Christopher Faulet

@felipewd glad to see your original issue is fixed. At least one victory this week... However, I'm totally desperate you're also hit by a similar issue to Luke's. I'm amazed...

That's indeed interesting. So all 3.0 are affected. I was looking for a way to drop data on client side. But it seems to be an issue on server side....

@lseelenbinder indeed, said this way it seems logical. However, `SD--` is not systematic. In addition, some bugs on the client side could lead HAProxy to report a server error. But...

We've spent the afternoon with Willy to try to find any issue without success. We are out of idea unfortunately. I guess you should enable some traces to inspect deeper...

Thanks. Your `show sess all` output is really useful. It seems some errors was triggered while buffers were congested. Because of the congestion, the errors were not properly handled. We...

For info, I'm able to reproduce the issue. Here is the scenario: * The server sends many data to the client, faster enough to fill the buffers on the response...

I pushed a commit that should fix your issue. I'm on backports. It should be soon available on 2.8

The commit was backported. I'm closing now, hoping it really fixes your issue. But reopen the issue if not. Thanks for your help to spot the issue !

Hum, it seems I failed the backport of fbc3de6e9e59679d2e9ece3984ce31b6a7dd418f. We should check if outgoing data are blocked to arm the write timeout. But it is armed if the channel is...

The fix was pushed to 2.8. For record the commit is: 67f6756633 ("BUG/MEDIUM: stconn: Really report blocked send if sends are blocked by an error")