Add RL + batch-inference interleaving guide

[aishukamal]

Stacked on #158; retarget to main after it merges.

Adds guides/rl-batch-interleaving/: a guide that time-slices one verl fully-async RL training job with a stock vLLM batch-inference server on the same GPU. The trainer has absolute priority (cuda-checkpoint C/R); vLLM harvests the trainer's idle valleys via its native sleep mode, driven through the snapshot agent's workload channel.

• The RL job manifest uses the timeslice-verl package from Add timeslice-verl package and verl fully-async time-slicing guide #158 (pkg/integrations/verl), with the verl lifecycle-hook and PG-pinning features installed directly from the fork branch aishukamal/verl@feat/fully-async-lifecycle-hooks.
• The vLLM supervisor and load generator live under examples/ and are marked example-only: the supervisor demonstrates the polite-tenant pattern (waiter poll, readiness-gate drain before sleep, workload-channel registration); production batch serving should sit behind a queue-based front end with retries. The rationale for draining before sleep is documented in the guide.
• Requires a snapshot-agent build with workload-channel default routing (config-less Snapshot/Restore resolves the job's registered workload channel) — the chart-default latest platform images include it; the guide's release-pin note covers pinning a tagged release once one containing orchestrator: fix cold-start Acquire deadlock for pre-provisioned jobs #152 and snapshot-agent: resolve each job's backend via explicit config, workload channel, pod annotation, then default #159 is cut.

Validation: validated e2e on GKE (2x 1-GPU H100 nodes; 80-minute run, 9/9 mixed-backend cycles — vLLM app-channel sleep/wake alongside trainer cuda-checkpoint C/R — with no crash, fault, or NCCL errors).

[coderabbitai]

Important

Review skipped

Draft detected.

Please check the settings in the CodeRabbit UI or the .coderabbit.yaml file in this repository. To trigger a single review, invoke the @coderabbitai review command.

⚙️ Run configuration

Configuration used: defaults

Review profile: CHILL

Plan: Pro Plus

Run ID: 832b753d-0de8-4e87-94db-e88a6d335663

You can disable this status message by setting the reviews.review_status to false in the CodeRabbit configuration file.

Use the checkbox below for a quick retry:

• 🔍 Trigger review

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

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