langchain-incident-runbook

# LangChain Incident Runbook ## Overview Standard operating procedures for LangChain production incidents: provider outages, error rate spikes, latency degradation, memory issues, and cost overruns. ## Severity Classification | Level | Description | Response Time | Example | |-------|-------------|---------------|---------| | SEV1 | Complete outage | 15 min | All LLM calls failing | | SEV2 | Major degradation | 30 min | >50% error rate, >10s latency | | SEV3 | Minor degradation | 2 hours | <10% errors, slow responses | | SEV4 | Low impact | 24 hours | Intermittent issues, warnings | ## Runbook 1: LLM Provider Outage ### Detect ```bash # Check provider status pages curl -s https://status.openai.com/api/v2/status.json | jq '.status' curl -s https://status.anthropic.com/api/v2/status.json | jq '.status' ``` ### Diagnose ```typescript async function diagnoseProviders() { const results: Record<string, string> = {}; try { const openai = new ChatOpenAI({ model: "gpt-4o-mini", timeout: 10000 }); await openai.invoke("ping"); results.openai = "OK"; } catch (e: any) { results.openai = `FAIL: ${e.message.slice(0, 100)}`; } try { const anthropic = new ChatAnthropic({ model: "claude-sonnet-4-20250514" }); await anthropic.invoke("ping"); results.anthropic = "OK"; } catch (e: any) { results.anthropic = `FAIL: ${e.message.slice(0, 100)}`; } console.table(results); return results; } ``` ### Mitigate ```typescript // Enable fallbac...