[tzachi-runai]

Adds a k3s smoke harness for direct PageBroker filesystem/S3 client operations and opt-in real Qwen GPU checkpoint/restore. The orchestrator deploys isolated synthetic Jobs, runs Qwen validation against an installed Snapshot release, and collects independently validated evidence.

Filesystem synthetic transfers use POSIX: three publications and two restores per publication, with 11 reported steps. S3 uses Model Streamer: three publications and 15 steps covering publication, metadata, repeated restore, prepared abort, and wrong-store rejection. Payload contents, sizes, permissions, bounded cleanup, and staging removal are checked.

Each synthetic Job runs a dedicated broker and client with a shared Unix socket, PID namespace and tmpfs staging. A pidfd identifies the exact daemon. Reports and resource samples survive process termination and are retrieved through a bounded read-only exporter. Credentials are restricted to the S3 components and an owned object prefix.

The opt-in Qwen cases use staged POSIX filesystem restores; the removed filesystem Model Streamer selector is rejected. Rendered native tuning contains shared and S3 settings only. Resource reports retain memory, descriptor, thread, and staging measurements; they do not infer a leak verdict.

Based on #715 (integration coverage after #469); #622 adds repeated endurance validation. Refs #267.

Validation: make check, make test, and make build pass at this PR head. The propagated smoke/endurance helper suite covers all 190 tests (the Kubernetes-model test was also run separately with its dependencies); 28 Qwen helper tests pass. Current changes were not executed on a physical GPU or k3s cluster.

[copy-pr-bot]

This pull request requires additional validation before any workflows can run on NVIDIA's runners.

Pull request vetters can view their responsibilities here.

Contributors can view more details about this message here.

[coderabbitai]

Important

Draft PR not reviewed

Draft PRs are not automatically reviewed by default.

• Trigger a manual review

To automatically review draft PRs, update your CodeRabbit configuration:

reviews:
  auto_review:
    drafts: true

• Autofix · Keep fixing CodeRabbit findings and required CI, and resolving merge conflicts

_{Comment @coderabbitai help to get the list of available commands.}