docs(memory): budgets size from the process cgroup limit, not only a container limit

[claudespice]

Summary

The docs described Spice's memory-budget sizing as container-aware / "the cgroup limit in a container". That was accurate for the old probe, which read the mountpoint root /sys/fs/cgroup/memory.max — a path that holds the limit only inside a cgroup-namespaced container.

spiceai/spiceai#12263 replaces it with telemetry::hardware::cgroup_memory_limit(), which walks the process's own cgroup path leaf → mountpoint root and takes the minimum limit found at any level (cgroup v2 memory.max, cgroup v1 memory.limit_in_bytes). A cap set by systemd-run -p MemoryMax=…, a capped parent slice, or a Kubernetes pod cgroup now binds sizing exactly like a container limit does; previously every one of those was invisible and budgets sized from full host RAM (an OOM kill within 25 minutes on a 121 GiB host with a 96 GiB MemoryMax, per the source PR).

runtime::resource_monitor::get_total_memory() delegates to it, so this is the figure behind both the runtime.query.memory_limit default and the DuckDB coordinated memory budget. get_host_memory() is deliberately unchanged — DuckDB's own default still sizes from host RAM, which is why the coordinated budget must project that ceiling from the host figure; the pages keep saying so.

Also notes the new startup INFO log when a cap binds (Memory budgets sized from the cgroup memory limit: … (host total: …), crates/runtime/src/resource_monitor.rs:110).

Source PRs

• fix(runtime): size memory budgets from the process's own cgroup limit spiceai#12263 — fix(runtime): size memory budgets from the process's own cgroup limit (fixes #12179)

Doc pages changed (5) — all vNext only

• reference/spicepod/runtime.md — runtime.query.memory_limit default
• reference/memory.md — DataFusion query memory management (2 spots)
• reference/performance-tuning.md — Memory Limit
• components/data-accelerators/duckdb/index.md — Coordinated memory budget
• components/data-accelerators/duckdb/deployment.md — Memory

Cross-page sweep: grep -rn 'container-aware\|container memory limit\|container limit' website/docs/ → 0 remaining hits. Scoped to vNext because the old container-only probe is what v2.1.2 and earlier actually ship, so the versioned snapshots are correct as written.

Test plan

• cd website && npm run build passes
• Versioned-docs propagation checked — current-behavior change on trunk only, not in v2.1.2
• Files updated: 5

[github-actions]

✅ Pull with Spice Passed

Passing checks:

• ✅ Title meets minimum length requirement (10 characters)
• ✅ Has at least one of the required labels: area/blog, area/docs, area/cookbook, dependencies
• ✅ No banned labels detected
• ✅ Has at least one assignee: claudespice

[github-actions]

🚀 deployed to https://45bdb7c7.spiceai-org-website.pages.dev

[claudespice]

@copilot review