Skip to content

shuhui-liu/tehuti

 
 

Repository files navigation

tehuti

Download

Build Status

Tehuti is a simple metrics library providing statistical measurement, reporting and quota functionalities.

Please subscribe to the mailing list to ask questions or discuss development.

Build

$ git clone git@github.com:tehuti-io/tehuti.git
$ cd tehuti
$ ./gradlew build

Usage

At its core, Tehuti has the concept of sensors, for recording data points, and metrics, for performing statistical measurements over those data points.

Here is an example usage:

// set up metrics:
MetricsRepository metrics = new MetricsRepository(); // this is the global repository of metrics and sensors
Sensor sensor = metrics.sensor("message-sizes");
sensor.add("message-sizes.avg", new Avg());
sensor.add("message-sizes.count", new Count());
sensor.add("message-sizes.max", new Max());
sensor.add("message-sizes.qps-throughput", new OccurrenceRate());
sensor.add("message-sizes.bytes-throughput", new Rate());
sensor.add("message-sizes.total", new Total());
sensor.add(new Percentiles(10 * 1024, // Histogram's memory consumption: max 10 KB
                           1024 * 1024, // Histogram's max value: 1 MB (i.e.: max expected message size)
                           Percentiles.BucketSizing.CONSTANT,
                           new Percentile("message-sizes.median", 50),
                           new Percentile("message-sizes.95thPercentile", 95),
                           new Percentile("message-sizes.99thPercentile", 99));

// as messages are sent we record the sizes
sensor.record(messageSize);

Note: you can also use metrics directly without sensors, although typically sensors are useful for the following reasons:

  • Sensors offer a convenient way to make sure that several metrics are measured off of the same data point.
  • Sensors are thread-safe. The synchronization happens at the sensor level, not the metric level.

Configuration

One can define a MetricConfig and apply it at the sensor level, or for specific metrics. This configuration defines how sampling and quota works.

Here is an example of custom configuration for a sensor and a metric:

// Initialize config so that sample windows roll over after 10 seconds or 1000 events, whichever comes first
MetricConfig sensorConfig = new MetricConfig().timeWindow(10, TimeUnit.SECONDS).eventWindow(1000);
 
// Initialize config so that there are 10 sample windows of 6000 milliseconds each
MetricConfig specificConfig = new MetricConfig().samples(10).timeWindow(6000, TimeUnit.MILLISECONDS);
 
// Use the first config for a sensor, and override one of the metrics within that sensor with the second config
MetricsRepository metrics = new MetricsRepository(); // this is the global repository of metrics and sensors
Sensor sensor = metrics.sensor("message-sizes", sensorConfig);
sensor.add("message-sizes.min", new Min());
sensor.add("message-sizes.max", new Max());
sensor.add("message-sizes.avg", new Avg(), specificConfig);
 
// as messages are sent we record the sizes
sensor.record(messageSize);

Here are the configuration properties and their defaults:

  • samples: The amount of sample windows to keep (and measure on). Once a sampled stat goes over this amount of samples, the oldest one will be purged and replaced by the new one. The default is 2 sample windows (the current and previous ones).
  • eventWindow: The amount of events that can come into a single window before we roll forward to the next one. The default is Long.MAX_VALUE.
  • timeWindow: The maximum time that can go by since the first event recorded in a window before we roll forward to the next one. The default is 30 seconds.
  • quota: A Quota instance defining an upper or lower bound that must be respected for a certain metric. The default is null (no quota).

Once a sampled stat goes over its maximum amount of samples, the oldest sample will be replaced by the new one. When measuring the value of a sampled stat, any sample whose beginning time is older than samples * timeWindow will be purged, so that the measurement is not affected by old data points.

Quotas

Tehuti also supports quotas. Quotas are passed within configurations, and can define a lower or upper bound for a specific metric. If the boundary is crossed, Tehuti will throw a QuotaViolationException. Applications can then catch (or bubble up) this exception to avoid doing an operation that would go beyond a desired threshold.

Here is an example on how to define quotas for queries per second and bytes per second off of the same sensor:

// Initialize config so that we bound operations to 1000 QPS and 100 MBps
MetricConfig queriesPerSecondQuota = new MetricConfig().quota(Quota.lessThan(1000))
MetricConfig bytesPerSecondQuota = new MetricConfig().quota(Quota.lessThan(100 * 1024 * 1024))
 
MetricsRepository metrics = new MetricsRepository(); // this is the global repository of metrics and sensors
Sensor sensor = metrics.sensor("messages");
sensor.add("message-sizes.min", new Min());
sensor.add("message-sizes.max", new Max());
sensor.add("message-sizes.avg", new Avg());
sensor.add("message-sizes.count", new Count());
sensor.add("message-sizes.qps-throughput", new OccurrenceRate(), queriesPerSecondQuota);
sensor.add("message-sizes.bytes-throughput", new Rate(), bytesPerSecondQuota);
 
// as messages are sent we record the sizes
sensor.record(messageSize);

Reporting

Finally, Tehuti can report metrics so that they can be picked up by external systems. Currently, Tehuti supports only JMX-based reporting, but the architecture is extendable to more types of reporting systems.

MetricsRepository metrics = new MetricsRepository(); // this is the global repository of metrics and sensors
metrics.addReporter(new JmxReporter("prefix.for.all.metrics.names."));

About

A simple metrics library.

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages

  • Java 99.4%
  • Shell 0.6%