https://bz.apache.org/bugzilla/show_bug.cgi?id=57832
regilero <[email protected]> changed: What |Removed |Added ---------------------------------------------------------------------------- CC| |[email protected] | |om --- Comment #1 from regilero <[email protected]> --- Created attachment 32685 --> https://bz.apache.org/bugzilla/attachment.cgi?id=32685&action=edit proposed patch on httpd 2.4.12 Here is a proposal. That's my first try with buckets and brigades so it needs a deep review. But "it works for me". Here I add a check on the stream BEFORE sending any new request, I check that the input stream is empty (but allow one remaining crlf for backward compatibility with old backends). If it's not empty the bakend keepalived connection is rejected and marked for closure. This is made in a retry loop (which was there to allow 2 retry on some sort of backend alive test), but I did not increment the retry counter. So if all backend connections are bad we can spend a quite long time in this loop, cleaning all bad connections, but I think this is better than sending 503 responses. I did not try to patch ap_proxy_connect_backend, where the backend connection is retrieved, because it may be used by some other protocols, with different expectations. -- You are receiving this mail because: You are the assignee for the bug. --------------------------------------------------------------------- To unsubscribe, e-mail: [email protected] For additional commands, e-mail: [email protected]
