Repository navigation
test(workloads): add workloads suite with a make workload - #5768
matthewgrossman wants to merge 9 commits into
Conversation
Signed-off-by: Matthew Grossman <mgrossman@nvidia.com>
Signed-off-by: Matthew Grossman <mgrossman@nvidia.com>
Signed-off-by: Matthew Grossman <mgrossman@nvidia.com>
Signed-off-by: Matthew Grossman <mgrossman@nvidia.com>
Signed-off-by: Matthew Grossman <mgrossman@nvidia.com>
|
Auto-sync is disabled for draft pull requests in this repository. Workflows must be run manually. Contributors can view more details about this message here. |
| - `test:e2e-kubernetes` runs Kubernetes E2E with the HA Helm overlay | ||
| (`replicaCount: 2` and bundled PostgreSQL) and the credential-driver suite | ||
| (Kubernetes Secrets plus Vault) in `Branch E2E Checks` | ||
| - `test:e2e-workloads` runs the workloads suite (`tests/suites/workloads`) on |
There was a problem hiding this comment.
for continuity I've used test:e2e:continuity so what about test:e2e:workloads?
Alternatively we could use a / as in: test:e2e/workloads?
There was a problem hiding this comment.
I like test:e2e:continuity ¯_(ツ)_/¯
| for (name, contents) in SOURCES { | ||
| fs::write(project.join(name), contents).context(format!("write {name}"))?; | ||
| } |
There was a problem hiding this comment.
Why don't we just build this into the workload image?
…ites/matthewgrossman Signed-off-by: Matthew Grossman <mgrossman@nvidia.com>
Signed-off-by: Matthew Grossman <mgrossman@nvidia.com>
Signed-off-by: Matthew Grossman <mgrossman@nvidia.com>
Signed-off-by: Matthew Grossman <mgrossman@nvidia.com>
|
/ok to test 7e7441c |
|
Label |
Summary
Add the
workloadstestkit family (tests/suites/workloads, programopenshell-workloads), where real tools do realistic tasks in sandboxes. The first workload builds a small C project withmake. Each workload has its own image, built during setup of a new tmachineubuntu-docker-rootful-workloadsenvironment, so other rows don't pay for it. The suite runs with thetest:e2e:workloadslabel (together withtest:e2e) and on manual runs.The make test fails today because of a real bug: make 4.3 makes a no-op
setresuidcall before each recipe, and the sandbox's seccomp filter denies it (#4102; the fix in #4104 has stalled). The test stays active, so the optional workloads row is red until #4102 is fixed.Related Issue
Part of #5726
Part of #5281
Changes
tests/suites/workloads: new workspace withmake::make_builds_a_c_program:makebuilds the project, the program prints the expected output, andmake -qfinds nothing to rebuild.tests/suites/workloads/make/Dockerfile: pinnednvcr.io/nvidia/base/ubuntu:24.04, gcc, make, thesandboxuser, and the C project in/sandbox/project. The test reads the image name fromOPENSHELL_TEST_WORKLOAD_MAKE_IMAGE.workloadsfeature on the Docker base. Its setup playbook builds the image and publishes the variable. The Dockerfile is a setup input, so changing it rebuilds the cached layer.test:e2e:workloadsadds aworkloads-integrationjob, wired liketest:e2e:continuity(test(continuity): add suites for gateway restart and package upgrade #3521): it needstest:e2efor the integration inputs and is not a required gate.TESTING.mdandCI.md.Testing
rust:lockfiles:checkactionlinton the changed workflows;ansible-playbook --syntax-check;tmachine layer-key ubuntu-docker-rootful-workloadsresolvesmake --versionpasses; the build fails withmake: cc: Operation not permitted(bug(sandbox): GNU Make cannot launch recipes with unchanged UID/GID #4102)ubuntu-docker-rootful-workloadsrow on CIChecklist