- added CustomHistogram type, allowing to parse and aggregate histogram with statically defined number of buckets and dynamic range, i.e.
some.metric:1|H1.2,1.8
- removed DiffCounter metric type due to being broken and counter intuitive
- sampling was not considered at all before, now it is a first class citizen
- rate aggregation has been added as
update_counter / aggregation_period
, with period provided externally - parser now parses floats from bytes without string conversion, giving around 10% speedup (see
benchmark-parser
branch for details) - capnp protocol schema v2 has been added as a rework of metric encoding into more logical and error-prone structure
- aggregation API has changed, supporting custom percentiles, specifying a list of aggregates, etc
- metric name is now a separate structure allowing to process tag information
- metrics can now have tags in (yet only one) graphite format
- tags are placed in sorted order during aggregation and parsing so they are always correct to be used without creating separate structure for them
- aggregation modes are not supported to put aggregate names in tags or name postfixes
- name prefixing is available per aggregate