how to set divisions for string indexed dask dataframes

Question

I have to load data from a database into a dask dataframe. I am using dd.read_sql_table for that. I need to pass a meta in order to make some int with nan columns use the pd.Int64 datatype. As the index column is string based (e.g. "-7874571842864554321-1403311221-11"), I need to set divisions. But I am failing to do so. What I tried so far:

divisions = [
    '9000000000000000000-0000000000-0',
    '8000000000000000000-0000000000-0',
    '7000000000000000000-0000000000-0',
    '6000000000000000000-0000000000-0',
    '5000000000000000000-0000000000-0',
    '4000000000000000000-0000000000-0',
    '3000000000000000000-0000000000-0',
    '2000000000000000000-0000000000-0',
    '1000000000000000000-0000000000-0',
    '0000000000000000000-0000000000-0',
    '-1000000000000000000-0000000000-0',
    '-2000000000000000000-0000000000-0',
    '-3000000000000000000-0000000000-0',
    '-4000000000000000000-0000000000-0',
    '-5000000000000000000-0000000000-0',
    '-6000000000000000000-0000000000-0',
    '-7000000000000000000-0000000000-0',
    '-8000000000000000000-0000000000-0',
    '-9000000000000000000-0000000000-0'
]
with ProgressBar():
    df = dd.read_sql_table(tablename, DB_CONNECT_STRING, index_col='id', meta=meta,
        divisions=divisions)
df.to_parquet(LOCAL_BUFFER_PATH, engine='pyarrow')

It would be fine if the data was just divided into n divisions, as the index does not say anything meaningful anyways. But what happens with this code right now is that it just seem to use one partition which overflows my memory. What I would expect is that 19 divisions get downloaded and saved separately. However, it just downloads data from my db until the memory is full without saving any.