Peidian Li created PARQUET-2152:
-----------------------------------
Summary: zstd compressor and decompressor use the same
configuration
Key: PARQUET-2152
URL: https://issues.apache.org/jira/browse/PARQUET-2152
Project: Parquet
Issue Type: Bug
Components: parquet-mr
Affects Versions: 1.12.2
Reporter: Peidian Li
I use spark to rewrite the parquet files that are compressed by zstd. And the
parquet version is 1.12.2. I want to read the parquet files compressed by
level 3 and compress them on another level. But the level can't be changed.
After I check the source, I found the problem was the codec was cached, and the
configuration will not be updated:
[https://github.com/apache/parquet-mr/blob/master/parquet-hadoop/src/main/java/org/apache/parquet/hadoop/CodecFactory.java#L144]
[https://github.com/apache/parquet-mr/blob/master/parquet-hadoop/src/main/java/org/apache/parquet/hadoop/CodecFactory.java#L226]
I think the problem is important. I found it when I try to use a different
level to compaction the files in the iceberg table. Asynchronous rewriting with
a higher level can lead to higher compression ratio. This is important to save
storage costs.
--
This message was sent by Atlassian Jira
(v8.20.7#820007)
