Ah, me too, just now.
I doubt that the update is the root cause; we upgraded on September 7, but the issues that Gaston reported appeared on Friday 11/09.
I’m checking the public expo on our side.
Hi guys,
I’ve run some checks on my end and we can confirm the following:
I’ve looked at two important Discourse cookies:
_t: The persistent session token.
_forum_session: Indicates a session that expires when the browser is closed, whereas _t is the one that keeps the account alive.
When I do this in DevTools:
console.log("Current user data:", Discourse.User.current());
I notice several details:
login_method: "oauth": So my account doesn’t log in with Direct Discourse credentials, but through an external identity provider (OAuth2/SSO), presumably managed by you.
no_password: true: Discourse doesn’t store a local password, so I assume it fully delegates access validity to the token issued by the Identity Provider.
previous_visit_at: "2026-09-15T12:37:40.360Z": My previous forum session was active today at 12:37 UTC (14:37 peninsular time). That rules out months‑long abandonment and confirms the session dropped only a few hours ago while I was browsing the forum :slightly_smiling_face:
When the variables maximum_session_age_hours and session_expiry_days return undefined, Discourse hides them in the frontend from users without admin roles, so the sudden logout could be due to either of these two scenarios:
- Expiration of the persistent _t cookie set by the backend.
- Periodic revocation of the Access/Refresh Token in the SSO backend.
We’ll continue investigating,
Sergio Turpín
More things... While I was writing the previous post, this message suddenly appeared:
I was logged out and had to log in again. I managed to capture this:
The error shows a HTTP 429 (Too Many Requests) against the endpoint /message-bus/.../poll, specifically triggered by onVisibilityChange and the user-presence.js module.
I understand that Discourse uses periodic calls to keep real‑time notifications, chat, and user status alive.
The OVHcloud reverse proxy (HAProxy, Cloudflare or Nginx) or Discourse’s own Rack::Attack middleware is treating those bursts as abuse or excessive requests per IP/client and blocks the communication with a 429 code ![]()
If the message-bus channel gets rate‑limited and token validations with SSO consecutively fail, the frontend assumes that the backend or the session is unavailable, which triggers loss of the authenticated state or session invalidation on the client.
Let's see if we can nail it together ![]()
Sergio Turpín
Hello everyone,
What news do you have on your side?
We made a few changes over the past two days, so you may have been disconnected a few times. But since yesterday at 11:30 a.m., you shouldn’t have been disconnected (unless it’s an intentional DDoS attempt
)
As for me, I’ve been logged in since yesterday at noon and I still wasn’t disconnected when I arrived this morning.
Thanks @Incognito
So far everything is fine on my Android smartphone. I’ll see how it holds up throughout the day.
It seems I was right and there really was a small issue on OVH’s servers ![]()
Nope, I'm the one who omitted a parameter in the deployment, now it's more optimized. At OVH we improve every day
![]()
Yeah! So far everything's fine.
I didn't doubt it ![]()
We'll keep you informed if we encounter any anomalies. We're a great QA Tester team ![]()
Stay in touch ![]()
Sergio Turpín
Hello @ all
Since yesterday afternoon at my place there have been no issues, I haven't needed to reconnect.
So far it looks good ![]()
Same in Belgium
it’s turning purple now
Hello everyone,
I continue to experience glitches when I visit certain threads.
I've run tests and the backend works and delivers the content without issues:
The net::ERR_HTTP2_PROTOCOL_ERROR error occurs only when Discourse's frontend requests that same JSON via AJAX/Fetch. I'm not sure if there's an HTTP/2 frame size limit on the front-end proxy.
I get the feeling that when the connection breaks at the protocol level, the thread request fails and the application state gets stuck.
I don't think it's related to cookie over‑injection in a thread, since sometimes that thread fails and other times it doesn't.
If you need me to check anything else on my end, let me know.
Note: I've tried other browsers and another PC and the symptom is the same. However, it only happens intermittently.
Regards,
Sergio Turpín
cc @Incognito
Hello @FabL and @Incognito
For the moment I no longer have any problems on my Android mobile phone as well as on my Windows 11 computer.
If @sturpin has any issues, that means there’s still a tiny setting to adjust. ![]()




