csringhofer
commented
1 year ago Floats that look like simple fractional numbers in decimal format are often not representable without loss of precision because the base of the exponent is 2, not 10: https://en.wikipedia.org/wiki/Single-precision_floating-point_format
In Impala shell: +-------------------------+ | cast(1.512345 as float) | +-------------------------+ | 1.5123449564 | +-------------------------+
So there isn't really a number like 1.512345 as float - only an approximation of it that has infinite digits as decimal and when the float is formatted to string it has to be decided how many digits to keep.
Impyla returns floats in binary format, so the print format is decided by the user. My guess is that the client used to get the Kudu value simply uses less digits during formatting.
If you want represent decimals without loss you have to use the DECIMAL data type instead of float/double. Decimals can be also used the control the number of digits returned. +-----------------------------------------------+ | cast(cast(1.512345 as float) as decimal(7,6)) | +-----------------------------------------------+ | 1.512345 | +-----------------------------------------------+
I have a float type column in a kudu table. With a value of 1.512345, the python program connecting through impyla will return 1.51234467023 from select. How do I keep the float/double value same as input? Thanks.