[PRA-391] Add log forwarding feature

[welpaolo]

No description provided.

[ckfbot]

🤖 Backport labels populated

Labels to this pull request were added automatically by the populate-labels.yaml action.

When the PR is merged, backport PRs according to the labels will be automatically created. To skip the backport creation, remove any unneeded labels before merging the PR.

[Copilot]

🟡 Changes recommended

Unresolved moderate findings affect relation-payload handling, image selection, and image provenance.

Get a fresh assessment by requesting another Copilot review.

Pull request overview

Adds Loki endpoint propagation and log-forwarding coverage for vLLM workloads.

Changes:

• Handles Loki relation data and injects LOKI_URL.
• Adds unit and integration observability tests.
• Updates the default vLLM image configuration.

File summaries

File	Summary	Final findings
tests/integration/test_observability.py	Verifies Loki forwarding end to end.	Nit (2 votes): Use the KSERVE_LLMISVC_IMAGES mapping instead of a hard-coded image.
tests/integration/helpers/cos.py	Adds workload environment lookup support.	None.
charms/kserve-llmisvc/tests/unit/test_loki.py	Tests Loki endpoint handling and rendering.	None.
charms/kserve-llmisvc/src/templates/llmisvc_configs_manifests.yaml.j2	Injects LOKI_URL into workload containers.	None.
charms/kserve-llmisvc/src/default-custom-images.json	Updates the vLLM image entry.	Moderate (1 vote): The vllm entry is not consumed by the scheduler. Moderate (1 vote): The mutable external image tag weakens provenance and reproducibility.
charms/kserve-llmisvc/src/charm.py	Processes Loki relation endpoints.	Moderate (3 votes): Valid JSON scalars can cause .get() to raise; only read url when the decoded payload is a dictionary.

Review details

Suppressed comments (2)

charms/kserve-llmisvc/src/default-custom-images.json:5

• This vllm entry is not consumed by the charm: _scheduler_context still selects images_context["llm_workload"] (and config.yaml exposes only the llm_* keys). Consequently, the default LLMInferenceServiceConfig resources continue to use ghcr.io/llm-d/llm-d-cuda:v0.4.0, so changing this image does not make the default workloads use the log-forwarding image. Wire the forwarding image into the key rendered by the scheduler, or change the scheduler/template image selection accordingly.

    "vllm": "ghcr.io/welpaolo/vllm-cpu:0.19.0",

charms/kserve-llmisvc/src/default-custom-images.json:5

• This changes the image used by the integration workload from a build-specific project image to a mutable tag in an externally controlled GHCR namespace. That weakens artifact provenance and reproducibility for CI and any consumers of this default; publish the log-forwarding image through a trusted project registry and pin an immutable digest, or keep the override local to this test.

    "vllm": "ghcr.io/welpaolo/vllm-cpu:0.19.0",

• Files reviewed: 6/6 changed files
• Comments generated: 2
• Review effort level: Lite

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.