Skip to content
A quantiles plugin for OpenTSDB using Ted Dunning's T-Digest
Branch: master
Clone or download
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Type Name Latest commit message Commit time
Failed to load latest commit information.

   ___                 _____ ____  ____  ____
  / _ \ _ __   ___ _ _|_   _/ ___||  _ \| __ )
 | | | | '_ \ / _ \ '_ \| | \___ \| | | |  _ \
 | |_| | |_) |  __/ | | | |  ___) | |_| | |_) |
  \___/| .__/ \___|_| |_|_| |____/|____/|____/
       |_|    The modern time series database.

Build Status Coverage Status


This plugin is used for storing and querying [Ted Dunning's T-Digests] ( as histograms with [OpenTSDB] ( Ted's data structure efficiently stores measurements rank based statistics. Because of the structure, quantiles at the extreme edges (99.999%, etc) have less error than those toward the mean, making this structure useful for tracking the best or worst of a time series.


  1. Download the source code and run mvn package to create the jar in the target/ directory. Copy this file to your OpenTSDB plugin directory as defined in the opentsdb config via tsd.core.plugin_path.
  2. Add the appropriate codec class to the tsd.core.histograms.config config in opentsdb.conf. E.g. { "MergingTDigestTDigestCodec":2 }
  3. Restart the TSD and make sure the plugin was loaded and associated with the proper ID. E.g. look in the logs for lines like:
2017-06-03 16:26:55,044 DEBUG [main] PluginLoader: Successfully added JAR to class loader: /Users/clarsen/Documents/opentsdb/plugins/opentsdb-tdigests-2.4.0-SNAPSHOT.jar
2017-06-03 16:26:55,550 INFO  [main] HistogramCodecManager: Successfully loaded decoder 'net.opentsdb.core.MergingTDigestTDigestCodec' with ID 1


Currently sketches are implemented in Java so here's an example of how to create sketches and send them to OpenTSDB. First, import the latest sketches-core jar for your project, e.g.

import java.nio.ByteBuffer;
import javax.xml.bind.DatatypeConverter;

import com.tdunning.math.stats.MergingDigest;
import com.tdunning.math.stats.TDigest;

public class DigestExample {

  private TDigest latency = MergingDigest.createDigest(100);
  public void updateAndEncode() {
    final ByteBuffer buf = ByteBuffer.allocate(latency.smallByteSize());
    final String b64 = DatatypeConverter.printBase64Binary(buf.array());
    // Send the b64 encoded string to the TSD via HTTP or Telnet.

This is an extremely contrived example to start playing with digests. In reality what you'll want to do is create a new digest on a given interval, for example every 60 seconds, update that digest with all of the measurements from your application, then at the end of 60 seconds, flush the sketch to TSD and start a new one.

NOTE: Do not continue updating the same digest and send it over and over again to TSD or your results will be incorrect when querying. Each digest is supposed to be a snapshot over a time period.


To send the digest over HTTP, create a JSON object like the following:

  "metric": "",
  "timestamp": 1346846400,
  "tags": {
    "host": "web01"

and POST it to <tsd>:4242/api/histogram. The value is the base 64 encoded value from the example above.


Similar to HTTP, encode the digest as a base 64 string and call:


You can’t perform that action at this time.