net::ERR_HTTP2_PROTOCOL_ERROR with 10.1.30

[Ahmed Ashour <as...@yahoo.com.INVALID>]

Hi all,
Even though the regression should have been fixed in 10.1.30, our team still sees it around once weekly. Twice so far.
With 10.1.29 it was very frequent, that the server can't be used, but with 10.1.30 it is much less, but sadly it seems on rare occasions to occur.

I understand the difficulty of reproducing it, but a hint about using Chrome could be beneficial, as it seems it doesn't happen with Firefox for example.
Remarking the UpgradeProtocol configuration makes the team uses the application server again.

Please let me know if anything to be done to help in this regard.

Thanks a lot,Ahmed Ashour

Re: net::ERR_HTTP2_PROTOCOL_ERROR with 10.1.30

[Ahmed Ashour <as...@yahoo.com.INVALID>]

 > How rare? Once in how many requests? Can you trigger this via automated testing e.g. with wrk?
After working for some time (from 10 minutes to an hour), making a request to a page (with subsequent JS/images) every one minute, the issue is shown, happens on Chrome and Brave.
The requests with wrk are all successful, tested for one hour, using the default settings.
Background:- There is a Tomcat server which is restarted daily, to clean the auto deploy of the applications.- Three members have seen the issue, in the last 3 days.- The main application is protected by basic authentication- Once the issue happens, even the sample non-protected application (html/js) is not accessible.

> Does setting discardRequestsAndResponses="true" help at all?

No, it doesn't.
The settings now are:
<UpgradeProtocol discardRequestsAndResponses="true" className="org.apache.coyote.http2.Http2Protocol" readTimeout="20000"/>

which still gives the error.
The current workaround is to remove the UpgradeProtocol element.

I wonder if low-level logs, or network sniffing can help, as it is a secure connection.
Thanks,Ahmed  

Re: net::ERR_HTTP2_PROTOCOL_ERROR with 10.1.30

[Mark Thomas <ma...@apache.org>]

On 30/09/2024 07:38, Ahmed Ashour wrote:
> Hi all,
> Even though the regression should have been fixed in 10.1.30, our team still sees it around once weekly. Twice so far.
> With 10.1.29 it was very frequent, that the server can't be used, but with 10.1.30 it is much less, but sadly it seems on rare occasions to occur.

How rare? Once in how many requests? Can you trigger this via automated 
testing e.g. with wrk?

> I understand the difficulty of reproducing it, but a hint about using Chrome could be beneficial, as it seems it doesn't happen with Firefox for example.
> Remarking the UpgradeProtocol configuration makes the team uses the application server again.
> 
> Please let me know if anything to be done to help in this regard.

Does setting discardRequestsAndResponses="true" help at all?

Mark


> 
> Thanks a lot,Ahmed Ashour


---------------------------------------------------------------------
To unsubscribe, e-mail: users-unsubscribe@tomcat.apache.org
For additional commands, e-mail: users-help@tomcat.apache.org