1. do not split compress data file
Some data file in hive is compressed with gzip, deflate, etc.
These kinds of file can not be splitted.
2. Support lz4 block codec
for hive scan node, use lz4 block codec instead of lz4 frame codec
4. Support snappy block codec
For hadoop snappy
5. Optimize the `count(*)` query of csv file
For query like `select count(*) from tbl`, only need to split the line, no need to split the column.
Need to pick to branch-2.0 after this PR: #22304
Sometimes, the partitions of a hive table may on different storage, eg, some is on HDFS, others on object storage(cos, etc).
This PR mainly changes:
1. Fix the bug of accessing files via cosn.
2. Add a new field `fs_name` in TFileRangeDesc
This is because, when accessing a file, the BE will get a hdfs client from hdfs client cache, and different file in one query
request may have different fs name, eg, some of are `hdfs://`, some of are `cosn://`, so we need to specify fs name
for each file, otherwise, it may return error:
`reason: IllegalArgumentException: Wrong FS: cosn://doris-build-1308700295/xxxx, expected: hdfs://[172.xxxx:4007](http://172.xxxxx:4007/)`
## Proposed changes
Refactor thoughts: close#22383
Descriptions about `enclose` and `escape`: #22385
## Further comments
2023-08-09:
It's a pity that experiment shows that the original way for parsing plain CSV is faster. Therefor, the refactor is only applied on enclose related code. The plain CSV parser use the original logic.
Fallback of performance is unavoidable anyway. From the `CSV reader`'s perspective, the real weak point may be the write column behavior, proved by the flame graph.
Trimming escape will be enable after fix: #22411 is merged
Cases should be discussed:
1. When an incomplete enclose appears in the beginning of a large scale data, the line delimiter will be unreachable till the EOF, will the buffer become extremely large?
2. What if an infinite line occurs in the case? Essentially, `1.` is equivalent to this.
Only support stream load as trial in this PR, avoid too many unrelated changes. Docs will be added when `enclose` and `escape` is available for all kinds of load.
This PR was originally #16940 , but it has not been updated for a long time due to the original author @Cai-Yao . At present, we will merge some of the code into the master first.
thanks @Cai-Yao @yiguolei
### 1
In previous implementation, for each FileSplit, there will be a `TFileScanRange`, and each `TFileScanRange`
contains a list of `TFileRangeDesc` and a `TFileScanRangeParams`.
So if there are thousands of FileSplit, there will be thousands of `TFileScanRange`, which cause the thrift
data send to BE too large, resulting in:
1. the rpc of sending fragment may fail due to timeout
2. FE will OOM
For a certain query request, the `TFileScanRangeParams` is the common part and is same of all `TFileScanRange`.
So I move this to the `TExecPlanFragmentParams`.
After that, for each FileSplit, there is only a list of `TFileRangeDesc`.
In my test, to query a hive table with 100000 partitions, the size of thrift data reduced from 151MB to 15MB,
and the above 2 issues are gone.
### 2
Support when setting `max_external_file_meta_cache_num` <=0, the file meta cache for parquet footer will
not be used.
Because I found that for some wide table, the footer is too large(1MB after compact, and much more after
deserialized to thrift), it will consuming too much memory of BE when there are many files.
This will be optimized later, here I just support to disable this cache.
Refactor the interface of create_file_reader
the file_size and mtime are merged into FileDescription, not in FileReaderOptions anymore.
Now the file handle cache can get correct file's modification time from FileDescription.
Add HdfsIO for hdfs file reader
pick from [Enhancement](multi-catalog) Add hdfs read statistics profile. #21442
1. fix concurrency bug of s3 fs benchmark tool, to avoid crash on multi thread.
2. Add `prefetch_read` operation to test prefetch reader.
3. add `AWS_EC2_METADATA_DISABLED` env in `start_be.sh` to avoid call ec2 metadata when creating s3 client.
4. add `AWS_MAX_ATTEMPTS` env in `start_be.sh` to avoid warning log of s3 sdk.
For routine load (kafka load), user can produce all data for different
table into single topic and doris will dispatch them into corresponding
table.
Signed-off-by: freemandealer <freeman.zhang1992@gmail.com>
Get the last modification time from file status, and use the combination of path and modification time to generate cache identifier.
When a file is changed, the modification time will be changed, so the former cache path will be invalid.
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+.
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.
A const reference member variables as class member stores a temporary object, which cannot be got after the temporary object being destroyed, cause be core dump while enable debug level log
_broker_addr has been destroyed in BrokerFileReader
1. Enhencement:
For single-charset column separator，csv_reader use another method of `split value`.
2. BugFix
Set `json` file format loading to be sensitive.
In previous implementation, when querying tvf, FE will get schema from BE.
And BE will try to open the first file to get its schema info, but for orc or parquet format,
if the file is empty, it will return error.
But even for an empty file, we can still get schema info from file's footer.
So we should handle the empty file to get schema info correctly.
Also modify the catalog doc to add some FAQ.
