Cold Starts

Cold starts happen when Globe's serverless platform needs to initialize a new container to process a request. This is common in environments where resources aren't always running. Since initialization takes time, cold starts introduce latency, which can affect performance and user experience, especially in time-sensitive applications.

Why Cold Starts Matter on Globe

Globe is designed for dynamic scaling and high availability, but this flexibility means containers may shut down when idle and reinitialize when traffic resumes. This process can delay response times for the first request, particularly in low-traffic regions or after a deployment.

How Globe Manages Starts

Globe uses three container start states, based on how recently a container was last active:

1. Cold start: Globe fetches and initializes a new container image.

   • Happens after long idle periods or in a region receiving its first request.
   • Takes around 500ms (currently being optimized).

2. Warm start: Globe resumes a paused container that was active recently.

   • Happens when a container was active recently but became idle.
   • Faster than cold starts, with reduced latency.

3. Hot start: Globe routes the request to an already-running container.

   • Happens when ongoing traffic keeps the container active.
   • Delivers the lowest latency.

Globe's infrastructure is engineered to minimize cold starts compared to other serverless platforms, with optimizations continuing during the preview period.

How to Minimize Cold Start Impact

Optimize Application Size

Smaller applications initialize faster. To reduce size:

• Remove unused dependencies from pubspec.yaml to decrease package payload
• Enable tree shaking when your dependencies support it to eliminate dead code
• Minimize included assets by optimizing images and removing unnecessary resources
• Split functionality into separate deployments when you need different scaling patterns or when components have significantly different resource requirements

Improve Startup Performance

Speed up your application's initialization process by:

• Deferring non-essential initialization until after the first request is handled
• Replacing synchronous operations with asynchronous alternatives during startup
• Implementing lazy loading for components that aren't needed for initial requests
• Caching computation results to avoid repeated expensive operations

Keep Containers Warm

For critical applications with strict latency requirements:

• Create scheduled tasks that periodically ping your application endpoints
• Distribute these pings across your highest-priority regions
• Focus warming efforts on business-critical endpoints that require immediate response

Monitor and analyze

Use Globe's temperature header to understand your application's performance:

• Track cold start frequency patterns by region, time of day, and day of week
• Identify which specific endpoints experience the most cold starts
• Measure actual latency impact to determine if optimization is needed

Cold Starts and Deployment Planning

When planning deployments, factor in how cold starts might affect performance:

• Deploy during lower-traffic windows: This reduces the risk of latency spikes impacting end users. Ideal for apps with predictable usage patterns.
• Warm up critical regions post-deploy: Schedule background pings or traffic simulators to keep containers hot, especially in regions with mission-critical traffic.
• Monitor early post-deploy latency: Use x-globe-temperature metrics to track cold start frequency and take corrective action if needed.

These strategies help maintain a smooth experience during and after deployment rollouts.