$job has a bot problem, (or maybe a shitty backend problem), cloudflare bot management to the rescue?
There are typically 3 types of resources we serve: page, static resource, api endpoint. A page is the initial request for a user, this could be a static html page or some dynamically generated one. Static resources are usually served from a CDN and api endpoints serve both the web frontend and mobile apps. What's important is that for pages, we can serve challenges (cloudflare making you wait a bit or making you solve a captcha). For static resources, we probably don't care that much, and for api endpoints, because the requests are made programmatically, there's not much we can do besides allow/block.
Cloudflare has quite a few blocking products that sit in front of their CDN, workers, and your servers. The order they happen in can be seen:
So why are the existing products not enough? The existing tools (IP, user agent, WAF, rate limiters...) are like scalpels, precise for their functions, but not very flexible to changes. Ex: IP and rate limit blocks are easy to get around by just spinning up new VMs in clouds, or if you really want (and don't mind IPs from dubious sources) with a large pool of proxies.
Sure, with enough investigation, you can tailor blocks using the existing tools to more or less only block specific attacks, but that's a lot of work for a fragile solution, and you inevitably end up with a bunch of rules that are no longer relevant as script kiddies move on.
If the existing tools were fishing rods and bait, bot management is like a dragnet. Requests are classed by heuristics, machine learning (they refuse to say what sort of machine learning is fast enough to run on every request), and behavioural analysis, and you get a 1-100 score you can use as part of your firewall rules.
Where previously your rules are more building up ORs of potentially bad sources,
ex (cf.client.bot is available to everyone as verified good bots):
1(
2 (
3 ip.geoip.asn in { xxxx yyyy zzzz }
4 or http.user_agent contains "fuzz"
5 or ip.src in { "a.b.c.d/32" "q.r.s.t/31" }
6 )
7 and not cf.client.bot
8 and http.host eq "example.com"
9)
With bot management it is more digging out exceptions for known good sources:
1(
2 cf.bot_management.score eq 1
3 and not cf.bot_management.static_resource
4 and not cf.bot_management.verified_bot
5 and not ip.src in $office
6 and not ip.src in $cloud
7 and not ip.src in $trusted_third_party
8 and not http.user_agent contains "company"
9 and not http.user_agent contains "seo.tool.we.use"
10 and http.host eq "example.com"
11)
With terraform it's slightly better since we can use variables and string interpolation to manage a bunch of rules, but it doesn't help with investigating all those potentially legitimate requests you might block
1resource "cloudflare_filter" "ex" {
2 # ...
3 expression = <<-EOF
4 (
5 cf.bot_management.score eq 1
6 and not cf.bot_management.static_resource
7 and not cf.bot_management.verified_bot
8 ${join("\n ", formatlist("and not ip.src in %s", local.known_ip_lists))}
9 ${join("\n ", formatlist("and not http.user_agent contains %s", local.known_user_agents))}
10 and http.host eq "example.com"
11 )
12 EOF
13}