Cold Start and Concurrency — Guardrails

🧭 Quick Return to Map

You are in a sub-page of Cloud_Serverless.
To reorient, go back here:

Cloud_Serverless — scalable functions and event-driven pipelines

WFGY Global Fix Map — main Emergency Room, 300+ structured fixes

WFGY Problem Map 1.0 — 16 reproducible failure modes

Think of this page as a desk within a ward.
If you need the full triage and all prescriptions, return to the Emergency Room lobby.

A platform-agnostic repair guide for serverless and edge runtimes. Use this page when first calls fail, warm instances drift in behavior, or concurrency spikes break your RAG/agent pipelines. Every step maps to a Problem Map page with measurable targets.

When to use this page

First request after deploy or scale-out fails, times out, or returns partial JSON.
Latency jumps on cold hits, then flips back after a few retries.
Thundering herd on a single key or hot shard.
Vector index not ready on the first few invocations.
Tool calls or multi-agent handoffs stall only under burst.

Acceptance targets

Coverage ≥ 0.70 to the target section after recovery.
ΔS(question, retrieved) ≤ 0.45 on cold and warm paths.
λ remains convergent across 3 paraphrases and 2 seeds.
First-hit success ≥ 0.98 after bootstrap guard is added.
No unbounded fan-out. Concurrency gates present for hot keys.

Fix in 60 seconds

Measure ΔS Compute ΔS(question, retrieved) and ΔS(retrieved, expected anchor). Thresholds: stable < 0.40, transitional 0.40–0.60, risk ≥ 0.60. Open: retrieval-playbook.md
Probe λ_observe Run the same request on a cold instance and a warm instance. If λ flips or ΔS stays high on cold only, add a warm-up fence and idempotent retries. Open: rag-architecture-and-recovery.md
Apply the structural guards

Boot order fence before any RAG step Open: bootstrap-ordering.md
Deadlock and circular wait checks in scale-out Open: deployment-deadlock.md
First-call collapse repair after fresh deploy Open: predeploy-collapse.md
Payload schema so warmers and real traffic use the same contract Open: data-contracts.md

Typical breakpoints → exact fix

Code executes before dependencies are ready Warmers fire, but vector store or secrets are not mounted yet. → bootstrap-ordering.md
First call uses an old build or missing env Cold path hits a skewed revision. → predeploy-collapse.md
Concurrency spike on a single key Multiple cold instances pull the same shard. Add keyed fences and backoff. → Pattern: Bootstrap Deadlock patterns/pattern_bootstrap_deadlock.md
High similarity yet wrong meaning on cold only Index metric or analyzer differs between warm and cold builds. → embedding-vs-semantic.md
Hybrid retriever unstable during fan-out Order changes across instances. Lock query split, then rerank deterministically. → patterns/pattern_query_parsing_split.md · rerankers.md
Long chains stall under burst Split the plan, bridge segments, add alternate paths. → BBPF and BBCR in the core docs rag-architecture-and-recovery.md

Minimal serverless recipe you can copy

Entry:
- Accept {source_id, revision, index_hash, shard_key?, warmup?}

Step 1: Warm-up fence
- If warmup==true:
  - Touch retriever with a read-only probe using the same headers and analyzer.
  - Preload secrets and confirm INDEX_HASH matches.
  - Return {ready:true}.

- If warmup is not set:
  - Check READY_KV[index_hash] == true and SECRETS_OK == true.
  - If not ready → delay 30–90s with capped retries, then fail fast with a fix tip.
  - Specs: bootstrap-ordering.md

Step 2: Concurrency gate
- gate_key = sha256(shard_key || source_id || index_hash)
- Acquire token in KV with TTL and single writer semantics.
- If lock not acquired, enqueue to a queue with jitter and backoff.

Step 3: Retrieval with schema
- Use explicit metric and analyzer.
- Return fields: {snippet_id, section_id, source_url, offsets, tokens}
- Contracts: data-contracts.md

Step 4: ΔS & λ probe
- Compute ΔS(question, retrieved), record λ_state.
- If ΔS ≥ 0.60, return minimal structural fix and stop.

Step 5: Reasoning
- LLM reads TXT OS + Problem Map, enforce cite-then-explain.

Step 6: Trace sink
- Store {ΔS, λ_state, index_hash, gate_key, cold_hit?} for live ops.

Observability you must add

Log cold_hit flag, time to first byte, ΔS, λ_state, INDEX_HASH, gate_key.
Alert when cold_hit P95 grows, or ΔS ≥ 0.60 on cold only. Open: ops/live_monitoring_rag.md

Verification

Run the same test three ways: cold, warm, burst.
Expect coverage ≥ 0.70 and ΔS ≤ 0.45 on all three. Open: eval/eval_rag_precision_recall.md

When to escalate

ΔS stays high after warm-up and gating → re-embed with the checklist and verify against a small gold set. Open: retrieval-playbook.md
Alternating answers only under scale → inspect memory namespace split and revision skew. Open: predeploy-collapse.md

🔗 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

Layer	Page	What it’s for
Proof	WFGY Recognition Map	External citations, integrations, and ecosystem proof
Engine	WFGY 1.0	Original PDF based tension engine
Engine	WFGY 2.0	Production tension kernel and math engine for RAG and agents
Engine	WFGY 3.0	TXT based Singularity tension engine, 131 S class set
Map	Problem Map 1.0	Flagship 16 problem RAG failure checklist and fix map
Map	Problem Map 2.0	RAG focused recovery pipeline
Map	Problem Map 3.0	Global Debug Card, image as a debug protocol layer
Map	Semantic Clinic	Symptom to family to exact fix
Map	Grandma’s Clinic	Plain language stories mapped to Problem Map 1.0
Onboarding	Starter Village	Guided tour for newcomers
App	TXT OS	TXT semantic OS, fast boot
App	Blah Blah Blah	Abstract and paradox Q and A built on TXT OS
App	Blur Blur Blur	Text to image with semantic control
App	Blow Blow Blow	Reasoning game engine and memory demo

If this repository helped, starring it improves discovery so more builders can find the docs and tools.

9.5 KiB Raw Blame History Unescape Escape