the configured target value.</td> <td>Gauge</td> </tr> <tr> <th rowspan="6"><strong>Task/Operator</strong></th> <td>numRecordsIn</td> <td>The total number of records this operator/task has received.</td> <td>Counter</td> </tr> <tr> <td>numRecordsInPerSecond</td> <td>The number of records this operator/task receives per second.</td> <td>Meter</td> </tr> <tr> <td>numRecordsOut</td> <td>The total number of records this operator/task has emitted.</td> <td>Counter</td> </tr> <tr> <td>numRecordsOutPerSecond</td> <td>The number of records this operator/task sends per second.</td> <td>Meter</td> </tr> <tr> <td>numLateRecordsDropped</td> <td>The number of records this operator/task has dropped due to arriving late.</td> <td>Counter</td> </tr> <tr> <td>currentInputWatermark</td> <td> The last watermark this operator/tasks has received (in milliseconds). <p><strong>Note:</strong> For operators/tasks with 2 inputs this is the minimum of the last received watermarks.</p> </td> <td>Gauge</td> </tr> <tr> <th rowspan="4"><strong>Operator</strong></th> <td>currentInput<strong>N</strong>Watermark</td> <td> The last watermark this operator has received in its <strong>N'th</strong> input (in milliseconds), with index <strong>N</strong> starting from 1. For example currentInput<strong>1</strong>Watermark, currentInput<strong>2</strong>Watermark, ... <p><strong>Note:</strong> Only for operators with 2 or more inputs.</p> </td> <td>Gauge</td> </tr> <tr> <td>currentOutputWatermark</td> <td> The last watermark this operator has emitted (in milliseconds). </td> <td>Gauge</td> </tr> <tr> <td>watermarkAlignmentDrift</td> <td> The current drift from the minimal watermark emitted by all sources/tasks/splits that belong to the same watermark group. <p><strong>Note:</strong> Available only when watermark alignment is enabled and the first common watermark is announced. You can configure the update interval in the WatermarkStrategy.</p> </td> <td>Gauge</td> </tr> <tr> <td>numSplitsProcessed</td> <td>The total number of InputSplits this data source has processed (if the operator is a data source).</td> <td>Gauge</td> </tr> </tbody> </table>

Connectors

Kafka Connectors

<table class="table table-bordered"> <thead> <tr> <th class="text-left" style="width: 15%">Scope</th> <th class="text-left" style="width: 18%">Metrics</th> <th class="text-left" style="width: 18%">User Variables</th> <th class="text-left" style="width: 39%">Description</th> <th class="text-left" style="width: 10%">Type</th> </tr> </thead> <tbody> <tr> <th rowspan="1">Operator</th> <td>commitsSucceeded</td> <td>n/a</td> <td>The total number of successful offset commits to Kafka, if offset committing is turned on and checkpointing is enabled.</td> <td>Counter</td> </tr> <tr> <th rowspan="1">Operator</th> <td>commitsFailed</td> <td>n/a</td> <td>The total number of offset commit failures to Kafka, if offset committing is turned on and checkpointing is enabled. Note that committing offsets back to Kafka is only a means to expose consumer progress, so a commit failure does not affect the integrity of Flink's checkpointed partition offsets.</td> <td>Counter</td> </tr> <tr> <th rowspan="1">Operator</th> <td>committedOffsets</td> <td>topic, partition</td> <td>The last successfully committed offsets to Kafka, for each partition. A particular partition's metric can be specified by topic name and partition id.</td> <td>Gauge</td> </tr> <tr> <th rowspan="1">Operator</th> <td>currentOffsets</td> <td>topic, partition</td> <td>The consumer's current read offset, for each partition. A particular partition's metric can be specified by topic name and partition id.</td> <td>Gauge</td> </tr> </tbody> </table>

