Skip to content

Instantly share code, notes, and snippets.

@jonymusky

jonymusky/replayconf-2024.md

Last active October 3, 2024 13:36
Show Gist options
  • Save jonymusky/8d1f2bbf2ed3168f37b1383b77fa2cfb to your computer and use it in GitHub Desktop.
Save jonymusky/8d1f2bbf2ed3168f37b1383b77fa2cfb to your computer and use it in GitHub Desktop.
Download ZIP
Conference Replay2024 - Temporal.io
Raw
replayconf-2024.md

Replay2024 - Temporal Notes

Replay2024 conference held in Seattle on September 19th and 20th, 2024.

Table of Contents

From the keynote

  • Temporal’s goal in the organization is to allow engineers to sleep peacefully and have their time while on-call.
  • System failures lead to chaos:
    • On-call issues slow down the speed of releasing new features.
    • Stress levels among engineers are high, with 27% changing jobs due to on-call stress.
    • With Temporal, no additional code is needed for recovery.

Temporal Cloud

Analyze why many self-hosted companies are transitioning to Temporal Cloud. Moving to Temporal Cloud is also part of Temporal’s philosophy, which is to allow engineers to sleep peacefully. Offering pain-free upgrades, improved monitoring, and optimized workflows. Engineers can focus on innovation rather than firefighting.

Key Updates Announced during the conference

  1. Workflow update performance: Improved speed from 50ms to 30ms in p90.
  2. Worker Auto-tune: Now automatically defines memory and CPU to avoid crashes, using Kubernetes metrics for horizontal pod scaling.
  3. Temporal Cloud: Migration to GPC (Google Cloud Platform).
    • Note: Ensure SSL setup.
  4. Temporal Nexus: Enables communication between multiple namespaces, each with its own rules.

Motivations about nexus: IMG_9094

Best Practices

  • Signals: Avoid overuse of signals as it can be difficult to trace status of a workflow.
  • Child Workflows: Consider using child workflows to paralellize tasks.
  • Workflow Continuity: Design workflows to continue-as-new, ensuring they upgrade smoothly when replay tests succeed.
  • Ideas for SDLC Changes IMG_9117 IMG_9113 2 Large

Strategy

In case a quick response is needed, consider having two workflows: one for the critical part that provides the response, and another that can continue handling non-critical operations at a slower pace

Advanced Techniques / Ideas

  • Parent workflows: Implement techniques to drain queues of child workflows efficiently.
    • Before: Every click created a new parallel workflow, leading to multiple parallel operations.
    • Now: Signals queue a child workflow but wait for the previous one to finish.
  • Scheduled Workflows: Schedule executions when timing is not critical. Avoiding times when we are under heavy traffic.
  • Compensation Workflows: IMG_9111 Large
  • Update with start (something new to replace the use of child workflows): IMG_9112 Large

Deployment & Versioning

  • Pinning Workflows: Introduce new workflow versions (e.g., V2 / V3).
  • Rainbow Deployments: Deploy multiple pods and move traffic dynamically for optimized load balancing.

Additional Insights

  • Workflow ID Naming: Java-based workflows have a naming limit of 256 characters.

  • On self-hosted: Check SSL Certificates expiration. IMG_9103

  • Implement HPA with Kubernetes IMG_9104

  • Info about Breaking Changes IMG_9115

References:

All screenshots were taken during the conference and are not my own. I highly recommend reviewing the talks once they are published and congratulating the engineers behind them

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment