database/doris
2
0
Fork 0
Code Issues Pull Requests Actions 3 Packages Projects Releases Wiki Activity

[opt](FileReader) InMemoryReader is only used in s3 (#23486)

Browse Source 
If file size < 8MB, the file will be read into memory, and this idea is from https://github.com/apache/hadoop/blob/trunk/hadoop-tools/hadoop-aws/src/site/markdown/tools/hadoop-aws/prefetching.md#s3inmemoryinputstream. However, in some cases, we only read one or two columns in a file, and the actually required bytes is only 1%, resulting in a multiple fold increase in the amount of data read. Therefore, `InMemoryReader` can only used in object storage, and reduce the threshold.
This commit is contained in:
Ashin Gau
2023-08-30 20:43:39 +08:00
committed by GitHub
parent b7404896fa
commit 449c595f9d
4 changed files with 13 additions and 5 deletions
Download Patch File Download Diff File Expand all files Collapse all files

3
be/src/common/config.cpp
View File

@ -850,6 +850,9 @@ DEFINE_Validator(jsonb_type_length_soft_limit_bytes,
// is greater than object_pool_buffer_size, release the object in the unused_object_pool.
DEFINE_Int32(object_pool_buffer_size, "100");
// Threshold of reading a small file into memory
DEFINE_mInt32(in_memory_file_size, "1048576"); // 1MB
// ParquetReaderWrap prefetch buffer size
DEFINE_Int32(parquet_reader_max_buffer_size, "50");
// Max size of parquet page header in bytes