Uh oh!
There was an error while loading. Please reload this page.
[opt](parquet-writer) Specify the row group size when writing data to Parquet files. - #35081
Conversation
doris-robot
commented
May 20, 2024
Thank you for your contribution to Apache Doris. Since 2024-03-18, the Document has been moved to doris-website. |
BePPPower
commented
May 20, 2024
run buildall |
clang-tidy review says "All clean, LGTM! 👍" |
doris-robot
commented
May 20, 2024
TeamCity be ut coverage result: |
BePPPower
commented
May 21, 2024
run p0 |
BePPPower
commented
May 31, 2024
run buildall |
clang-tidy review says "All clean, LGTM! 👍" |
doris-robot
commented
May 31, 2024
TeamCity be ut coverage result: |
doris-robot
commented
May 31, 2024
TPC-H: Total hot run time: 41482 ms |
doris-robot
commented
May 31, 2024
TPC-DS: Total hot run time: 169074 ms |
doris-robot
commented
May 31, 2024
ClickBench: Total hot run time: 30.56 s |
PR approved by at least one committer and no changes requested. |
PR approved by anyone and no changes requested. |
kaka11chen
commented
May 31, 2024
This solution will consume a lot of memory, so we need to discuss a better solution. @BePPPower |
BePPPower
commented
Jun 5, 2024
run buildall |
clang-tidy review says "All clean, LGTM! 👍" |
doris-robot
commented
Jun 5, 2024
TeamCity be ut coverage result: |
BePPPower
commented
Jun 6, 2024
run performance |
PR approved by at least one committer and no changes requested. |
| builder.created_by( | ||
| fmt::format("{}({})", doris::get_short_version(), parquet::DEFAULT_CREATED_BY)); | ||
| _parquet_writer_properties = builder.build(); | ||
| _parquet_writer_properties = builder.created_by("DorisBE")->build(); |
doris-robot
commented
Jun 6, 2024
TPC-H: Total hot run time: 41257 ms |
doris-robot
commented
Jun 6, 2024
TPC-DS: Total hot run time: 172761 ms |
doris-robot
commented
Jun 6, 2024
ClickBench: Total hot run time: 30.36 s |
… Parquet files. (apache#35081) Previously, each Block data would write a row group every time when data was written to a Parquet file. Using Arrow's `WriteRecordBatch` interface allows caching data to be written to Parquet. We can control the size of the row group cache." Currently, simply accumulate the size of the blocks to determine the cache size.
… Parquet files. (#35081) Previously, each Block data would write a row group every time when data was written to a Parquet file. Using Arrow's `WriteRecordBatch` interface allows caching data to be written to Parquet. We can control the size of the row group cache." Currently, simply accumulate the size of the blocks to determine the cache size.
… Parquet files. (apache#35081) Previously, each Block data would write a row group every time when data was written to a Parquet file. Using Arrow's `WriteRecordBatch` interface allows caching data to be written to Parquet. We can control the size of the row group cache." Currently, simply accumulate the size of the blocks to determine the cache size.
Previously, each Block data would write a row group every time when data was written to a Parquet file.
Which causing too many row group in a single parquet file. The the reading performance on this kind of file
is very bad.
This PR use Arrow's
WriteRecordBatchinterface, which allows caching data to be written to Parquet.We can control the size of the row group cache."
Currently, simply accumulate the size of the blocks to determine the cache size.