[remotego]

What type of PR is this?

• chore / dependency bump

What this PR does and why

Moves the chart's decision-gen image to 0.8.1. That release:

• reads bodylog_service_replicas_ready with role=~"|router". bodylog-exporter 0.2.13 (chart bodylog-exporter 0.3.6) splits a PD release's series by role, and older decision-gen tags then read a replica-typed CR on a DisaggregatedSet as permanently draining, so it could not scale up. Upgrade this chart before the exporter.
• books a workload with no GPU-product pin against the cluster's largest schedulable pool, not always H100.

Changes in charts/llm-slo-decision-gen:

• values.yaml: decisionGen.image.tag goes from 0.8.0 to 0.8.1, with a comment on why.
• Chart.yaml: version goes from 0.3.8 to 0.3.9, appVersion to 0.8.1.
• README.md: image row updated.

Behaviour to check on upgrade:

• On a mixed cluster whose largest pool is not H100, unpinned services move to that pool. Set DEFAULT_GPU_POOL in decisionGen.env to keep a fixed pool.
• With exporter 0.2.13, pdengine services start scaling on their SLOs. That path has not run live yet.

Which issue(s) this PR fixes

Refs modelsphere/slo-scaler-decision-gen#13, modelsphere/llm-openresty#27

How it was tested

• helm lint --strict charts/llm-slo-decision-gen → 1 chart(s) linted, 0 chart(s) failed
• helm template ci charts/llm-slo-decision-gen | grep llm-scaler-decision-gen: → 0.8.1
• hack/check-chart-version.sh origin/main → exit 0
• 4pdosc/llm-scaler-decision-gen:0.8.1 is on Docker Hub (amd64, arm64), from the release run. check-chart-images.sh runs in CI because crane is not installed locally.

Checks

• Chart version bumped
• Docs updated where behaviour, a config key or a chart value moved
• No internal hostnames, IPs, registries or credentials

🤖 Generated with Claude Code