I did some more research and here are some findings: 1) Based on https://httpd.apache.org/docs/2.4/mod/mod_proxy.html , enablereuse default to On. This means if I remove it, it should be the same as I set it to On. Unfortunately, this contradicts my observation in my original email.
2) https://www.mail-archive.com/dev@httpd.apache.org/msg74177.html seems to say things have changed from prior to 2.4.4x: > + /* Prior to 2.4.4x, regex workers with dollar substitution > + * were never matched and fell into the generic worker. To > + * avoid dns/connection reuse compat issues, let's set > + * disablereuse by default, which can still be overwritten > + * by an explicit enablereuse=on. > + */ This explains my observations in Debian 11 with Apache 2.4.54. When I remove the explicit "enablereuse=on", it defaults to disablereuse. Based on all above, it sounds like Apache enablereuse=on won't work with ProxyPassMatch. We probably should update the Apache docs. Now the question is if mod_proxy_http2, will ProxyPassMatch support connection reuse? Frank On Thu, Sep 22, 2022 at 4:48 PM Frank Liu <gfrank...@gmail.com> wrote: > Hi, > > I am using the bundled apache on Debian 10 and Debian 11 as a reverse > proxy. I found one problem that is only in Debian 11 which bundles apache > 2.4.54. The Debian 10 bundled apache 2.4.38 seems to work as expected. > > ProxyPassMatch ^/(prod|dev)/([-a-z0-9]+)/(.*)$ http://$2.internal/$1/$2/$3 > enablereuse=on keepalive=on > > When the user chrome browses the apache using /prod/host1/something, > apache proxies the request to backend server host1.internal but then when > chrome browses /prod/host2/something, apache again sends to host1.internal. > I would expect the request to go to host2.internal. > > Again, the issue is only with Apache 2.4.54 on Debian11. The old Debiain > 10 (apache 2.4.38) works correctly using the same configuration. The > request goes to host1.internal and host2.internal based on the request URI. > > If I remove "enablereuse=on" from Debian 11 apache config, it will work > correctly. Is this a feature or bug? > > Thanks! > Frank > > > >
