Feature fill missing steps #5
Conversation
With this extension all influxdb columns (beside of time and sequence_number) will be available in Graphite. This allows to use timeseries automaticaly calculated by continuous queries, where the column name is named after the group by function, e.g. sum, mean. The column-name is separated by the constant string which is configurable in the config-file. This string enables the possibitity to identify leafs and therefore to shortcut in case of rendering a graph. This may be replaced by caching in the finder.
Provide functional compatibility to whisper db, where missing values generate None (null). Without filling the gaps, Graphite-Web is generating wrong graphics.
|
How do you know the step interval? When there is no retention (schema) config. I guess the step interval depends on the time range? I havent looked at the code but did try to fill in missing steps in grafana recently and it was tricky as I do not know the the rate of measurements so what is a missing step? In graphite we know if there is a measurement missing but not with influxdb. Lets say you input data every 10 seconds, and look at only 10min range what will the step interval be then? If the step interval is below 10seconds then nulls will be inserted between every measurement. |
|
torkelo you're exactly right. we were getting the step as timedelta(2ndpoint-1stpoint) but as I found out yesterday, this didn't work in case of missing values, so i pushed this fix 35e0ad7 we basically need a source of truth, like carbon has storage-schemas.conf ; for now it's hardcoded in graphite-influxdb but at some point i'ld like this info to be available in influx and then we'd just query influx |
An other enhancement to fill missing steps (gaps in timeseries) with None (null) values. This allows to generate correct graphs in Graphite-Web even if there are for some reason some values missing (e.g. caused by a network interrupt).