proomt

Search

Search posts, papers, and topics

All posts

TemporalGeorge He7 min readintermediate

How LlamaIndex uses Temporal to scale reliable document orchestration

Summary

LlamaIndex migrated from RabbitMQ to Temporal for orchestrating complex, multi-stage document processing workflows, scaling to tens of millions of pages daily. This shift enabled them to replace ad-hoc state management and concurrency controls with Temporal's durable execution and workflow primitives, improving reliability and resource efficiency.

  • Traditional message queues require significant custom logic for durable state, complex retries, and fairness in multi-step distributed workflows.
  • Temporal Workflows provide durable execution and state, simplifying the orchestration of complex, long-running distributed tasks.
  • A 'SemaphoreWorkflow' pattern using Temporal's start_workflow with WorkflowIDConflictPolicy.USE_EXISTING and signals can manage concurrency and resource locking.
  • Dedicated Task Queues for critical control plane workflows (like semaphores) prevent application backpressure from impacting core orchestration logic.

Engineers building complex, stateful distributed systems, especially those involving LLM pipelines or document processing, can learn how to leverage workflow orchestration platforms to avoid common pitfalls of ad-hoc solutions.

7/10

Related reading

  1. Durable Digest: August highlights

    Temporal’s August digest rolls out a suite of new features: Serverless Workers for AWS Lambda (public preview), Projects for organizing Cloud resources, Standalone Nexus Operations, Google GenAI and Deep Agents integrations, Activity Operations and Workflow Pause controls, poller autoscaling GA, and .NET/TypeScript Nexus SDKs. It also adds a Web UI news feed, new tutorials, validated patterns, an…

    Temporaltemporal.io5 min
  2. Chaining Activities — from text to vectors

    This post details how Temporal Activities orchestrate an RAG pipeline, from text embedding to vector storage, ensuring PII compliance and idempotency. It covers using an on-premise embedding model, caching vectors in Redis, and upserting to Pinecone with robust retry mechanisms.

    Temporaltemporal.io5 min
  3. Constraint Decay: The Fragility of LLM Agents in Backend Code Generation

    A systematic evaluation of LLM agents generating multi‑file backend code shows a sharp drop in correctness when structural constraints (framework conventions, ORM usage, API contracts) are added. Across 100 tasks in 8 Python web frameworks, assertion pass rates fall ~27 points, with data‑layer bugs (bad queries, ORM violations) driving most failures. Mid‑size models cope with minimal frameworks (…

    arXiv cs.SE (Software Engineering)arxiv.org1 minpaperHN287197