Kinesis 源

<table class="table table-bordered"> <thead> <tr> <th class="text-left" style="width: 15%">范围</th> <th class="text-left" style="width: 18%">指标</th> <th class="text-left" style="width: 18%">用户变量</th> <th class="text-left" style="width: 39%">描述</th> <th class="text-left" style="width: 10%">类型</th> </tr> </thead> <tbody> <tr> <th rowspan="1">Operator</th> <td>millisBehindLatest</td> <td>stream, shardId</td> <td>消费者落后于流头部的毫秒数，对每个Kinesis分片，表示费者落后当前时间多久。可以通过流名称和分片id指定一个特定分片的指标值。值为0表示记录处理已完成，并且没有新记录在此时处理。值为-1表示尚未报告指标值。 </td> <td>Gauge</td> </tr> <tr> <th rowspan="1">Operator</th> <td>sleepTimeMillis</td> <td>stream, shardId</td> <td>消费者在从Kinesis获取记录之前睡眠的毫秒数。可以通过流名称和分片id指定特定分片的指标值。 </td> <td>Gauge</td> </tr> <tr> <th rowspan="1">Operator</th> <td>maxNumberOfRecordsPerFetch</td> <td>stream, shardId</td> <td>消费者在对Kinesis的单个getRecords调用中请求的最大记录数。如果ConsumerConfigConstants.SHARD_USE_ADAPTIVE_READS 设置为true，自适应计算该值，以最大化来自Kinesis的2Mbps读取限制。 </td> <td>Gauge</td> </tr> <tr> <th rowspan="1">Operator</th> <td>numberOfAggregatedRecordsPerFetch</td> <td>stream, shardId</td> <td>消费者在对Kinesis的单个getRecords调用中获取的聚合的Kinesis记录数。 </td> <td>Gauge</td> </tr> <tr> <th rowspan="1">Operator</th> <td>numberOfDeggregatedRecordsPerFetch</td> <td>stream, shardId</td> <td>消费者在对Kinesis的单个getRecords调用中获取的非聚合的Kinesis记录数。 </td> <td>Gauge</td> </tr> <tr> <th rowspan="1">Operator</th> <td>averageRecordSizeBytes</td> <td>stream, shardId</td> <td>以字节为单位的Kinesis记录的平均大小，由消费者在单个getRecords调用中获取。 </td> <td>Gauge</td> </tr> <tr> <th rowspan="1">Operator</th> <td>runLoopTimeNanos</td> <td>stream, shardId</td> <td>消费者在运行循环中花费的实际时间（纳秒）。 </td> <td>Gauge</td> </tr> <tr> <th rowspan="1">Operator</th> <td>loopFrequencyHz</td> <td>stream, shardId</td> <td>一秒钟内调用getRecords的次数。 </td> <td>Gauge</td> </tr> <tr> <th rowspan="1">Operator</th> <td>bytesRequestedPerFetch</td> <td>stream, shardId</td> <td>在对getRecords的单个调用中请求的字节数（2 Mbps / loopFrequencyHz）。 </td> <td>Gauge</td> </tr> </tbody> </table>

Kinesis 接收器

HBase Connectors

System resources

System resources reporting is disabled by default. When metrics.system-resource is enabled additional metrics listed below will be available on Job- and TaskManager. System resources metrics are updated periodically and they present average values for a configured interval (metrics.system-resource-probing-interval).

System resources reporting requires an optional dependency to be present on the classpath (for example placed in Flink's lib directory):

com.github.oshi:oshi-core:6.1.5 (licensed under MIT license)

Including it's transitive dependencies:

net.java.dev.jna:jna-platform:jar:5.10.0
net.java.dev.jna:jna:jar:5.10.0

Failures in this regard will be reported as warning messages like NoClassDefFoundError logged by SystemResourcesMetricsInitializer during the startup.

System CPU

<table class="table table-bordered"> <thead> <tr> <th class="text-left" style="width: 20%">Scope</th> <th class="text-left" style="width: 25%">Infix</th> <th class="text-left" style="width: 23%">Metrics</th> <th class="text-left" style="width: 32%">Description</th> </tr> </thead> <tbody> <tr> <th rowspan="12"><strong>Job-/TaskManager</strong></th> <td rowspan="12">System.CPU</td> <td>Usage</td> <td>Overall % of CPU usage on the machine.</td> </tr> <tr> <td>Idle</td> <td>% of CPU Idle usage on the machine.</td> </tr> <tr> <td>Sys</td> <td>% of System CPU usage on the machine.</td> </tr> <tr> <td>User</td> <td>% of User CPU usage on the machine.</td> </tr> <tr> <td>IOWait</td> <td>% of IOWait CPU usage on the machine.</td> </tr> <tr> <td>Irq</td> <td>% of Irq CPU usage on the machine.</td> </tr> <tr> <td>SoftIrq</td> <td>% of SoftIrq CPU usage on the machine.</td> </tr> <tr> <td>Nice</td> <td>% of Nice Idle usage on the machine.</td> </tr> <tr> <td>Load1min</td> <td>Average CPU load over 1 minute</td> </tr> <tr> <td>Load5min</td> <td>Average CPU load over 5 minute</td> </tr> <tr> <td>Load15min</td> <td>Average CPU load over 15 minute</td> </tr> <tr> <td>UsageCPU*</td> <td>% of CPU usage per each processor</td> </tr> </tbody> </table>

