search

ccin2p3 / samplerr

Round robin timeseries middleware based on riemann and elasticsearch
Eclipse Public License 1.0
15 stars 4 forks source link

time precision #8

Closed faxm0dem closed 8 years ago

faxm0dem commented 8 years ago

It seems the time of events doesn't exactly match the provided step. For a step of 20 for instance, it seems more like 22 is being effective:

2016-03-18T09:56:47.000Z load/load/longterm/avg/20 4.03
2016-03-18T09:57:09.000Z load/load/longterm/avg/20 4.01
2016-03-18T09:57:09.000Z load/load/longterm/avg/20 3.98
2016-03-18T09:57:31.000Z load/load/longterm/avg/20 3.895
2016-03-18T09:57:53.000Z load/load/longterm/avg/20 3.81222222222222
2016-03-18T09:58:15.000Z load/load/longterm/avg/20 3.73625
2016-03-18T09:58:44.780Z load/load/longterm/avg/20 N/A
2016-03-18T09:58:37.000Z load/load/longterm/avg/20 3.68714285714286
2016-03-18T09:58:59.000Z load/load/longterm/avg/20 3.60666666666667
2016-03-18T09:59:21.000Z load/load/longterm/avg/20 3.54

Meybe this is duee to the fact we're using the riemann ttl and somehow a two-second delay is being introduced?

faxm0dem commented 8 years ago

looking at the events from a lower-resolution archive (t/minutes 10) it seems the two second delay is confirmed

samplerr commented 8 years ago

we should definitely use something else than riemann.streams/expired?

faxm0dem commented 8 years ago

this should be fixed in v0.3.3 or worst case: obsoleted