feat: add io_uring micro-benchmark playground

[xgerman]

What

Adds documentdb-playground/io-uring-benchmark/ — a reproducible cloud (AKS/EKS) playground that measures PostgreSQL 18 io_method = io_uring vs worker vs sync on DocumentDB, driving the documentdb/micro-benchmarks Locust read-query suite.

Why / proven feasibility

A spike on a local kind cluster (kernel 6.12, io_uring_disabled=0) established the full enable path:

Finding	Evidence
PG18 18-minimal-trixie ships io_uring	initdb bootstrap postgres attempts io_uring queue setup
The only blocker is container seccomp	RuntimeDefault → FATAL: could not setup io_uring queue: Operation not permitted
Relaxing seccomp turns it on	Unconfined → PG18.3 starts, SHOW io_method = io_uring, pg_stat_io reads flow

CNPG natively exposes spec.seccompProfile, but the DocumentDB CR has no seccomp field, so the playground relaxes seccomp on the CNPG postgres pods via a Kyverno mutate policy.

Contents

• CR variants manifests/documentdb-{sync,worker,io_uring}.yaml — differ only in spec.postgres.parameters.io_method.
• seccomp — Kyverno policies (Unconfined quick-start + hardened Localhost), the curated CNPG io_uring profile (from CNPG #10446), and a node-installer DaemonSet.
• harness — an in-cluster Locust runner + an 00…99 script pipeline + analyze.py (median + speedup-vs-sync table).
• README — proven feasibility, the seccomp deep-dive, and an I/O-bound methodology (dataset > RAM, one variable, repeats + median, pg_stat_io corroboration).

How to run

cd documentdb-playground/io-uring-benchmark
./scripts/00-prereqs.sh && ./scripts/10-deploy-operator.sh && ./scripts/20-deploy-seccomp.sh
STORAGE_CLASS=<fast-ssd> ./scripts/30-deploy-documentdb.sh
DOCUMENT_COUNT=10000000 ./scripts/40-load-data.sh
./scripts/50-run-matrix.sh && ./scripts/60-collect.sh

Notes

• Depends on operator support for spec.postgres.parameters (feat: add PostgreSQL parameter tuning with memory-aware defaults #307). The README's link to postgresql-tuning.md resolves once feat: add PostgreSQL parameter tuning with memory-aware defaults #307 lands.
• Headline numbers are intended to come from a managed cloud cluster with fast NVMe/premium-SSD storage; local kind is a smoke test only (storage is not representative).
• Draft until feat: add PostgreSQL parameter tuning with memory-aware defaults #307 merges and a cloud run populates results/.

Co-authored-by: Copilot 223556219+Copilot@users.noreply.github.com

[documentdb-triage-tool]

🤖 Auto-triaged by documentdb-triage-tool.

Applied: ecosystem, documentation, enhancement
Project fields suggested: Component playground · Priority P3 · Effort XL · Status In Progress
Confidence: 0.85 (mixed)

Reasoning

component from path globs (playground, docs); effort from diff stats (2067+0 LOC, 23 files); LLM: Adds a new multi-file io_uring benchmarking playground (manifests, scripts, Kyverno policies, harness, README) scoped entirely to the playground component; it's a nice-to-have experimental feature with no functional or blocking impact.

If a label is wrong, remove it manually and ping @patty-chow so the rules can be tuned. The bot will not re-label items that already have component labels.

[xgerman]

Superseded by #407, which makes io_uring a first-class operator-native opt-in feature gate (spec.featureGates.IOUring: true). The operator now sets io_method=io_uring and relaxes the postgres container seccomp profile via CNPG's native Cluster.Spec.SeccompProfile — no Kyverno policy required. The benchmark harness in this PR remains available on branch developer/io-uring-benchmark for reference; closing in favor of the productized feature.