System memory

System network

预测执行

以下指标可以用来衡量预测执行的有效性。

Async State Processing

<table class="table table-bordered"> <thead> <tr> <th class="text-left" style="width: 15%">Scope</th> <th class="text-left" style="width: 10%">Infix</th> <th class="text-left" style="width: 20%">Metrics</th> <th class="text-left" style="width: 50%">Description</th> <th class="text-left" style="width: 5%">Type</th> </tr> </thead> <tbody> <tr> <th rowspan="4"><strong>Operator</strong></th> <td rowspan="4">asyncStateProcessing</td> <td>numInFlightRecords</td> <td>The number of in-flight records in the async execution controller's buffers.</td> <td>Gauge</td> </tr> <tr> <td>activeBufferSize</td> <td>The number of records which are pending to be processed.</td> <td>Gauge</td> </tr> <tr> <td>blockingBufferSize</td> <td>The number of records which are blocked by the ongoing records.</td> <td>Gauge</td> </tr> <tr> <td>numBlockingKeys</td> <td>The number of different keys are blocked in async execution controller.</td> <td>Gauge</td> </tr> </tbody> </table>

End-to-End latency tracking

Flink allows to track the latency of records travelling through the system. This feature is disabled by default. To enable the latency tracking you must set the latencyTrackingInterval to a positive number in either the [Flink configuration]({{< ref "docs/deployment/config" >}}#metrics-latency-interval) or ExecutionConfig.

At the latencyTrackingInterval, the sources will periodically emit a special record, called a LatencyMarker. The marker contains a timestamp from the time when the record has been emitted at the sources. Latency markers can not overtake regular user records, thus if records are queuing up in front of an operator, it will add to the latency tracked by the marker.

Note that the latency markers are not accounting for the time user records spend in operators as they are bypassing them. In particular the markers are not accounting for the time records spend for example in window buffers. Only if operators are not able to accept new records, thus they are queuing up, the latency measured using the markers will reflect that.

The LatencyMarkers are used to derive a distribution of the latency between the sources of the topology and each downstream operator. These distributions are reported as histogram metrics. The granularity of these distributions can be controlled in the [Flink configuration]({{< ref "docs/deployment/config" >}}#metrics-latency-interval). For the highest granularity subtask Flink will derive the latency distribution between every source subtask and every downstream subtask, which results in quadratic (in the terms of the parallelism) number of histograms.

Currently, Flink assumes that the clocks of all machines in the cluster are in sync. We recommend setting up an automated clock synchronisation service (like NTP) to avoid false latency results.

<span class="label label-danger">Warning</span> Enabling latency metrics can significantly impact the performance of the cluster (in particular for subtask granularity). It is highly recommended to only use them for debugging purposes.

State access latency tracking

Flink also allows to track the keyed state access latency for standard Flink state-backends or customized state backends which extending from AbstractStateBackend. This feature is disabled by default. To enable this feature you must set the state.latency-track.keyed-state-enabled to true in the [Flink configuration]({{< ref "docs/deployment/config" >}}#state-backends-latency-tracking-options).

Once tracking keyed state access latency is enabled, Flink will sample the state access latency every N access, in which N is defined by state.latency-track.sample-interval. This configuration has a default value of 100. A smaller value will get more accurate results but have a higher performance impact since it is sampled more frequently.

As the type of this latency metrics is histogram, state.latency-track.history-size will control the maximum number of recorded values in history, which has the default value of 128. A larger value of this configuration will require more memory, but will provide a more accurate result.

<span class="label label-danger">Warning</span> Enabling state-access-latency metrics may impact the performance. It is recommended to only use them for debugging purposes.

State key/value size tracking

Flink also allows to track the keyed state key/value size for standard Flink state-backends or customized state backends which extending from AbstractStateBackend. This feature is disabled by default. To enable this feature you must set the state.size-track.keyed-state-enabled to true in the [Flink configuration]({{< ref "docs/deployment/config" >}}#state-backends-size-tracking-options).

Once tracking keyed state key/value size is enabled, Flink will sample the state size every N access, in which N is defined by state.size-track.sample-interval. This configuration has a default value of 100. A smaller value will get more accurate results but have a higher performance impact since it is sampled more frequently.

