Append ignores time series index when data is identical?

[josa69]

Given that pystore is still maintained...

A pandas dataframe with date as index (type DateTimeIndex) is stored with Pystore. Last entered date is eg 2020-01-01.
I call append to add a row with index 2020-01-02, with all data in all dataframe columns being identical (np.nan) to the row with index 2020-01-01 then only the last row (2020-01-02) is stored.
I suspect the line "combined = dd.concat([current.data, new]).drop_duplicates(keep="last")" in collection.py is the reason.

IRL perhaps unlikely that two days have 100% identical data (EOD stock data) but shouldn't the time series index be honored in this case?

[gnzsnz]

I can't reproduce your issue, but I get something quite similar. basically after an append the index get out of order. data is OKish, but not the index.

is a small file this is not an issue as results can be sorted, but on large files it's very slow

import pandas as pd
import pystore

# create new store
pystore.set_path(path='/tmp/store')
store = pystore.store(datastore='datastore', engine='pyarrow')
test_eod = store.collection(collection='TEST.EOD')

# generate sample data
df = pd.DataFrame(
      list(range(10)), index=pd.date_range(start='2020-1-1', periods=10), columns=['data']
)
df
            data
2020-01-01     0
2020-01-02     1
2020-01-03     2
2020-01-04     3
2020-01-05     4
2020-01-06     5
2020-01-07     6
2020-01-08     7
2020-01-09     8
2020-01-10     9
# generate 2 overlaping sets of data
dfa = df[:-3].copy()
dfb = df[-5:].copy()
# 
dfa
            data
2020-01-01     0
2020-01-02     1
2020-01-03     2
2020-01-04     3
2020-01-05     4
2020-01-06     5
2020-01-07     6

dfb
            data
2020-01-06     5
2020-01-07     6
2020-01-08     7
2020-01-09     8
2020-01-10     9

# write data
test_eod.write('STOCK',data=dfa)
# query data
test_eod.item('STOCK').to_pandas()
            data
2020-01-01     0
2020-01-02     1
2020-01-03     2
2020-01-04     3
2020-01-05     4
2020-01-06     5
2020-01-07     6
# append data
test_eod.append('STOCK',data=dfb)
# query again
test_eod.item('STOCK').to_pandas()
            data
2020-01-03     2
2020-01-04     3
2020-01-08     7
2020-01-01     0
2020-01-02     1
2020-01-05     4
2020-01-06     5
2020-01-07     6
2020-01-09     8
2020-01-10     9

pystore manages overlaping data, so i'm quite sure that's not the issue. The issue seems to be coming from dask, however there's no "sort_index" in dask only sort values and set index, which is not what we need here.

i wonder if this is caused by dask droping support for fastparquet