[top posting reodered] On Sun, May 8, 2016 at 7:21 PM, Stefan Eissing <[email protected]> wrote: > >> Am 08.05.2016 um 16:30 schrieb Rainer Jung <[email protected]>: >> >> If that would be consensus, it would mean, we should only reset the request >> info at the start of a connection. For sync connections, that's exactly what >> happens already. For http2 I'm not sure. F > > Isn't conn_rec->keepalives an indicator?
Yes, !c->keepalives would work equivalently, but the issue is that with mpm_event, at BUSY_READ time, the previous scoreboard entry is not necessarily for the same connection, hence we could keep the request-line/vhost related to another connection (mixup discussed in PR 59333). That's why I proposed to restore the previous request's informations saved in the conn_rec itself, and since at the start of the connection those would be blank, this also blanks the scoreboard entry when necessary.
