Problem description:
I query a dataset from a local InfluxDB which Data Explorer UI queries and plots in 0.02 seconds.

When I load exactly the same dataset into pandas.DataFrame using query_data_frame function it takes 184 seconds. 157 seconds of which cProfile attributes to pandas.to_datetime function, called by PandasDateTimeHelper for each and every row's timestamp. This performance problem seems to originate from the fact that the query receives _time column as text in format %Y-%m-%dT%H:%M:%SZ, which must then be parsed. Complete cProfile output.

It looks like this conversion of _time column represented as uint64 number of nanoseconds to string in ISO datetime format in InfluxDB and then parsing that string back into uint64 number of nanoseconds in influxdb_client is totally unnecessary, if I understand the dataflow correctly.

How do I get rid of these unnecessary conversions for _time column, please?

Is there a way to receive _time column as uint64 number of nanoseconds, which I could cast to pandas.datetime64 with one call to pandas.to_datetime on entire column?

Specifications:

• Client Version: 1.13.0
• InfluxDB Version: 2.0.3
• Platform: Ubuntu 18.04.5 LTS, amd64