As the type of this key/value size metrics is histogram, state.size-track.history-size will control the maximum number of recorded values in history, which has the default value of 128. A larger value of this configuration will require more memory, but will provide a more accurate result.

<span class="label label-danger">Warning</span> Enabling state-size metrics may impact the performance. It is recommended to only use them for debugging purposes. If state.ttl is enabled, the size of the value will include the size of the TTL-related timestamp. The value size of AggregatingState is not accounted for because AggregatingState returns a result processed by a user-defined AggregateFunction, whereas currently, only the actual stored data size in the state can be tracked.

REST API integration

Metrics can be queried through the [Monitoring REST API]({{< ref "docs/ops/rest_api" >}}).

Below is a list of available endpoints, with a sample JSON response. All endpoints are of the sample form http://hostname:8081/jobmanager/metrics, below we list only the path part of the URLs.

Values in angle brackets are variables, for example http://hostname:8081/jobs/<jobid>/metrics will have to be requested for example as http://hostname:8081/jobs/7684be6004e4e955c2a558a9bc463f65/metrics.

Request metrics for a specific entity:

/jobmanager/metrics
/taskmanagers/<taskmanagerid>/metrics
/jobs/<jobid>/metrics
/jobs/<jobid>/vertices/<vertexid>/subtasks/<subtaskindex>

Request metrics aggregated across all entities of the respective type:

/taskmanagers/metrics
/jobs/metrics
/jobs/<jobid>/vertices/<vertexid>/subtasks/metrics
/jobs/<jobid>/vertices/<vertexid>/jm-operator-metrics

Request metrics aggregated over a subset of all entities of the respective type:

/taskmanagers/metrics?taskmanagers=A,B,C
/jobs/metrics?jobs=D,E,F
/jobs/<jobid>/vertices/<vertexid>/subtasks/metrics?subtask=1,2,3

<span class="label label-danger">Warning</span> Metric names can contain special characters that you need to escape when querying metrics. For example, "a_+_b" would be escaped to "a_%2B_b".

List of characters that should be escaped:

<table class="table table-bordered"> <thead> <tr> <th>Character</th> <th>Escape Sequence</th> </tr> </thead> <tbody> <tr> <td>#</td> <td>%23</td> </tr> <tr> <td>$</td> <td>%24</td> </tr> <tr> <td>&</td> <td>%26</td> </tr> <tr> <td>+</td> <td>%2B</td> </tr> <tr> <td>/</td> <td>%2F</td> </tr> <tr> <td>;</td> <td>%3B</td> </tr> <tr> <td>=</td> <td>%3D</td> </tr> <tr> <td>?</td> <td>%3F</td> </tr> <tr> <td>@</td> <td>%40</td> </tr> </tbody> </table>

Request a list of available metrics:

GET /jobmanager/metrics

json

[
  {
    "id": "metric1"
  },
  {
    "id": "metric2"
  }
]

Request the values for specific (unaggregated) metrics:

GET taskmanagers/ABCDE/metrics?get=metric1,metric2

json

[
  {
    "id": "metric1",
    "value": "34"
  },
  {
    "id": "metric2",
    "value": "2"
  }
]

Request aggregated values for specific metrics:

GET /taskmanagers/metrics?get=metric1,metric2

json

[
  {
    "id": "metric1",
    "min": 1,
    "max": 34,
    "avg": 15,
    "sum": 45
  },
  {
    "id": "metric2",
    "min": 2,
    "max": 14,
    "avg": 7,
    "sum": 16
  }
]

Request specific aggregated values for specific metrics:

GET /taskmanagers/metrics?get=metric1,metric2&agg=min,max

json

[
  {
    "id": "metric1",
    "min": 1,
    "max": 34
  },
  {
    "id": "metric2",
    "min": 2,
    "max": 14
  }
]

Dashboard integration

Metrics that were gathered for each task or operator can also be visualized in the Dashboard. On the main page for a job, select the Metrics tab. After selecting one of the tasks in the top graph you can select metrics to display using the Add Metric drop-down menu.

Task metrics are listed as <subtask_index>.<metric_name>.
Operator metrics are listed as <subtask_index>.<operator_name>.<metric_name>.

Each metric will be visualized as a separate graph, with the x-axis representing time and the y-axis the measured value. All graphs are automatically updated every 10 seconds, and continue to do so when navigating to another page.

There is no limit as to the number of visualized metrics; however only numeric metrics can be visualized.