Repository files navigation

This is a Redis module for thet-digest data structure which can be used for accurate online accumulation of rank-based statistics such as quantiles and cumulative distribution at a point. The implementation is based on theMerging Digest implementation by the author.

Before going ahead, make sure that the Redis server you're using has support forRedis modules.

First, you'll have to build the Redis t-digest module from source.

make

This should generate a shared library calledtdigest.so in the root folder. You can now load it into Redis by using the followingredis.conf configuration directive:

loadmodule /path/to/tdigest.so

Alternatively, you can load it on an already running Redis server by issuing the following commands:

MODULE LOAD /path/to/tdigest.so

Initializes akey to an empty t-digest structure with thecompression provided or with the default compression of400.

Reply:"OK"

Adds avalue with the specifiedcount. Ifkey is missing, an empty t-digest structure is initialized with a default compression of400. Returns the sum of counts for all values added.

Reply:long long

Merges one or moresourcekey intodestkey. Ifdestkey is missing, an empty t-digest structure is initialized with a default compression of400.

Reply:"OK"

Returns the cumulative distribution for all provided values.value must be a double. The cumulative distribution returned for all values is between0..1.

Reply:double array ornil if key missing

Returns the estimate values at all provided quantiles.quantile must be adouble between0..1.

Reply:double array ornil if key missing

Prints debug information about the t-digest.

Reply: bulk strings array

The reply is of the form:

1) TDIGEST (<compression>, <num_centroids>, <memory size>)2)   CENTROID (<mean>, <weight>)3)   CENTROID (<mean>, <weight>)4)   CENTROID (<mean>, <weight>)5)   ...

Centroids are printed in sorted order with respect to their mean.

The integration tests require a running Redis server so you must haveredis-server on yourPATH or pass its location in an environment variable calledREDIS_SERVER. Tests are written in Python and use thepytest unit testing library.

make test

Bug reports, feature and pull requests are welcome! Please add tests for any non-trivial changes you submit.