Cold Start Issue on Cloudflare Workers Paid Plan

Hello Cloudflare Community,

I’m reaching out to share an experience I’ve been having with Cloudflare Workers, particularly around cold start times, and to seek advice or solutions from the community.

Despite being on the Workers Paid Plan, where I expected minimal cold start delays, I’ve consistently encountered cold starts up to 250ms for the first query on my workers. While subsequent queries are much faster, around 60ms, which is within my acceptable range, the initial delay is a significant concern for me.

This performance is quite unexpected and seems to detract from the otherwise excellent experience I’ve had with Cloudflare Workers. I’ve done my due diligence in terms of optimizing my worker scripts and ensuring my configurations are as efficient as possible, but the issue persists.

I am hoping to gather insights from this knowledgeable community:

  1. Has anyone else experienced similar cold start times, and how did you address them?
  2. Are there any specific optimizations or configurations within the Cloudflare Workers environment that could help mitigate this issue?

Any feedback, insights, or guidance would be greatly appreciated. I’m keen to resolve this issue and continue enjoying the benefits of Cloudflare Workers for my applications.

Thank you all in advance for your help and support.

Best,
Bosco

Hi,

Thank you for reaching out to us today. Can you please submit a ticket so we can take a closer look and provide us with more details regarding the issue you’re experiencing?

Kindly,

Hello @Lili84,

Regrettably, as a paid CF Workers subscriber, I’m unable to initiate a support ticket with CF. It appears I must upgrade to a Business account to have the ability to submit a ticket. I find this quite surprising; it seems reasonable to expect access to customer support as a paying customer. Currently, my only recourse is to seek assistance through the CF community forum.

Best regards,
Bosco

Dear @Lili84,

I hope this message finds you well. I am reaching out to follow up on a concern I previously raised. As a paid customer, I am unable to submit a ticket through the official channels, which has necessitated my reliance on the Cloudflare Community for support. Despite submitting my query three days ago, I have yet to receive a resolution or guidance.

I must express that this situation has led to some disappointment with the customer service experience provided by Cloudflare. I am keen to resolve this matter promptly and would greatly appreciate your assistance in facilitating a solution.

Thank you for your understanding and support.

Kind regards,
Bosco

Is there anyone from the Cloudflare team monitoring these discussions? As a subscribing customer, my patience is wearing thin due to the lack of support. The level of service I’m receiving is less than what I’d expect when buying a toothbrush from a supermarket.

To those considering Cloudflare Workers, be skeptical of the claims regarding ‘zero cold start’ times as advertised here: Eliminating cold starts with Cloudflare Workers. My experience suggests that this may be misleading and could warrant being flagged for deceptive marketing practices.

I’ve deployed a basic “Hello World” Cloudflare worker coded in Rust and compiled to WebAssembly (wasm). Despite being a paying customer, I’m experiencing an 821ms delay on initial requests.

The absence of assistance from the Cloudflare team on this matter is astounding. It’s been several days without a resolution to my issue. I’m now considering transitioning my services to AWS due to this loss of trust in Cloudflare, and I’ll no longer endorse their services.

I have also experienced cold starts of up to a 1 seconds sometimes. While it seems better than other serverless options, I find the claims that Cloudflare Workers has eliminated cold starts entirely very deceitful. Cold starts are very much still a thing with CL workers.

I am also seeing ~400-500ms on first call that drops to less than 10ms on next calls. For a Cf worker that just displays the value in a stored KV. (Hono App with KV Binding)

I’ve done some testing and it’s a a definite 50 second spin down time.

One could use secondary “keep alive workers” to ping your site every 50secs.

Multiple t2 micros across all of AWS infrastructure (one per aws account will do the job for free :wink: or use CF geo location traffic redirection to redirect based on location to your ec2 instance that is always running.

1/2 second is just not acceptable.

1 Like