Overview

Doris Streamloader is a client tool designed for loading data into Apache Doris. In comparison to single-threaded load using curl, it reduces the load latency of large datasets by its concurrent loading capabilities. It comes with the following features:

• Parallel loading: multi-threaded load for the Stream Load method. You can set the parallelism level using the workers parameter.
• Multi-file load: simultaneously load of multiple files and directories with one shot. It supports recursive file fetching and allows you to specify file names with wildcard characters.
• Path traversal support: support path traversal when the source files are in directories
• Resilience and continuity: in case of partial load failures, it can resume data loading from the point of failure.
• Automatic retry mechanism: in case of loading failures, it can automatically retry a default number of times. If the loading remains unsuccessful, it will print the command for manual retry.

Installation

Version 1.0

Source Code: <https://github.com/apache/doris-streamloader/>

note

The obtained result is the executable binary.

How to use

2. doris-streamloader --source_file={FILE_LIST} --url={FE_OR_BE_SERVER_URL}:{PORT} --header={STREAMLOAD_HEADER} --db={TARGET_DATABASE} --table={TARGET_TABLE}

1. FILE_LIST support:

• Single file

E.g. Load a single file

1. ```json

3. doris-streamloader --source_file="dir" --url="http://localhost:8330" --header="column_separator:|?columns:col1,col2" --db="testdb" --table="testtbl"

5. ```

• Single directory

E.g. Load a single directory

1. doris-streamloader --source_file="dir" --url="http://localhost:8330" --header="column_separator:|?columns:col1,col2" --db="testdb" --table="testtbl" - File names with wildcard characters (enclosed in quotes)

E.g. Load file0.csv, file1.csv, file2.csv

1. doris-streamloader --source_file="file*" --url="http://localhost:8330" --header="column_separator:|?columns:col1,col2" --db="testdb" --table="testtbl" - A list of files separated by commas

E.g. Load file0.csv, file1.csv, file2.csv

1. doris-streamloader --source_file="file0.csv,file1.csv,file2.csv" --url="http://localhost:8330" --header="column_separator:|?columns:col1,col2" --db="testdb" --table="testtbl" - A list of directories separated by commas

E.g. Load dir1, dir2, dir3

1. doris-streamloader --source_file="dir1,dir2,dir3" --url="http://localhost:8330" --header="column_separator:|?columns:col1,col2" --db="testdb" --table="testtbl"

2. STREAMLOAD_HEADER supports all streamload headers separated with ‘?’ if there is more than one

Example:

1. doris-streamloader --source_file="data.csv" --url="http://localhost:8330" --header="column_separator:|?columns:col1,col2" --db="testdb" --table="testtbl"

The parameters above are required, and the following parameters are optional:

Result description

A result will be returned no matter the data loading succeeds or fails.

Examples:

• If the loading succeeds, you will see a result like:

1. Load Result: { 2. "Status": "Success", 3. "TotalRows": 120, 4. "FailLoadRows": 0, 5. "LoadedRows": 120, 6. "FilteredRows": 0, 7. "UnselectedRows": 0, 8. "LoadBytes": 40632, 9. "LoadTimeMs": 971, 10. "LoadFiles": [ 11. "basic.csv", 12. "basic_data1.csv", 13. "basic_data2.csv", 14. "dir1/basic_data.csv", 15. "dir1/basic_data.csv.1", 16. "dir1/basic_data1.csv" 17. ] 18. } - If the loading fails (or partially fails), you will see a retry message:

1. load has some error and auto retry failed, you can retry by : 2. ./doris-streamloader --source_file /mnt/disk1/laihui/doris/tools/tpch-tools/bin/tpch-data/lineitem.tbl.1 --url="http://127.0.0.1:8239" --header="column_separator:|?columns: l_orderkey, l_partkey, l_suppkey, l_linenumber, l_quantity, l_extendedprice, l_discount, l_tax, l_returnflag,l_linestatus, l_shipdate,l_commitdate,l_receiptdate,l_shipinstruct,l_shipmode,l_comment,temp" --db="db" --table="lineitem1" -u root -p "" --compress=false --timeout=36000 --workers=3 --batch=4096 --batch_byte=943718400 --max_byte_per_task=1073741824 --check_utf8=true --report_duration=1 --auto_retry="2,1;1,1;0,1" --auto_retry_times=0 --auto_retry_interval=60

You can copy and execute the command. The failure message will also be provided:

1. Load Result: {
2. "Status": "Failed",
3. "TotalRows": 1,
4. "FailLoadRows": 1,
5. "LoadedRows": 0,
6. "FilteredRows": 0,
7. "UnselectedRows": 0,
8. "LoadBytes": 0,
9. "LoadTimeMs": 104,
10. "LoadFiles": [
11. "/mnt/disk1/laihui/doris/tools/tpch-tools/bin/tpch-data/lineitem.tbl.1"
12. ]
13. }

Best practice

Parameter suggestions

1. Required parameters: --source_file=FILE_LIST --url=FE_OR_BE_SERVER_URL_WITH_PORT --header=STREAMLOAD_HEADER --db=TARGET_DATABASE --table=TARGET_TABLE If you need to load multiple files, you should configure all of them at a time in source_file.
2. The default value of workers is the number of CPU cores. When that is large, for example, 96 cores, the value of workers should be dialed down. The recommended value for most cases is 8.
3. max_byte_per_task is recommended to be large in order to reduce the number of load versions. However, if you encounter a “body exceed max size” and try to avoid adjusting the streaming_load_max_mb parameter (which requires restarting the backend), or if you encounter a -238 TOO MANY SEGMENT error, you can temporarily dial this down. For most cases, this can remain as default.

Two parameters that impacts the number of versions:

• workers: The more workers, the higher concurrency level, and thus the more versions. The recommended value for most cases is 8.
• max_byte_per_task: The larger max_byte_per_task , the larger data size in one single version, and thus the less versions. However, if this is excessively high, it could easily cause an -238 TOO MANY SEGMENT error. For most cases, this can remain as default.

Recommended commands

In most cases, you only need to set the required parameters and workers.

1. ./doris-streamloader --source_file="demo.csv,demoFile*.csv,demoDir" --url="http://127.0.0.1:8030" --header="column_separator:," --db="demo" --table="test_load" --u="root" --workers=8

FAQ

• Before resumable loading was available, to fix any partial failures in loading would require deleting the current table and starting over. In this case, Doris Streamloader would retry automatically. If the retry fails, a retry command will be printed so you can copy and execute it.
• The default maximum data loading size for Doris Streamloader is limited by BE config streaming_load_max_mb (default: 100GB). If you don’t want to restart BE, you can also dial down max_byte_per_task.

To show current streaming_load_max_mb:

1. curl "http://127.0.0.1:8040/api/show_config" - If you encounter an -238 TOO MANY SEGMENT error, you can dial down max_byte_per_task.