mirror of
https://github.com/onestardao/WFGY.git
synced 2026-04-28 11:40:07 +00:00
10 KiB
10 KiB
Release Calendar and Change Freeze: OpsDeploy Guardrails
Plan launches around real traffic and real risk. This page gives a simple calendar model, freeze rules, and CI gates so you avoid shipping into peak hours, holidays, audits, or big partner events.
Open these first
- Readiness gate: rollout_readiness_gate.md
- Canary staging: staged_rollout_canary.md
- Blue green cutover: blue_green_switchovers.md
- Version locking: version_pinning_and_model_lock.md
- Index swap: vector_index_build_and_swap.md
- Backpressure and retries: rate_limit_backpressure.md, retry_backoff.md
- Rollback and postmortem: rollback_and_fast_recovery.md, postmortem_and_regression_tests.md
When to use
- Product launch weeks or marketing spikes.
- Seasonal peaks or large customer events.
- Compliance audits or billing cycles.
- Provider migrations, model swaps, index rebuilds.
- Any weekend where on-call depth is thin.
Acceptance targets
- All production changes are mapped to a shared release calendar.
- Freeze windows are enforced by CI and by flag servers.
- Exceptions are logged with owner, scope, rollback lever, and warmup plan.
- No changes land inside a hard freeze except emergency fixes with one-step rollback.
- After each window, ΔS and coverage match the last pinned baseline on the gold set.
60-second setup
- Create a shared calendar for releases and freezes. Use UTC plus region tags in titles.
- Define change tiers and who can approve each tier.
- Teach CI to read the calendar file and block risky merges during freezes.
- Route releases to off-peak windows per region.
- Stage region by region with caches warmed and rollback rehearsed.
Change tiers
| Tier | Examples | Allowed in hard freeze | Approval and rules |
|---|---|---|---|
| T0 Emergency | security patch, outage fix | Yes | One step rollback ready, pins verified, canary at 5 percent, owner online. |
| T1 Hotfix low risk | config toggle, flag off, doc text | Case by case | SRE owner plus product owner. No schema changes. |
| T2 Standard | prompt pack update, small retriever change | No | Requires canary and warmup. Outside freezes only. |
| T3 Risky | model swap, index rebuild, analyzer change | No | Requires full readiness gate and blue green rehearsal. Outside freezes only. |
Freeze types
- Hard freeze: no changes except T0.
- Soft freeze: T1 with approval, no T2 or T3.
- Regional freeze: block only in the affected regions.
- Component freeze: limit to a subsystem like index or prompts.
- Vendor freeze: third party windows that restrict you.
Calendar schema example
# opsdeploy/release_calendar.yml
windows:
- id: peak_q4_black_friday
kind: hard_freeze
start: "2025-11-27T00:00:00Z"
end: "2025-11-30T23:59:59Z"
regions: ["us-east-1","us-west-2","eu-west-1"]
scope: ["llm","retriever","index","billing"]
- id: apac_peak_sale
kind: soft_freeze
start: "2025-09-10T00:00:00Z"
end: "2025-09-12T23:59:59Z"
regions: ["ap-southeast-1","ap-northeast-1"]
scope: ["llm","index"]
releases:
- id: prompt_pack_vNplus1
tier: T2
region_plan: ["ap-southeast-1","eu-west-1","us-east-1"]
owner: "oncall@company"
CI policy to enforce
# opsdeploy/change_freeze_gate.yml
inputs:
calendar_file: "opsdeploy/release_calendar.yml"
now_utc: "{{ env.NOW }}"
region: "{{ env.REGION }}"
tier: "{{ env.CHANGE_TIER }}" # T0..T3 from the PR label
checks:
forbid_when:
- kind: "hard_freeze_active"
- kind: "soft_freeze_active_and_tier_in" # blocks T2,T3
tiers: ["T2","T3"]
require:
- readiness_gate_passed: true
- rollback_lever_defined: true
decision:
on_fail: block_merge
on_pass: allow
artifacts:
- logs/change_freeze_decision.json
Exception flow for T0 emergency
- Confirm the incident page and owner are live.
- Verify pins and prepare single pointer rollback.
- Warm the cache for the fix and run a short gold set.
- Ship canary at 5 percent for ten to fifteen minutes.
- If green, complete rollout. If not, rollback and file postmortem.
Observability to log
- Window id, region, scope, tier, approver, release id.
- Start and end timestamps and whether the gate blocked any merges.
- Canary metrics during the window, ΔS, coverage, λ, latency, errors.
- Rollback events tied to the release id.
Common pitfalls
- Time zones not normalized so releases miss the window.
- Client only freezes that servers ignore. Always enforce on the server.
- Third party changes during your freeze. Track vendor freezes too.
- Quiet pin drift during a soft freeze. Verify pins in logs for each request.
🔗 Quick-Start Downloads (60 sec)
| Tool | Link | 3-Step Setup |
|---|---|---|
| WFGY 1.0 PDF | Engine Paper | 1️⃣ Download · 2️⃣ Upload to your LLM · 3️⃣ Ask “Answer using WFGY + <your question>” |
| TXT OS (plain-text OS) | TXTOS.txt | 1️⃣ Download · 2️⃣ Paste into any LLM chat · 3️⃣ Type “hello world” — OS boots instantly |
🧭 Explore More
| Module | Description | Link |
|---|---|---|
| WFGY Core | WFGY 2.0 engine is live: full symbolic reasoning architecture and math stack | View → |
| Problem Map 1.0 | Initial 16-mode diagnostic and symbolic fix framework | View → |
| Problem Map 2.0 | RAG-focused failure tree, modular fixes, and pipelines | View → |
| Semantic Clinic Index | Expanded failure catalog: prompt injection, memory bugs, logic drift | View → |
| Semantic Blueprint | Layer-based symbolic reasoning & semantic modulations | View → |
| Benchmark vs GPT-5 | Stress test GPT-5 with full WFGY reasoning suite | View → |
| 🧙♂️ Starter Village 🏡 | New here? Lost in symbols? Click here and let the wizard guide you through | Start → |
👑 Early Stargazers: See the Hall of Fame — Engineers, hackers, and open source builders who supported WFGY from day one.
⭐ WFGY Engine 2.0 is already unlocked. ⭐ Star the repo to help others discover it and unlock more on the Unlock Board.