milvus/tests/benchmark
Cai Yudong 84110d2684 Add tests/benchmark and tests/python_test using new python SDK
Signed-off-by: Cai Yudong <yudong.cai@zilliz.com>
2021-02-25 17:35:36 +08:00
..
assets
ci
executors
handlers
results
runners
scheduler
suites
task
README.md
__init__.py
client.py
docker_runner.py
docker_utils.py
helm_utils.py
k8s_runner.py
local_runner.py
locust_file.py
locust_flush_task.py
locust_get_entity_task.py
locust_insert_task.py
locust_search_task.py
locust_task.py
locust_tasks.py
locust_test.py
locust_user.py
main.py
mix_task.py
operation.py
parser.py
requirements.txt
runner.py
search_task.py
test_loop.py
utils.py

README.md

Quick start

Description

This project is used to test performance/reliability/stability for milvus server

  • Test cases can be organized with yaml
  • Test can run with local mode or helm mode

Usage:

pip install requirements.txt

if using local mode, the following libs is optional

pymongo==3.10.0

kubernetes==10.0.1

Demos

  1. Local test

    python3 main.py --local --host=*.* --port=19530 --suite=suites/gpu_search_performance_random50m.yaml

Definitions of test suites

Testers need to write test suite config if adding a customizised test into the current test framework

  1. search_performance: the test typealso we havebuild_performance,insert_performance,accuracy,stability,search_stability
  2. tables: list of test cases
  3. The following fields are in the table field
    • server: run host
    • milvus: config in milvus
    • collection_name: currently support one collection
    • run_count: search count
    • search_params: params of query

Test result

Test result will be uploaded if tests run in helm mode, and will be used to judge if the test run pass or failed