On Tue, 30 Mar 2021, Yann Ylavic wrote:
Hi Niklas,
On Tue, Mar 30, 2021 at 4:34 PM Niklas Edmundsson <[email protected]> wrote:
I currently see this on a Ubuntu 18.04 machine, from server-status
(raw cut&paste, so the formatting isn't the best):
--------------------8<---------------------------
Server Version: Apache/2.4.46 (Unix) OpenSSL/1.1.1
Is that Ubuntu's httpd? It looks like Bionic is bundling an older one.
It's our own build of the unpatched httpd.apache.org source. We're in
the process of rolling these hosts to Ubuntu 20.04LTS which has the
same httpd version but an OS-specific build.
Also, which modules are you loading?
server-info says:
Loaded Modules
core.c, event.c, http_core.c, mod_access_compat.c, mod_alias.c,
mod_authn_core.c, mod_authz_core.c, mod_authz_host.c, mod_autoindex.c,
mod_cache.c, mod_cache_disk_largefile.c, mod_dir.c, mod_env.c,
mod_expires.c, mod_include.c, mod_info.c, mod_log_config.c,
mod_logio.c, mod_mime.c, mod_reqtimeout.c, mod_rewrite.c,
mod_setenvif.c, mod_so.c, mod_socache_shmcb.c, mod_ssl.c,
mod_status.c, mod_systemd.c, mod_unixd.c, mod_version.c,
Out of these, mod_cache_disk_largefile nonstandard, it's our own take
on mod_cache_disk to suit our workload. It feels unlikely to have
these side effects, although anything is possible I guess...
What kind of traffic (static, cgi, proxy, websocket, http2..)?
static files and directory listings, http and https, this is one of
the frontend hosts for ftp.acc.umu.se (the file archive run by
Academic Computer Club, Umeå University).
How long do connections last?
The frontends serve only files smaller than 4 MiB, anything bigger
gets redirected to offload hosts. There are real slow downloads, but
they tend to finish sooner rather than later...
Looking now, I see that we have set GracefulShutdownTimeout 3600 for
these frontend hosts, and seem to remember that time was chosen to
allow for the vast majority of transfers to finish.
So, what strikes me as very odd is that we have a bunch of PIDs that
are marked as "old gen", but are not stopping (and thus still
accepting new connections). Shouldn't "old gen" processes by default
stop accepting new connections?
Indeed, provided they get the signal to stop..
Ah, so the "old gen" marker in server-status isn't the same mechanism
that's used by the processes for the accepting/stopping logic?
Could you set the global LogLevel to trace6 please and provide the error_log?
If all your virtual hosts have their own LogLevel, this shouldn't
affect the verbosity of the traffic, only the administration.
Ah, nice trick. I'll get that sorted if I get bored over easter (I
don't think we set LogLevel per-vhost so likely need to rearrange the
config a bit).
/Nikke
--
-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-
Niklas Edmundsson, Admin @ {acc,hpc2n}.umu.se | [email protected]
---------------------------------------------------------------------------
Pain: Giving a Klingon a noogie.
=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=
