Skip to content

Latest commit

 

History

History
131 lines (69 loc) · 3.13 KB

README_DATA.md

File metadata and controls

131 lines (69 loc) · 3.13 KB
Raw

Testing Data

Here we spell out the conventions for how to get data needed to test H2O for different test environments. All data is retrieved using the gradlew wrapper that comes as a part of this repository. Please keep in mind that anywhere below that says $ ./gradlew may require you run C:\> gradlew.bat on windows machines instead. All sync commands use file size and "last modified" time stamps of each local file to determine whether the file needs to be updated.

Use cases

Pre-push tests

Pre-push tests are intended to be fast and run often. Running time for all tests should be a few minutes.

Hardware assumptions:

  • 8GB or better RAM.
  • Five 1GB JVMs get started.

Data assumptions:

  • All test data is either generated or exists in the s3://h2o-public-test-data/smalldata/ directory.

Test environment:

  • Driven by java unit test runners

How to get the data:

$ ./gradlew syncSmalldata

How to run tests:

$ ./gradlew test

Laptop (or small server)

Laptop tests are meant to run stressful workloads on a powerful laptop. Running time for all tests should be less than an hour.

Hardware assumptions:

  • At least 12 GB of RAM. At least 2 CPUs. (Most H2O developers use a 16 GB Macbook Pro with 4 CPUs and 8 hardware threads.)

Data assumptions:

  • Max 5 GB of data to download.

  • Search for data in the following order:

    1. path specified by environment variable H2O_BIGDATA + "/laptop"
    2. ./bigdata/laptop (A "magic" directory in your git workspace)
    3. /home/h2opublictestdata/bigdata/laptop
    4. /mnt/h2o-public-test-data/bigdata/laptop

Test environment:

  • RUnit tests
  • Python tests

How to get the data:

$ ./gradlew syncBigdataLaptop

How to run tests:

./gradlew testLaptop

Big server

Big server tests are meant to run stressful workloads on modern server hardware with lots of resources. Many servers may be used at once to reduce running time.

Hardware assumptions:

  • At least 256 GB of RAM. Lots of CPUs.

Data assumptions:

  • Max 50 GB of data to download.

  • Take advantage of soft and hard links to make bigger datasets.

  • Search for data in the following order:

    1. path specified by environment variable H2O_BIGDATA + "/bigserver"
    2. ./bigdata/bigserver (A "magic" directory in your git workspace)
    3. /home/h2opublictestdata/bigdata/bigserver
    4. /mnt/h2o-public-test-data/bigdata/bigserver

Test environment:

  • RUnit tests
  • Python tests

How to get the data:

CAUTION: Don't do this at home.

$ ./gradlew syncBigdataBigserver

How to run tests:

./gradlew testBigserver

Cluster in EC2

Environment assumptions:

  • Infinite RAM. Lots of cores.

Data assumptions:

  • Data lives in S3. Huge.

  • Search for data in the following order:

    1. s3://h2o-public-test-data/bigdata

Test environment:

  • JVMs running directly in EC2 instances

How to get the data:

You don't. Just point your test directly to s3://h2o-public-test-data/bigdata. Definitely do not copy the data to EBS disks.

How to run tests:

Jenkins launches them nightly or on-demand. (Need instructions for how to do this.)