write ColumnMetadata after the column chunk data, not the ColumnChunk
#1947
Conversation
Codecov Report
@@ Coverage Diff @@
## master #1947 +/- ##
=======================================
Coverage 83.48% 83.48%
=======================================
Files 221 221
Lines 57054 57068 +14
=======================================
+ Hits 47629 47641 +12
- Misses 9425 9427 +2
Continue to review full report at Codecov.
|
|
@tustvold PTAL |
There was a problem hiding this comment.
I think this makes sense, I left some minor nits. I think with a simple unit test, this can go in.
FWIW I did some digging and tbh I'm not actually sure if anything actually ever reads this data, parquet-mr doesn't even write it, arrow C++ writes it but doesn't appear to ever read it (apache/parquet-cpp#224). Hive and Impala allegedly makes use of it, but I can't find out where/how/why...
I can't help wondering if this was an oversight in the original parquet specification, not collocating column chunk metadata in the footer, that has since been papered over. All readers I can find simply read the ColumnChunkMetadata from the footer and ignore everything else.
| @@ -611,6 +611,29 @@ impl ColumnChunkMetaData { | |||
| encrypted_column_metadata: None, | |||
| } | |||
| } | |||
|
|
|||
| /// Method to convert to Thrift `ColumnMetaData` | |||
| pub fn to_column_metadata_thrift(&self) -> ColumnMetaData { | |||
There was a problem hiding this comment.
Could we change to_thrift above to use this method
parquet/src/file/writer.rs
Outdated
| /// Returns Ok() if there are not errors serializing and writing data into the sink. | ||
| #[inline] | ||
| fn serialize_column_chunk(&mut self, chunk: parquet::ColumnChunk) -> Result<()> { | ||
| fn serialize_column_chunk( |
There was a problem hiding this comment.
Could we just remove this method and move its content into write_metadata?
There was a problem hiding this comment.
I don't have any preferences
I have the same confuse like you about this metadata. But from the definition of the format https://github.com/apache/parquet-format/blob/54e53e5d7794d383529dd30746378f19a12afd58/src/main/thrift/parquet.thrift#L790, we can know the |
liukun4515
force-pushed
the
parquet_column_append_metadata
branch
from
b104d64
to
82c5534
Compare
|
Many system or reader just read the footer and get the metadata, I think we should just follow the parquet-format. |
alamb
changed the title
ColumnMetadata after the column chunk data, not the ColumnChunk
Which issue does this PR close?
Closes #1946
Rationale for this change
What changes are included in this PR?
Are there any user-facing changes?