[erlang-questions] [ANN] pulsedb: storage for time series — library and daemon

[Max Lapshin]

Hi.

I want to announce our library that I've extracted from flussonic (former
erlyvideo):

https://github.com/flussonic/pulsedb

You can save uniform time series (ordered list of (utc,integer) pairs) with
this library with second step and read later for building graphics.

This library was designed to be a replacement for http://opentsdb.net/

It is important to understand that it is not a storage for events, it is
not a replacement for statsd. It cannot work with events that happen
non-uniformly in time.



Right now it has following feature set:

1) you can store data with following structure:

metric_name utc int_value  tag1=value1,tag2=value2

2) you can query data from metrics with different names and tell how to
filter by tags:

sum:total_clients{datacenter=dc01}

3) it uses own file format for storing data on disk. Right now it uses
about 190 kb per one day for storing one metrics (about 2 bytes per value)

4) it can write data with non-monotonic timestamps (really cool, comparing
to others!)

5) you can use it as a library in your software or as a service.

6) you can subscribe to live feeds with the same aggregation as in reading
from disk. I don't know who else has such feature.

7) it has cool network protocol that uses not more than 13 bytes for single
data point (metric name, utc and value)

8) it can be used in a server mode with external authorization and with
websockets feed for real-time graphic.


It can be compared with rrd, whisper (graphite) or opentsdb.
It is closer to opentsdb by features, because it can have many time series
under same name, and distinguish them by tags.


I've written some docs for using it as a library and will later document
server mode.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://erlang.org/pipermail/erlang-questions/attachments/20140128/4f58b88a/attachment.htm>