Ober the last few weeks we have had some sporadic alerts for:
(ProbeDown) firing: (2) Service mirror1001:443 has failed probes (http_mirrors_wikimedia_org_ip4) - https://wikitech.wikimedia.org/wiki/Runbook#mirror1001:443 -
In the Apache logs we see the following
[Fri Nov 10 10:20:29.712676 2023] [mpm_event:error] [pid 3244005:tid 140056102948160] AH00484: server reached MaxRequestWorkers setting, consider raising the setting
And looking at netstat we can see hundreds of ip's from byte dance address space e.g. bytespider-110-249-202-187.crawl.bytedance.com.
As a side note and possibly a red herring but i notice that IPv6 seems to continue to accept connections which suggests to me that appears may have dedicate threads for v6 and v4. with all v4 threads at max but some free thread available for v6?
As a fix we can increase MaxRequestWorkers but i wonder if we can also add something to ask bytdance to be a bit nicer. When i checked theyre were crawling us ffrom multiple (100+) ip's from at least the following prefixes 111.225.149.0/24, 110.249.201.0/24, 111.225.148.0/24, 110.249.202.0/24, 10.249.201.0/24, 111.225.149.0/24, 111.225.148.0/24