Redis Timeseries, rolling sum with LUA

I have a timeseries I want to compute a rolling sum of. My idea (which is probably suboptimal already) is to have a script to compute each data point then run it from python in a pipeline.

I struggle with the script.

return 
redis.call('TS.ADD',
           KEYS[1],
            ARGV[1],
            redis.call('TS.RANGE',
                       'ts:1s',
                        ARGV[2],
                        ARGV[3],
                        'AGGREGATION',
                        'sum',
                        120000)[1][2])

When I EVAL this on redis-cli, with args 1 ts:rolling:sum 1609011855000 1609013655000

it returns:
(error) ERR Error running script (call to f_e57a3287fc55b792d33ba7c21e3aae715c5ee3e5): @user_script:1: @user_script: 1: Lua redis() command arguments must be strings or integers

the first call

eval "return redis.call('TS.RANGE',
            'ts:1s',
             ARGV[2],
             ARGV[3],
             'AGGREGATION',
             'sum',
             120000)" 0 1609011855000 1609013655000

gives me

1) 1) (integer) 1609012800000
   2) 2057

Strange thing is, I can sum the timestamps but not the values.

I have never touched lua before, so I may be missing something totally obvious here

I’m not sure I’m being clear so here’s in python, with pandas, what I’d like to achieve:

import pandas as pd
df = pd.DataFrame(my_data_granular_to_1s, columns=['date','val']).set_index(date)
df.val.resample('1s').sum().rolling(120).sum() # <= that