Skip to content

Latest commit

 

History

History
80 lines (54 loc) · 2.45 KB

README.md

File metadata and controls

80 lines (54 loc) · 2.45 KB
Raw

1. Functional test to Neptune Server

  • Export the following Environmen variables:
    • export NEPTUNE_PROJECT=<Your Project>
    • export NEPTUNE_API_TOKEN=<Your API Token>
  • Run the following script:
    • python3 functional_test.py

2. Load generator to Neptune Server

Installation and setup

Please install the latest, with parallelized sync support version of neptune client:

  pip install git+https://github.com/neptune-ai/neptune-client.git@partitioned

Alternatively, here is release candidate but it doesn't support the parallelized sync.

pip install neptune==1.8.3rc2

You need to remember about setting neptune token and project enviroments. Use the token of the Neptune instance you want to use.

export NEPTUNE_API_TOKEN="YOUR TOKEN"
export NEPTUNE_PROJECT="YOUR PROJECT" # "dzwiedziu/load-test"

There are additional ones, that load generators sets and are adviced for benchmark set-up.

export NEPTUNE_ALLOW_SELF_SIGNED_CERTIFICATE=TRUE
export NEPTUNE_REQUEST_TIMEOUT=90

It is also adviced to increase a limit of open files (neptune-client has separate sync-files for each run).

ulimit -Sn 1000000

Example configurations

20 runs, sending 20 steps every 10 seconds, in each step 100 series and 100 atoms

python3 load_generator.py --runs 20 --steps 20 --series 100 --atoms 100 --step-time=10 --run-name='warmup' --indexed-split=0.5 --sync-partitions=32

50 runs, sending 20 steps every 25 seconds, in each step 5000 series and 100 atoms

python3 load_generator.py --runs 50 --steps 20 --series 5000 --atoms 100 --step-time=25 --run-name='low 50x5000' --indexed-split=0.5 --sync-partitions=32

100 runs, sending 100 steps every 25 seconds, in each step 9000 series and 500 atoms

python3 load_generator.py --runs 100 --steps 20 --series 9000 --atoms 500 --step-time=25 --run-name='medium 100x9500' --indexed-split=0.5 --sync-partitions=32

100K attributes runs

For bigger runs you may want to use experimental version of Neptune Client, that parallelize the syncronization process of a run.

Normally, you also need to set env variables but in this case, load generator will do it for you via setting --sync-partitions=32 flag.

export NEPTUNE_MODE=experimental
export NEPTUNE_ASYNC_PARTITIONS_NUMBER=32

Check it out with 5 runs

 python3 load_generator.py --runs 4 --steps 20 --series 100000 --step-time=25 --run-name='100k 5 runs' --indexed-split=0.1 --sync-partitions=32