0

I've noticed weird behaviour when exporting data from Vertica and trying to read it later with parquet (python). Let's say I want to have table dump to parquet:

EXPORT TO PARQUET (directory = '/data/table_name') over (partition by event_date) 
AS select * from table;

it gives me next structure:

/data/table_name
 - event_date=2019-01-01
 - event_date=2019-01-02
 - event_date=2019-01-03
...

Then I'm trying to read it with pyarrow:

import pyarrow.parquet as pq
df = pq.read_table('/data/table_name')

But I'm getting an error of inconsistent schema:

ValueError: Schema in partition[event_date=0] ./event_date=2019-01-01/84087de6-node0001-139759025940222.parquet was different.
user_id: string
event_id: int64
event_name: string
install_date: int32
event_date: int32
site_id: string

vs

user_id: string
event_id: int64
event_name: string
install_date: int32
site_id: string

How come?

P.S. If I read each dir separately - it works fine.

df1 = pq.read_table('/data/table_name/event_date=2019-01-01')
df2 = pq.read_table('/data/table_name/event_date=2019-01-02')
df3 = pq.read_table('/data/table_name/event_date=2019-01-02')

df1.schema == df2.schema == df3.schema
> True
Dmitriy Apollonin
  • 1,418
  • 2
  • 16
  • 30

1 Answers1

1

You need to exclude the partition column (event_date) from the export query:

EXPORT TO PARQUET (directory = '/data/table_name') over (partition by event_date) 
AS SELECT user_id,
          event_id,
          event_name,
          install_date,
          site_id
FROM table;
ya24
  • 490
  • 1
  • 4
  • 16