docs(d1): document primary reads after replica failures

[harlan-zw]

🔗 Linked issue

Related to #393

❓ Type of change

• 📖 Documentation
• 🐞 Bug fix
• 👌 Enhancement
• ✨ New feature
• 🧹 Chore
• ⚠️ Breaking change

📚 Description

Cloudflare traced failed Skill reads to replicas repeatedly restoring and disconnecting.
This runbook records the disabled-replication operating rule, its latency trade-off, and the commands for checking production.
The architecture reference now explains session routing when replication is disabled.

🤖 AI disclosure: Harlan Agent Kit modified this description. My AI open-source policy.

[harlan-zw]

Codex wrote this production verification report.

Cloudflare case 02370111 identified replicas stuck in a restoration loop.
The agent changed skilld-db read replication from auto to disabled on October 13, 2026, UTC.
A separate cf d1 get confirmed read_replication.mode: disabled.

All 20 pnpm production:smoke checks passed on their first attempt, including the Skill page.
A direct Skill query returned success: true, served_by_primary: true, and served_by_region: OC.
It read one row and wrote zero rows.

Existing replicas may process requests for up to 24 hours, according to Cloudflare support.
This smoke run does not prove regional replica retirement or the absence of future errors.
The October 9 primary storage timeout remains a separate failure path.

[harlan-github-agent]

🤖 MERGED

Harlan Agent Kit posted this automated review. It is not Harlan's personal review or approval. AI open source policy. Last updated: 2026-10-13 16:10 UTC.

GitHub merged this pull request.

No material findings were recorded.

f0020b3e-9dce-441c-b1ae-4d7a8eb9b3ca

[harlan-github-agent]

Harlan Agent Kit posted this automated review. AI open source policy.

Automated Review: READY.

Read the Review results