Skip to content
Notifications
Clear all

Anyone else having sudden latency spikes with API Gateway HTTP APIs?

1 Posts
1 Users
0 Reactions
3 Views
(@jennyk8)
Estimable Member
Joined: 1 week ago
Posts: 78
Topic starter   [#16059]

Hey everyone, I'm hoping to crowdsource some wisdom here. For the last few weeks, my team's monitoring has been catching these really unpredictable latency spikes on our AWS API Gateway HTTP APIs (the v2 kind). They're not tied to any of our code deployments or obvious load patterns, which is the frustrating part.

We use these APIs as the public face for a few internal Tableau and Power BI data service backends, so when latency jumps, it directly impacts our self-serve analytics users with slower dashboard loads. We're talking about p95 latencies usually sitting comfortably around 80ms suddenly jumping to 1.5-2 seconds for several minutes at a time, then vanishing. It feels completely arbitrary.

I've gone through my usual methodical checklist, comparing metrics across different stages and routes:
* **Backend Services:** Lambda functions and Fargate tasks show no corresponding CPU or memory spikes. Their execution times are flat.
* **Authentication:** We're using JWT authorizers. No change in token validation times during the spikes.
* **Throttling:** No 429s in the logs, and our usage is well below configured throttles.
* **Region:** It's happening in us-east-1, but I'm wondering if it's isolated to a specific AZ or underlying AWS host.

Has anyone else experienced this recently? I'm trying to determine if this is:
1. Something in our configuration we've overlooked (maybe a caching issue?).
2. A known, intermittent issue with the HTTP API service itself.
3. A networking/routing blip between API Gateway and our VPC-backed integrations.

I'd be incredibly grateful if anyone could share similar experiences or data points. If you've run benchmarks or have comparison tables tracking performance across different serverless API providers (like a comparison of Azure API Management, Google Cloud Endpoints, or even API Gateway v1), I'd love to see your real-world findings. It would help us understand the trade-offs we're making.

For now, I'm diving deeper into CloudWatch Logs Insights for the gateway itself. If I find a smoking gun, I'll be sure to post a follow-up with our resolution path.

~jenny


Let the data speak.


   
Quote