> Am 11.01.2023 um 15:50 schrieb Yann Ylavic <ylavic....@gmail.com>:
>
> On Wed, Jan 11, 2023 at 10:10 AM Stefan Eissing via dev
> <dev@httpd.apache.org> wrote:
>>
>>> Am 10.01.2023 um 22:39 schrieb Christophe JAILLET
>>> <christophe.jail...@wanadoo.fr>:
>>>
>>> 1 issue with pytest:
>>> test/modules/http2/test_600_h2proxy.py .....F....
>>> (details at the end of the mail)
>>>
>>> Don't know if expected or not. Some pytest commits are only in trunk and
>>> have not been backported to 2.4.x. I don't know if it is linked to this
>>> failing test.
>>
>> This one works on my machine(tm). Odd indeed. This tries to verify proxy
>> behaviour in regard to "enable_reuse" and var substitution in the urls
>> authority. See <https://github.com/icing/mod_h2/issues/235>, discussed
>> further in
>> <https://lists.apache.org/thread/tlzfbvopg5k61nz8mhjq518oowkmm43f>.
>>
>> The test has a flexible proxypass using a part of the path to construct the
>> backend url. Accessing first one path and then the other creates backend
>> urls with different port numbers. Those backend resources produce a JSON
>> response carrying the port number used.
>>
>> With "enable_reuse=on" the test expects the backend connection from the
>> first request to be reused on the second, therefore producing a JSON that
>> carries to first port number and not the second.
>
> Does curl reuse the same connection (keepalive) for the two requests?
> Otherwise I think we need "ServerLimit 1" or something for the test to
> be reliable, because the two requests could be handled by two
> different child processes due to TCP queuing/scheduling (and the
> backend connection would not be reused obviously).
Good point. We could limit the server count in this test to avoid contacting
the "other" instance.
For the vote: I see no impact here. Code is behaving as it should, the test is
not reliable.
Cheers,
Stefan