flink-issues mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From tillrohrmann <...@git.apache.org>
Subject [GitHub] flink pull request #2158: [FLINK-4116] Metrics documentation
Date Mon, 27 Jun 2016 11:58:40 GMT
Github user tillrohrmann commented on a diff in the pull request:

    --- Diff: docs/apis/common/index.md ---
    @@ -1350,3 +1350,211 @@ You may specify program arguments before the job is executed.
The plan visualiza
     the execution plan before executing the Flink job.
     {% top %}
    +Flink exposes a metric system that allows users gather and expose metrics to external
    +### Registering metrics
    +You can access the metric system from any user function that extends [RichFunction]({{
site.baseurl }}/apis/common/index.html#rich-functions) by calling getRuntimeContext().getMetricGroup().
    +This method returns a MetricGroup object on which you can create and register new metrics.
    +If you want to count the number of records your user function has received you could
use a Counter like this:
    +{% highlight java %}
    +public class MyMapper extends RichMapFunction<String, Integer> {
    +  private Counter counter;
    +  @Override
    +  public void open(Configuration config) {
    +    // create and register a counter
    +    this.counter = getRuntimeContext().getMetricGroup().counter("myCounter");
    +    ...
    +  }
    +  @public Integer map(String value) throws Exception {
    +    // increment counter
    +    this.counter.inc();
    +    ...
    +  }
    +{% endhighlight %}
    +### Metric types
    +Flink supports Counters, Gauges and Histograms.
    +A Counter is used to count something. The current value can be in- or decremented using
`inc([long n])` or `dec([long n])`.
    +You can create and registers a Counter by calling `counter(String name)` on a MetricGroup.
Alternatively, you can pass an instance of your own Counter implementation along with the
    +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 not restriction for the type of the returned value.
    +You can register a gauge by calling `gauge(String name, Gauge gauge)` on a MetricGroup.
    +A Histogram measure the distribution of long values.
    +You can register one by calling histogram(String name, Histogram histogram) on a MetricGroup.
    +Flink only provides an interface for Histograms, but offers a Wrapper that allows usage
of Codahale/DropWizard Histograms. (org.apache.flink.dropwizard.metrics.DropWizardHistogramWrapper)
    +### Scope
    +Every registered metric has an automatically assigned scope which represent the entities
it is tied to. By default a metric that is registered in a user function will be scoped to
the Operator in which the function runs, the Task/Job it belongs to and the TaskManager/Host
it is executed on. This is referred to as the "system scope".
    +You can define an additonal "user scope" by calling the MetricGroup#addGroup((int/String)
name) method.
    +{% highlight java %}
    +counter2 = getRuntimeContext().getMetricGroup().addGroup("MyMetrics").counter("myCounter2");
    +{% endhighlight %}
    +The name under which a metric is exported is based on both scopes and the name passed
in the `counter() call. The order is always \<system_scope>\<user_scope>\<name>.
    +The system scope allows the reported name to contain contextual information like the
name of job it was registered in without requiring the user to pass this information manually.
    +How the system scope affects the reported name for a metric can be modified by setting
the following keys in the flink-conf.yaml. 
    +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
    +- `metrics.scope.jm.job`
    +  - Default: \<host>.jobmanager.\<job_name>
    +- `metrics.scope.tm`
    +  - Default: \<host>.taskmanager.\<tm_id>
    +- `metrics.scope.tm.job`
    +  - Default: \<host>.taskmanager.\<tm_id>.\<job_name>
    +- `metrics.scope.task`
    +  - Default: \<host>.taskmanager.\<tm_id>.\<job_name>.\<task_name>.\<subtask_index>
    +- `metrics.scope.operator`
    +  - Default: \<host>.taskmanager.\<tm_id>.\<job_name>.\<operator_name>.\<subtask_index>
    +The following is a list of all variables available:
    +- JobManager: \<host>
    +- TaskManager: \<host>, \<tm_id>
    +- Job: \<job_id>, \<job_name>
    +- Task: \<task_id>, \<task_name>, \<task_attempt_id>, \<task_attempt_num>,
    +- Operator: \<operator_name>, \<subtask_index>
    +There is no restriction on the order of variables.
    +Which format is applied to a metric depends on which entities it was scoped to.
    +The `metrics.scope.operator` format will be applied to all metrics that were scoped to
an Operator.
    +The `metrics.scope.tm.job` format will be applied to all metrics that were only scoped
to a Job and a TaskManager.
    +This allows you to define explicitly how metrics on every level should be reported.
    +### Reporter
    +Metrics can be exposed to an external system by configuring a reporter in the `conf/flink-conf.yaml`.
    +- `metrics.reporter.class`: The class of the reporter to use.
    +  - Example: org.apache.flink.metrics.reporter.JMXReporter)
    +- `metrics.reporter.arguments`: A list of named parameters that are passed to the reporter.
    +  - Example: --host localhost --port 9010)
    +- `metrics.reporter.interval`: The interval between reports.
    +  - Example: 10 SECONDS)
    +By default Flink uses JMX to expose metrics.
    +The port for JMX can be configured by setting the `metrics.jmx.port` key. This parameter
expects either a single port
    +or a port range, with the default being 9010-9025. Which port is used in the end is shown
in the relevant Job-/TaskManager log.
    +All non-JMXReporters are not part of the distribution and have to be added to the classpath
manually. (usually by putting the jar into /lib)
    +You can write your own Reporter by implementing the `org.apache.flink.metrics.reporter.MetricReporter`
    +If the Reporter should send out reports regularly you have to implement the Scheduled
interface as well.
    +Flink supports the following systems with their respective configuration parameters:
    +#### Ganglia (org.apcahe.flink.metrics.ganglia.GangliaReporter)
    +- `host`
    +- `port`
    +- `dmax`
    +- `tmax`
    +- `ttl`
    +- `addressingMode` (UNICAST/MULTICAST)
    --- End diff --
    Which flink module do I have to add for this reporter? Maybe give code snippet which you
can easily add to your `pom.xml` file.

If your project is set up for it, you can reply to this email and have your
reply appear on GitHub as well. If your project does not have this feature
enabled and wishes so, or if the feature is enabled but not working, please
contact infrastructure at infrastructure@apache.org or file a JIRA ticket
with INFRA.

View raw message