RE: Child processes stuck in R status
Gregory ROCHER <[email protected]> Wed, 17 Sep 2025 16:47:48 +0200
| Newsgroups | gmane.comp.apache.user |
|---|---|
| Message-ID | <[email protected]> |
On 2025/09/16 18:12:14 [email protected] wrote: > I'm wondering if anyone has seen this and/or how it is even possible. > > We're having some issues that at least on the surface look like a > slowloris attack or variant; on the server-status page multiple child > processes appear to be stuck in R status with large seconds-since > values. Hi, We are seeing this too recently on one of our reverse proxy The scoreboard is almost saturated with R requests > __R___R___R_R_RR________RRRRW_RRRRRRR_R__RR_RR_RWRRRRRRRRRRRRRRR > WRRRRRRRRWRRRRW_RR_RR_RRR__RW_RR_____WR_R____R__________R_______ > ________R__R__________RRRRRR_RR_RR_RRRRR_RRR_RRCRRRWRRRWRRRWW_RR > RRRRRRRRRR_RRRRRRWRRRRRWRRWRRRRRRRRRRRRRRRRRRWRWRR_WRRRRRRRRRWRR > RRRRRRWRRRRRRRRRRRRRRRRRRWRRRRRRRRRRRWRRRRRRR_RRRRRRR_RRRR_RRRRR > __RRRRRRRRRR_RRRWRRRR_RRRRRRRRRRRRRRRRRRWRRRWRRRRRRRRRWRRRRRWRRR > WRRRRRRRRRRRRR_RRWRRRRRRRRRRRRRRRRRRRRRRR__________________RRRR_ > _RW__RRRRR_RR____WRR_R_RRR_RRRRRRWWRRRRRR_RRRRRRRR_R............ > ..G..........RRRRRRRRRRRCRRRRRRRR_RRRR___________R____RRRR_____R > R____RRRRRWRR_R_RRR_R___RRRRRRR_RRRRRRRRRR_RRRRRWRRRRRRWR_RRRRRR > W_RRRRRRRR____R_R_W__RR_RR_R__RRR___R___RWR___R____R_____R_RRRRR > RRRR_RRRRRR__RRRRRRRRRR_RRRRRRRWRRRR_RRRRR_RR______________R____ > _____R_____________W______WR_______________________________R____ > __________________RRRRRRRRRRRRRRRRRRRRRRRRR_____R_______________ > ______W__W__RR_R_____________................................... > ........................................ We have changed the number of available workers in the mpm config in order to mitigate the saturation : 150 (default on Debian) to 1000 (now) During this, on lot of R workers server-status shows - the Client ip - the VHost domain.example.com:port - the Request GET or POST /some/path… This apache instance use > Server Version: Apache/2.4.65 (Debian) OpenSSL/1.1.1w mod_qos/11.63 > Server MPM: event On an older apache instance serving the same type of sites > Server Version: Apache/2.4.10 (Debian) OpenSSL/1.0.1t > Server MPM: worker We don't observe these saturations due to R workers And in server-status the VHost and Request aren't present on R workers, only an ip address What MPM are you using ? -- Grégory Rocher