Presto: Error reading Parquet file

[l1x]

Query:

select internal_id from hive.db.table limit 10;

Output:

Query 20190821_080531_00002_dj2ve failed: com.facebook.presto.spi.type.VarcharType

Same query runs with Hive or Athena perfectly.

Table create:

CREATE TABLE hive.db.table (
   ...
   internal_id varchar,
   ...
)

Parquet schema:

{
  "type": "record",
  "name": "flatschema",
  "namespace": "ns",
  "fields": [ {
    "name": "internal_id",
    "type": ["null", "string"],
    "default": null
  }]
}

Version:

presto-cli --version
Presto CLI 0.220

[babrar]

There is a Stack Overflow thread for this problem:
https://stackoverflow.com/questions/51918860/unable-to-query-parquet-data-with-nested-fields-in-presto-db

Problem: Presto is reading data based on the column indices instead of the column names.

Solution: Add hive.parquet.use-column-names=true to your hive.properties file to force presto to read data using column names instead of column indices.

[l1x]

Thanks @babrar but it still fails:

Query 20190822_083805_00007_2sbhh failed: com.facebook.presto.spi.type.VarcharType

I can load the file with parquet tools and Python. Not sure why Presto cannot read it.

[babrar]

Seems like the same issue as before.
Make sure you restart Presto after adding the new configuration to allow the changes to take effect.

[l1x]

I have made the change and restarted Presto.

It is visible in the live session:

 hive.parquet_fail_with_corrupted_statistics     | false                 | true                  | boolean | Parquet: Fail when
 hive.parquet_use_column_names                   | true                  | true                  | boolean | Experimental: Parq
 hive.parquet_writer_block_size                  | 134217728B            | 134217728B            | varchar | Parquet: Writer bl
 hive.parquet_writer_page_size                   | 1048576B              | 1048576B              | varchar | Parquet: Writer pa
 hive.partition_statistics_sample_size           | 100                   | 100                   | integer | Maximum sample siz

[l1x]

We figured out with @maugly24. The problem is that Presto cannot read from a table where the table create field order does not match the Parquet field order.

Parget field order -> A,B,C
Table in Hive -> A,C, B

It does not matter what is the value hive.parquet_use_column_names, unless I am missing something. Does this value need to be set on all nodes or only on the coordinator node?

[ioah86]

@babrar and I have set this one up on all nodes, worker and coordinator.
We have actually had the scenario where suddenly a new column appeared which was positioned differently in the parquet compared to the Hive metastore, and there was not an issue.

[l1x]

So the setting has to be set on all node I guess.