PARQUET-2373: Improve I/O performance with bloom_filter_length

[zhangjiashen]

The spec PARQUET-2257 has added bloom_filter_length for reader to load the bloom filter in a single shot. This implementation alters the code to make use of the 'bloom_filter_length' field for loading the bloom filter (consisting of the header and bitset) in order to enhance I/O scheduling.

Jira

• My PR addresses the following Parquet Jira issues and references them in the PR title. For example, "PARQUET-1234: My Parquet PR"
  • https://issues.apache.org/jira/browse/PARQUET-2373
  • relates to: https://issues.apache.org/jira/browse/PARQUET-2257, PARQUET-41: Add bloom filter  #757

Tests

• Unit tests and existing tests in TestParquetFileWriter, TestParquetMetadataConverter, ParquetRewriterTest etc.

Commits

• My commits all reference Jira issues in their subject lines. In addition, my commits follow the guidelines from "How to write a good git commit message":
  1. Subject is separated from body by a blank line
  2. Subject is limited to 50 characters (not including Jira issue reference)
  3. Subject does not end with a period
  4. Subject uses the imperative mood ("add", not "adding")
  5. Body wraps at 72 characters
  6. Body explains "what" and "why", not "how"

Documentation

• In case of new functionality, my PR adds documentation that describes how to use it.
  • All the public functions and the classes in the PR contain Javadoc that explain what it does

[wgtmac]

Thanks for the change! I have left some comments.

[wgtmac]

@zhangjiashen This can be rebased to adopt parquet-format 2.10.0

[zhangjiashen]

@zhangjiashen This can be rebased to adopt parquet-format 2.10.0

@wgtmac I just rebased with master branch and please help take a look when you get a chance?

[mapleFU]

FYI, I've update a BloomFilter with length for testing: apache/parquet-testing#43

[wgtmac]

@zhangjiashen This needs to be rebased again.

[wangyum]

@zhangjiashen@wgtmac There is a data issue after this improvement. The result is empty after this patch. My query is:

CREATE TEMPORARY VIEW parquetTable
USING org.apache.spark.sql.parquet
OPTIONS (
path"viewfs://cluster/path/to/file/00000-62-eff62af9-bea8-4eb6-b9c9-c3ee29fd795e-0-00001.parquet"
);
SELECT*FROM parquetTable
WHERE JNL_DEDUP_ID ='PG.PG_CHARGE_ID.25030465176'AND ACCT_ID =4920012120302AND SEQ_NUM =734343586;

We build bloom filter on the column of JNL_DEDUP_ID using Parquet 1.13.1(Read data using Parquet 1.15.2), and it's data type is string type.

[wgtmac]

Did you mean filtering on bloom filter returns wrong result? It worked on older versions? Is there any way to reproduce it?

[wangyum]

Yes, the results are broken for Parquet 1.14+, but work fine on older versions (Parquet 1.13).
It's not easy to reproduce locally. If needed, I can provide a Parquet file for you to reproduce.

I also make a PR to fix this issue: #3250. Please help to review it.