Subject: Questions about Ingress timeout and streaming support for Dedicated Server

Waiting for author response

?梨PRO5/3/2026

Hi Zeabur team,

I’m planning to subscribe to Zeabur and register my own public-IP server as a Zeabur Dedicated Server.

I plan to deploy NashiYard, an OpenAI-compatible API gateway. Its core functionality depends heavily on SSE / chunked streaming responses, especially for long-running LLM streaming requests.

I found two related forum cases:

- https://zeabur.com/forum/posts/69f62c1cc0bbffb242d45430
- https://zeabur.com/forum/posts/69aa51f920754fa5cc0c9555

Both cases mention that an older ingress version, specifically ingress 1.1.15, had an approximately 600-second / 10-minute timeout limit for streaming responses, and that Zeabur staff could manually upgrade the ingress component to remove this limitation.

Before I subscribe and register my server, could you please help confirm the following?

1. For newly registered Dedicated Servers, does the current Zeabur ingress version still have the 10-minute streaming response timeout?
2. If this timeout still exists, can Zeabur upgrade the ingress component for my server to the newer version that removes this limitation?
3. Is there any maximum total HTTP request duration, even if the response is continuously streaming data?
4. What are the current default stream timeout and idle timeout values?
5. Does Zeabur Public Networking / Ingress perform any response buffering for `text/event-stream` or chunked streaming responses?
6. Does Zeabur enable gzip or Brotli compression for streaming responses?
7. Does Zeabur Ingress preserve application response headers such as `X-Accel-Buffering: no`, `Cache-Control: no-cache`, and `Content-Type: text/event-stream`?
8. Are there any behavioral differences for streaming responses between HTTP/1.1 and HTTP/2 on Zeabur Public Networking?
9. Are long-lived SSE connections officially supported for API gateway use cases?
10. Do you currently support user-configurable timeout settings?
11. Do you support custom ingress annotations or per-service ingress configuration for Dedicated Servers?
12. If I need full control over reverse proxy settings, is it supported to run my own Nginx/Caddy on the same Dedicated Server, or would that conflict with Zeabur’s K3s/Ingress components using ports 80/443?

This is important for my use case because NashiYard needs reliable long-lived OpenAI-compatible streaming connections.

Thank you.

2 Replies

BCBohan ChenEMPLOYEE5/4/2026

Hi, thanks for the detailed pre-sales rundown — that level of due diligence for an SSE-heavy gateway is exactly the right thing to do. Quick answers below. Note that our ingress is in-house (not nginx-ingress), so the semantics are slightly different from the typical k8s answers.

Streaming / timeout (Q1–Q4, Q9, Q10)

Default state of a brand-new Dedicated Server today: ingress-controller 1.1.15, which enforces a hard ~600s (10-min) total request duration cap. So your long-lived SSE streams would be cut at ~10 minutes out of the box.
Yes, we can upgrade your ingress to remove this — same operation as in the two cases you linked. After the upgrade there is no fixed total request duration; the connection lives as long as your application keeps it open. Multiple customers run OpenAI-compatible streaming gateways on this version today.
Idle timeout after upgrade: no fixed idle timeout that breaks an actively streaming connection. SSE works fine because the server keeps sending bytes.
User-configurable timeout UI: on the roadmap but not GA yet. Until then we upgrade manually on request, and the upgraded ingress simply removes the cap.
Long-lived SSE officially supported: yes, this is an explicitly supported pattern after the ingress upgrade.

Buffering / compression / headers (Q5–Q7)

No buffering on text/event-stream or chunked responses — passes through.
No gzip/Brotli applied to text/event-stream — it is served as-is.
Headers preserved end-to-end — Cache-Control: no-cache, Content-Type: text/event-stream, X-Accel-Buffering: no etc. all reach the client untouched.

HTTP/1.1 vs HTTP/2 (Q8)

Both supported, both stream correctly. HTTP/2 multiplexes on a single TCP connection; per-stream SSE lifetime semantics are equivalent in practice.

Custom ingress / your own reverse proxy (Q11, Q12)

Custom annotations / per-service ingress config: our ingress is custom-built and does not consume nginx-style Ingress annotations. Per-service overrides are not user-exposed today (overlaps with the roadmap item in Q10).
Running your own nginx/Caddy on the same Dedicated Server: ports 80/443 are owned by our ingress-controller DaemonSet (HostNetwork: true), so you cannot take those over. You can run your own reverse proxy on a different port inside a service (e.g. 8080) and bind a Zeabur domain to it through the normal flow — but for NashiYard specifically you almost certainly do not need this. Our ingress already does the streaming pass-through correctly.

TL;DR

For NashiYard (long-lived OpenAI-compatible SSE): yes, this works on a Zeabur Dedicated Server after we run a one-time ingress upgrade on your machine — same op the two prior threads got. Once you have subscribed and provisioned your server, drop the server ID(s) here and we will handle the upgrade.

AAnonymous5/8/2026

This post has been inactive for a while. We will be closing it in 2 days if there is no new activity.