CDEF works wrong on short distance

[BombermaG]

RRDgraph show wrong data on graph when using CDEF and python callback. Code listing with bad result (from this example with little modification):

import rrdtool
import sys
import math

def my_callback(filename, cf, start, end, step):
    items = int((end - start) / step)
    return {
        'start': start,
        'step': 300,
        'data': {
            'a': [math.sin(x / 200) for x in range(0, items)],
            'b': [math.cos(x / 200) for x in range(10, items)],
            'c': [math.sin(x / 100) for x in range(100, items)],
        }
    }

def main():
    graphv_args = [
        'test.png',
        '--title', 'Callback Demo',
        '--start', '1424540800',
        '--end', 'start+1d',
        '--lower-limit=0',
        '--interlaced',
        '--imgformat', 'PNG',
        '--width=450',
        'DEF:a=cb//extrainfo:a:AVERAGE',
        'DEF:b=cb//:b:AVERAGE',
        'DEF:c=cb//:c:AVERAGE',
        'LINE:a#00b6e4:a',
        'LINE:b#10b634:b',
        'LINE:c#503d14:c',
        'CDEF:test=a,2,*',
        'LINE:test#ff391a:test',
        'VDEF:av=a,AVERAGE',
        'PRINT:av:%8.6lf'
    ]

    rrdtool.register_fetch_cb(my_callback)
    rrdtool.graphv(*graphv_args)

if __name__ == "__main__":
    main()

Result: Look at red line test. It is straight, but must be like a 2, because CDEF:test=a,2, After changing --end to start+2d all looks fine.

import rrdtool
import sys
import math

def my_callback(filename, cf, start, end, step):
    items = int((end - start) / step)
    return {
        'start': start,
        'step': 300,
        'data': {
            'a': [math.sin(x / 200) for x in range(0, items)],
            'b': [math.cos(x / 200) for x in range(10, items)],
            'c': [math.sin(x / 100) for x in range(100, items)],
        }
    }

def main():
    graphv_args = [
        'test.png',
        '--title', 'Callback Demo',
        '--start', '1424540800',
        '--end', 'start+2d',
        '--lower-limit=0',
        '--interlaced',
        '--imgformat', 'PNG',
        '--width=450',
        'DEF:a=cb//extrainfo:a:AVERAGE',
        'DEF:b=cb//:b:AVERAGE',
        'DEF:c=cb//:c:AVERAGE',
        'LINE:a#00b6e4:a',
        'LINE:b#10b634:b',
        'LINE:c#503d14:c',
        'CDEF:test=a,2,*',
        'LINE:test#ff391a:test',
        'VDEF:av=a,AVERAGE',
        'PRINT:av:%8.6lf'
    ]

    rrdtool.register_fetch_cb(my_callback)
    rrdtool.graphv(*graphv_args)

if __name__ == "__main__":
    main()

Result:

Something wrong with CDEF on short period.

Debian 10.6, Python 3.7.3 rrdtool 1.7.1-2, python3-rrdtool 1.7.1-2 installed by apt.

[yrebrac]

You haven't shown the data for a. Have you checked the sin function isn't actually returning all zeros?

[BombermaG]

You haven't shown the data for a. Have you checked the sin function isn't actually returning all zeros?

As you can see, "a" is not the zero line on the chart.

[yrebrac]

2 x 0 = 0. So if series a is all zero, line test will be all zeros

[BombermaG]

2 x 0 = 0. So if series a is all zero, line test will be all zeros

Look at line "a" in the first chart.

[yrebrac]

Different start and end = equals potentially different dataset. rrdtool is resampling and recomputing entirely for each new time range. I'm not saying zero data is the problem, I'm saying you should rule it out :)