gpuCI
gpuCI is the name for our GPU-backed CI service based on custom plugins and Jenkins. This allows us to use Docker containers as the build environment for testing GoAi projects through the use of nvidia-docker for GPU pass-through to the containers.
For new pull requests (PRs) and pushes to master triggers jobs on the gpuCI server to build from source and run the included tests. The results are available on the server itself, or if triggered by a PR they are embedded in the PR with the high level result and a link for detailed build information.
Any push to master triggers a gpuCI job that builds and tests the current code, linking and showing the results as a build status icon on the README
When a new PR is opened or an existing one has been updated, gpuCI will trigger a build and embed the results of the build into the status of the PR.
NOTE: There may be a request in the form of a comment on the PR of Can one of the admins verify this patch?, if this appears one of the members of the GoAi team needs to respond with ok to test to allow this PR to be tested. However, if the response is add to whitelist the author of the PR will be allowed to trigger builds without need for authorization on each PR.
Comments of rerun tests or run tests are used in a PR to trigger a retest; as the code has not changed.
This is helpful when trying to troubleshoot issues with the build process and/or to see if the encountered error is persistent. Use gpuCI to examine the Console Output to find more detail.
- To approve PR author for just the current PR
ok to test
- To approve PR author for all PRs in the future
add to whitelist
- To manually trigger tests in a PR
run testsrerun tests
- To skip CI testing (for documentation changes) add the following to the title or body of the PR
skip ci
If you run into issues and need help with gpuCI, please file an issue here