29692785423
68%

Ran 19 Jul 2026 03:28PM UTC

Jobs 1

Files 42

Run time 1min

Badge

Embed ▾

Committed 19 Jul 2026 03:24PM UTC coverage: 68.299%. Remained the same

Build # 29692785423

Build Type

push

github

Committed by

web-flow

Commit Message

feat(runtimes): make MPI launcher depend on worker readiness (#3748)

* feat(runtimes): make MPI launcher depend on worker readiness

The DeepSpeed and MLX ClusterTrainingRuntimes are MPI-based: the launcher
runs mpirun and SSHes into the worker (node) pods, which only accept
connections once sshd is up. Previously the launcher and node replicated
jobs started concurrently, so the launcher could fail its initial SSH
attempts before the workers were ready.

Use JobSet's dependsOn API so the launcher is created only after the node
replicated job reaches Ready. Because JobSet requires a dependency to be
enumerated before the dependent job (and forbids dependsOn on the first
replicated job), the node job is now listed before the launcher in all
four runtime manifests. Tests and the JobSet test wrapper are updated to
cover the new ordering and dependency.

Fixes: #2751
Signed-off-by: Andrey Velichkevich <andrey.velichkevich@gmail.com>
Co-authored-by: Claude <noreply@anthropic.com>

* Add securityContext to MPI Runtimes to fix permission

Signed-off-by: Andrey Velichkevich <andrey.velichkevich@gmail.com>

---------

Signed-off-by: Andrey Velichkevich <andrey.velichkevich@gmail.com>
Co-authored-by: Claude <noreply@anthropic.com>

Coverage Stats

2525 of 3697 relevant lines covered (68.3%)

0.8 hits per line

Jobs

ID	Job ID	Ran	Files	Coverage
1	29692785423.1	19 Jul 2026 03:28PM UTC	42	68.3	GitHub Action Run

kubeflow / trainer / 29692785423
68%

README BADGES
x

Markdown

Textile

RDoc

HTML

Rst

Jobs

Source Files on build 29692785423

kubeflow / trainer / 29692785423 68%

README BADGES x

Markdown

Textile

RDoc

HTML

Rst

Jobs

Source Files on build 29692785423

kubeflow / trainer / 29692785423
68%

README BADGES
x