Esun Kim
dd3a355cc1
|
1 year ago | |
---|---|---|
.. | ||
test | 1 year ago | |
BUILD | 1 year ago | |
README.md | 1 year ago | |
build_defs.bzl | 1 year ago | |
dockerimage_current_versions.bzl | 1 year ago | |
generate_dockerimage_current_versions_bzl.sh | 1 year ago | |
grpc_build_artifact_task.sh | 1 year ago | |
grpc_build_artifact_task_build_test.sh | 1 year ago | |
grpc_repo_archive.sh | 1 year ago | |
grpc_run_bazel_distribtest_test.sh | 1 year ago | |
grpc_run_cpp_distribtest_test.sh | 1 year ago | |
grpc_run_distribtest_test.sh | 1 year ago | |
grpc_run_simple_command_test.sh | 1 year ago | |
grpc_run_tests_harness_test.sh | 1 year ago | |
workspace_status_cmd.sh | 1 year ago |
README.md
Non-Bazel native tests
This directory contains logic that wraps builds and tests from the non-bazel realm to make them runnable under bazel.
Examples: cmake builds, run_tests.py tests, artifacts, distribtests etc.
NOTE: all tests and their setup under this directory are currently EXPERIMENTAL.
How it works
The //tools/bazelify_tests:repo_archive
target produces an archive that contains grpc at the current head with all its submodules. The rule uses a few tricks to achieve this:
- Uses a workspace status command to obtain the commit SHAs of grpc and all submodules from the workspace.
- When running, it actually jailbreaks from the bazel execroot to access the bazel workspace and create the necessary archives.
- The produced archives are deterministic (they have the same checksum if neither grpc or its submodules have changed).
- The target is defined in such a way so that it behaves "reasonably" from bazel's perspective (always re-runs if commit SHAs have changed, can be cached if not).
After grpc source code is archived, the "bazelified" tests basically depend on the archive and they run a script under a docker container. The script unpacks the archived grpc code and creates a temporary workspace (under bazel's target execroot) and then performs the actions that are needed (e.g. run the run_tests.py test harness, run cmake build etc).
There are two ways the test targets can run under a docker container:
- When running on RBE, all actions run under a docker container by definition.
- When running locally, bazel will start a docker container for each action when docker sandbox is used. (Note that the docker sandbox currently doesn't work on windows)
In both cases, the docker image which is used for any given action is determined by the action's exec_properties
and can be specified as a default
(e.g. by RBE toolchain or by setting --experimental_docker_image
flag) or explicitly for each action. For most tests in this directory,
the test rules actually configure the exec_properties
for you, based on selecting one of the gRPC's testing docker images.
Run tests on RBE
# "--genrule_strategy=remote,local" allows fallback to local execution if action doesn't support running remotely
# (required to be able to run the //tools/bazelify_tests:repo_archive target).
tools/bazel --bazelrc=tools/remote_build/linux.bazelrc test --genrule_strategy=remote,local --workspace_status_command=tools/bazelify_tests/workspace_status_cmd.sh //tools/bazelify_tests/test:basic_tests_linux
Run tests locally under bazel's docker sandbox
tools/bazel --bazelrc=tools/remote_build/linux_docker_sandbox.bazelrc test --workspace_status_command=tools/bazelify_tests/workspace_status_cmd.sh //tools/bazelify_tests/test:basic_tests_linux