Eternita Edge · intelligence
The edge sees every request before your app does, so it is where you learn what your traffic is: who is a crawler, who is a customer, which rule acted and on what — and where your team gets told, with the detail no metric carries.
# conf/bots.conf — name the visitors you care about
@type customer like guest no-store
@type bot-verified like bot
customer | cookie:JSESSIONID ; request-type:toplevel | customer
googlebot-verified | ua~(?i)googlebot ; client-ip:@lists/googlebot.txt | bot-verified
# Prometheus: the share of pages served to each type
sum(rate(frontcache_visitor_requests_total[5m])) by (type)
Declaring a type changes nothing until something names it — a TTL, a rule, a fragment — and needs no cache flush. It is counted from the first request.
Send us an access log and we tell you what share of your traffic is crawlers and how much of it is cacheable. Free; we reply within one business day.
Who Is at Your Edge
Every request is classified before anything else happens — by User-Agent, address, cookies, headers and path. Bot and guest are built in; up to 16 more are yours, each behaving like one of the two everywhere you do not name it.
A type is evidence, not a label. A verified crawler is one from its vendor’s own network; a customer is one with a session; a visitor who passed a challenge holds a pass signed by your node. Decide a type on what a client cannot forge, and every rule built on it holds.
Then use it. A TTL per type in the response header, a guard rule per type, a fragment
only one type sees — and no-store, so a customer’s page is never cached for
anyone else.
# set by your app on a page or a fragment
x-frontcache-component-maxage: bot:30d, guest:1h, customer:0
Crawlers from cache for a month, guests for an hour, customers always fresh.
A Visitor Types screen with each type and the hits of the rules deciding it; the Cache View shows an entry’s expiry per type.
A type label on the visitor metric, and the type as the last column of every request
log line.
What Happened
Everything below is in the Apache-2.0 engine. Each channel answers a different question, and none leaves your infrastructure unless you send it.
Prometheus at /fc-metrics, or OTLP push to a collector, Grafana Alloy or your vendor.
Hit ratio by tier, latency, circuits, rule matches, visitors by type. Nothing is labelled by URL
— deliberately, so a scrape can never take your Prometheus down.
A line per request and per fragment — cache status, latency, bytes, client, visitor type — plus logs of refusals, errors and fallbacks. The only per-URL channel, with a ready-made Elastic setup and four dashboards in the repository.
The console’s realtime page: request rates and circuit state per node as they happen, cache inspection and invalidation. And events after the fact — rule matches, rate limits, origin errors — for any plugin to act on.
Pro Alerts
From the node itself, with the detail a metric cannot carry: which rule, which URL, which address — hashed unless you choose otherwise. To Slack, Microsoft Teams, PagerDuty, Opsgenie, a webhook (optionally signed), mail or syslog.
Three shapes of rule. Every matching event, with a cooldown; a threshold over a window — including a challenge pass rate that falls; or a state that lasts, resolved when it ends. On rule matches, rate limits, circuits, origin errors, fallbacks, config reloads, invalidations and challenges.
Never in a visitor’s way. Alerts are sent from the node’s own threads, never
on a request, and only to the sinks your file names; a web sink must be https. Nothing happens
until you write a rule; edit it in the console and it applies without a restart.
# conf/alerts.conf (Pro)
alerts {
sinks {
ops-slack { type = slack, url = ${?FC_ALERT_SLACK_URL} }
}
rules = [
{ name = search-flood, on = guard.summary,
rule = search-limit, measure = rejected,
threshold = 500, window = 5m, severity = warning,
notify = [ops-slack], cooldown = 15m }
{ name = origin-down, on = circuit.changed,
command = Cache-Miss, to = open, for = 1m,
severity = critical, notify = [ops-slack],
resolve = true }
]
}
A search rule refusing more than 500 requests in five minutes is a warning; the origin’s circuit open for a minute is critical, and resolves itself when it closes. The webhook URL stays in the environment, out of the file.
Pro The Fleet View
What a team running several nodes and several sites needs on top of one node’s view.
Each node’s card on the realtime monitor shows what its whole cluster is taking, beside what the node itself is — added up in your browser, at no cost to the nodes.
Tenants and sites, each site called with its own API key, each user seeing and acting on only the tenants granted to them — with a role on each.
Several users with roles, single sign-on, and a tamper-evident audit log of who did what, from where.
Free and Pro
Everything that describes your traffic is in the Apache-2.0 engine. Pro adds alerts and the views a fleet needs — $3,600 per node per year, 2-node minimum.
| Capability | Community — free | Pro |
|---|---|---|
| Visitor types | Bot, guest and up to 16 declared types | Included; the cache warmer can warm as a declared type |
| Metrics | Prometheus and OTLP export | Included |
| Logs and events | Request, refusal, error and fallback logs; events for plugins | Included |
| Realtime console | Per node, single-user sign-in | Plus cluster request rate; users with roles, SSO, audit log |
| Alerts | Alert on the metrics in your own tools | Slack, Teams, PagerDuty, Opsgenie, webhook, mail, syslog — with rule, URL and address |
| Many sites in one console | One console for all your nodes, an edge with a key of its own if it needs one | Tenants and sites, each user seeing only the ones granted to them |
What It Is Not
Know where the line is before you buy anything.
“Intelligence” here means knowing what happens at your edge. A visitor’s type comes from rules you can read, not a model; there is no risk score and no fingerprinting.
No visitor profiles, no tracking cookie, no sessions stitched together. It counts requests by type and logs them on your servers; your analytics tool stays where it is.
PagerDuty and Opsgenie fold a fleet’s alerts into one incident; a chat channel gets one message per node, each naming its node.
Next Step
Send an access log for a free first look at your crawlers and visitors — or book 45 minutes with an engineer, $200, credited toward an Assessment or Pro.