Grouping data with Flux in InfluxDB

Grouping happens based on group keys which effectilve are the name of columns used for the grouping. A group then consists of rows having identical values in the group key columns.

Example

Let’s consider the http api request duration data.

from(bucket: "trial_bucket")
  |> range(start: -15m)
  |> filter(fn: (r) => r._measurement == "http_api_request_duration_seconds")

In respect to the time columns, it is ‘grouped’, by default, by its ‘start’ and ‘end’ times.

“Default grouping”

“Default grouping as chart”

Grouping

Now, with the following snippet, we group data along the _measurement and _field columns.

from(bucket: "trial_bucket")
  |> range(start: -15m)
  |> filter(fn: (r) => r._measurement == "http_api_request_duration_seconds")
  |> group(columns: ["_measurement", "_field"])

This will change the group keys:

And will also shape the charts:

As observations of the same time now belong to the same data, plotting them as a line chart on the time axis produces vertical lines.

Aggregate and ungroup

For the next step let’s calculate the mean values of the request durations.

from(bucket: "trial_bucket")
  |> range(start: -15m)
  |> filter(fn: (r) => r._measurement == "http_api_request_duration_seconds")
  |> group(columns: ["_time"])
  |> mean()
  |> group()  

While we calucalted the mean values for each timestamp, this also resulted in separate tables consisting of a single value. We can use the group() function to ungroup data, that is, to merge all tables into a single table.

This will generate the following chart.

We can produce the same result by explicitly turning off grouping on specific columns with the mode parameter.

mode

The group() function also accepts a mode parameter with the following values:

• by: Group by the columns defined in the columns parameter.
• except: Group by the columns not defined in the columns parameter.

Accordingly, we can modify the above code to get the same result:

from(bucket: "trial_bucket")
  |> range(start: -15m)
  |> filter(fn: (r) => r._measurement == "http_api_request_duration_seconds")
  |> group(columns: ["_time"])
  |> mean()
  |> group(columns: ["_time", "_value"], mode: "except")