The performance of ClickBench Q30 is affected by batch_size:
| batch_size | 1024 | 4096 | 20480 |
| -- | -- | -- | -- |
| Q30 query time | 2.27 | 1.08 | 0.62 |
Because aggregation operator will create a new result block for each batch block, and Q30 has 90 columns, which is time-consuming. Larger batch_size will decrease the number of aggregation blocks, so the larger batch_size will improve performance.
Doris internal reader will read at least 4064 rows even if batch_size < 4064, so this PR keep the process of reading external table the same as internal table.
Support set skip line number for stream load to load csv file.
Usage `-H skip_lines:number`:
```
curl --location-trusted -u root: -T test.csv -H skip_lines:5 -XPUT http://127.0.0.1:8030/api/testDb/testTbl/_stream_load
```
Skip line number also can be used in mysql load as below:
```sql
LOAD DATA
LOCAL
INFILE '${mysql_load_skip_lines}'
INTO TABLE ${tableName}
COLUMNS TERMINATED BY ','
IGNORE 2 LINES
PROPERTIES ("auth" = "root:");
```
The main purpose of this pr is to import `fileCache` for lakehouse reading remote files.
Use the local disk as the cache for reading remote file, so the next time this file is read,
the data can be obtained directly from the local disk.
In addition, this pr includes a few other minor changes
Import File Cache:
1. The imported `fileCache` is called `block_file_cache`, which uses lru replacement policy.
2. Implement a new FileRereader `CachedRemoteFilereader`, so that the logic of `file cache` is hidden under `CachedRemoteFilereader`.
Other changes:
1. Add a new interface `fs()` for `FileReader`.
2. `IOContext` adds some statistical information to count the situation of `FileCache`
Co-authored-by: Lightman <31928846+Lchangliang@users.noreply.github.com>
Fix three bugs when read iceberg v2 tables:
1. The `delete position` in `delete file` represents the position of delete row in the entire file, but the `read range` in
`RowGroupReader` represents the position in current row group. Therefore, we need to subtract the position of first
row of current row group from `delete position`.
2. When only reading the partition columns, `RowGroupReader` skips processing the `delete position`.
3. If the `delete position` has delete all rows in a row group, the `read range` is empty, but we read the whole row
group in such case.
Optimize four performance issues:
1. We change `delete position` to `delete range`, and then merge `delete range` and `read range` into the final read
ranges. This process is too tedious and time-consuming. . we can merge `delete position` and `read range` directly.
2. `delete position` is ordered in a `delete file`, so we can use merge-sort, instead of ordered-set.
3. Initialize `RowGroupReader` when reading, instead of initialize all row groups when opening a `ParquetReader`, to
save memory usage, and the same as `IcebergReader`.
4. Change the recursive call of `_do_lazy_read` to loop logic.
BE will crash when querying partitioned hive table with text format
and put partition column at first of select items.
1. FE should use file slots to set the column mapping index of csv file.
2. BE should use `get_by_name` of block to get right column in a block in csv reader.
This pr does three things：
1. Modified the framework of table-valued-function(tvf).
2. be support `fetch_table_schema` rpc.
3. Implemented `S3(path, AK, SK, format)` table-valued-function.
Support running transactional insert operation with new scan framework. eg:
admin set frontend config("enable_new_load_scan_node" = "true");
begin;
insert into tbl1 values(1,2);
insert into tbl1 values(3,4);
insert into tbl1 values(5,6);
commit;
Add some limitation to transactional insert
Do not support non-literal value in insert stmt
Fix some issue about array type:
Forbid cast other non-array type to NESTED array type, it may cause BE crash.
Add getStringValueForArray() method for Expr, to get valid string-formatted array type value.
Add useLocalSessionState=true in regression-test jdbc url
without this config, the jdbc driver will send some init cmd each time it connect to server, such as
select @@session.tx_read_only.
But when we use transactional insert, after begin command, Doris do not support any other type of
stmt except for insert, commit or rollback.
So adding this config to let the jdbc NOT send cmd when connecting.
1. Missing field and line delimiter
2. When query external table with text(csv) format, we should pass the column position map to BE,
otherwise the column order is wrong.
TODO:
1. For now, if we query csv file with non-exist column, it will return null.
But it should return null or default value of that column.
2. Add regression test after hive docker is ready.
1. remove FE config `enable_array_type`
2. limit the nested depth of array in FE side.
3. Fix bug that when loading array from parquet, the decimal type is treated as bigint
4. Fix loading array from csv(vec-engine), handle null and "null"
5. Change the csv array loading behavior, if the array string format is invalid in csv, it will be converted to null.
6. Remove `check_array_format()`, because it's logic is wrong and meaningless
7. Add stream load csv test cases and more parquet broker load tests
1. Refactor the file reader creation in FileFactory, for simplicity.
Previously, FileFactory had too many `create_file_reader` interfaces.
Now unified into two categories: the interface used by the previous BrokerScanNode,
and the interface used by the new FileScanNode.
And separate the creation methods of readers that read `StreamLoadPipe` and other readers that read files.
2. Modify the StreamLoadPlanner on FE side to support using ExternalFileScanNode
3. Now for generic reader, the file reader will be created inside the reader, not passed from the outside.
4. Add some test cases for csv stream load, the behavior is same as the old broker scanner.
