Optimize the strategy of merging small IO to prevent severe read amplification, and turn off merged IO when file cache enabled.
Adjustable parameters:
```
// the max amplified read ratio when merging small IO
max_amplified_read_ratio=0.8
// the min segment size
file_cache_min_file_segment_size = 1048576
```
We encountered one confusing situation where buffered reader were trapped in one endless loop when calling readat. Then we found out that it was all due to the return data size is less than requested.
As the following picture shows, the actual data size is about 2M, and when we called readat it only retrieved about 1MB.
#18976 introduced merge small IO facility to optimize performance, and used by parquet reader.
This PR support this facility in orc reader. Current ORC reader implementation need to reposition parent present stream when reading lazy columns in lazy materialization facility. So let it works by removing `DCHECK_GE(offset, cached_data.end_offset)`.
Add file cache metrics and management.
1. Get file cache metrics
> If the performance of file cache is not efficient, there are currently no metrics to investigate the cause. In practice, hit ratio, disk usage, and segments removed status are very important information.
API: `http://be_host:be_webserver_port/metrics`
File cache metrics for each base path start with `doris_be_file_cache_` prefix. `hits_ratio` is the hit ratio of the cache since BE startup; `removed_elements` is the num of removed segment files since BE startup; Every cache path has three queues: index, normal and disposable. The capacity ratio of the three queues is 1:17:2.
```
doris_be_file_cache_hits_ratio{path="/mnt/datadisk1/gaoxin/file_cache"} 0.500000
doris_be_file_cache_hits_ratio{path="/mnt/datadisk1/gaoxin/small_file_cache"} 0.500000
doris_be_file_cache_removed_elements{path="/mnt/datadisk1/gaoxin/file_cache"} 0
doris_be_file_cache_removed_elements{path="/mnt/datadisk1/gaoxin/small_file_cache"} 0
doris_be_file_cache_normal_queue_max_size{path="/mnt/datadisk1/gaoxin/file_cache"} 912680550400
doris_be_file_cache_normal_queue_max_size{path="/mnt/datadisk1/gaoxin/small_file_cache"} 8500000000
doris_be_file_cache_normal_queue_max_elements{path="/mnt/datadisk1/gaoxin/file_cache"} 217600
doris_be_file_cache_normal_queue_max_elements{path="/mnt/datadisk1/gaoxin/small_file_cache"} 102400
doris_be_file_cache_normal_queue_curr_size{path="/mnt/datadisk1/gaoxin/file_cache"} 14129846
doris_be_file_cache_normal_queue_curr_size{path="/mnt/datadisk1/gaoxin/small_file_cache"} 14874904
doris_be_file_cache_normal_queue_curr_elements{path="/mnt/datadisk1/gaoxin/file_cache"} 18
doris_be_file_cache_normal_queue_curr_elements{path="/mnt/datadisk1/gaoxin/small_file_cache"} 22
...
```
2. Release file cache
> Frequent segment files swapping can seriously affect the performance of file cache. Adding a deletion interface helps users clean up the file cache.
API: `http://be_host:be_webserver_port/api/file_cache?op=release&base_path=${file_cache_base_path}`
Return the number of released segment files. If `base_path` is not provide in url, all cache paths will be released.
It's thread-safe to call this api, so only the segment files not been read currently can be released.
```
{"released_elements":22}
```
3. Specify the base path to store cache data
> Currently, regression testing lacks test cases of file cache, which cannot guarantee the stability of file cache. This interface is generally used in regression testing scenarios. Different queries use different paths to verify different usage cases and performance.
User can set session variable `file_cache_base_path` to specify the base path to store cache data. `file_cache_base_path="random"` as default, means chosing a random path from cached paths to store cache data. If `file_cache_base_path` is not one of the base paths in BE configuration, a random path is used.
Add `MergeRangeFileReader` to merge small IO to optimize parquet&orc read performance.
`MergeRangeFileReader` is a FileReader that efficiently supports random access in format like parquet and orc.
In order to merge small IO in parquet and orc, the random access ranges should be generated when creating the
reader. The random access ranges is a list of ranges that order by offset.
The range in random access ranges should be reading sequentially, can be skipped, but can't be read repeatedly.
When calling read_at, if the start offset located in random access ranges, the slice size should not span two ranges.
For example, in parquet, the random access ranges is the column offsets in a row group.
When reading at offset, if [offset, offset + 8MB) contains many random access ranges,
the reader will read data in [offset, offset + 8MB) as a whole, and copy the data in random access ranges into small
buffers(name as box, default 1MB, 64MB in total). A box can be occupied by many ranges,
and use a reference counter to record how many ranges are cached in the box. If reference counter equals zero,
the box can be release or reused by other ranges. When there is no empty box for a new read operation,
the read operation will do directly.
## Effects
The runtime of ClickBench reduces from 102s to 77s, and the runtime of Query 24 reduces from 24.74s to 9.45s.
The profile of Query 24:
```
VFILE_SCAN_NODE (id=0):(Active: 8s344ms, % non-child: 83.06%)
- FileReadBytes: 534.46 MB
- FileReadCalls: 1.031K (1031)
- FileReadTime: 28s801ms
- GetNextTime: 8s304ms
- MaxScannerThreadNum: 12
- MergedSmallIO: 0ns
- CopyTime: 157.774ms
- MergedBytes: 549.91 MB
- MergedIO: 94
- ReadTime: 28s642ms
- RequestBytes: 507.96 MB
- RequestIO: 1.001K (1001)
- NumScanners: 18
```
1001 request IOs has been merged into 94 IOs.
## Remaining problems
1. Add p2 regression test in nest PR
2. Profiles are scattered in various codes and will be refactored in the next PR
3. Support ORC reader
Currently, there are some useless includes in the codebase. We can use a tool named include-what-you-use to optimize these includes. By using a strict include-what-you-use policy, we can get lots of benefits from it.
Fix tow bugs:
1. Enabling file caching requires both `FE session` and `BE` configurations(enable_file_cache=true) to be enabled.
2. `ParquetReader` has not used `IOContext` previously, but `CachedRemoteFileReader::read_at` needs `IOContext` after PR(#17586).
Co-authored-by: ByteYue <[yj976240184@gmail.com](mailto:yj976240184@gmail.com)>
This PR is an optimization for https://github.com/apache/doris/pull/17478:
1. Change the buffer size of `LineReader` to 4MB to align with the size of prefetch buffer.
2. Lazily prefetch data in the first read to prevent wasted reading.
3. S3 block size is 32MB only, which is too small for a file split. Set 128MB as default file split size.
4. Add `_end_offset` for prefetch buffer to prevent wasted reading.
The query performance of reading data on object storage is improved by more than 3x+.
Follow #17586.
This PR mainly changes:
Remove env/
Remove FileUtils/FilesystemUtils
Some methods are moved to LocalFileSystem
Remove olap/file_cache
Add s3 client cache for s3 file system
In my test, the time of open s3 file can be reduced significantly
Fix cold/hot separation bug for s3 fs.
This is the last PR of #17764.
After this, all IO operation should be in io/fs.
Except for tests in #17586, I also tested some case related to fs io:
clone
concurrency query on local/s3/hdfs
load error log create and clean
disk metrics
See #17764 for details
I have tested:
- Unit test for local/s3/hdfs/broker file system: be/test/io/fs/file_system_test.cpp
- Outfile to local/s3/hdfs/broker.
- Load from local/s3/hdfs/broker.
- Query file on local/s3/hdfs/broker file system, with table value function and catalog.
- Backup/Restore with local/s3/hdfs/broker file system
Not test:
- cold & host data separation case.
