Skip to main content
Revetment
Install

The evidence behind the homepage

What our own production shows, and how we counted it.

Every production figure on this site comes from one snapshot of our own cluster, taken by hand on . This page prints the whole snapshot, the command behind each number, and what each number does and does not count.

One cluster. Read every number at that size.

This is one DigitalOcean Kubernetes cluster with six nodes, running eight of our own services across four namespaces. It has 23 HTTPRoutes. 18 of them carry a RevetmentPolicy, and all 18 are blocking. That is the whole estate. It is enough to show that Revetment runs in blocking mode on real traffic without taking down the services behind it. It is not a claim about your estate, which has different routes and different traffic.

23 HTTPRoutes · 18 with a policy18 blocking · 0 in shadow
Fig. 1Every HTTPRoute on the cluster. A ticked, stippled square carries a policy in blocking mode; a dashed square has no policy and is not inspected.

The policy ledger

18 policies. 18 blocking. 0 in shadow.

Snapshot taken 17 November 2026. One row per workload; hostnames and route names left out on purpose.
WorkloadModeAccepted
Email platform (SMTP bootstrap, mailbox and media routes)blockingTrue
DNS hosting APIblockingTrue
Monitoring and alerting serviceblockingTrue
Search serviceblockingTrue
Bot and spam gateblockingTrue
Document and inbox appblockingTrue
Banking demo with three routesblockingTrue
The Revetment console itselfblockingTrue
18 policies · 18 blocking · 0 in shadow · 18 accepted

The whole snapshot, field by field. The cluster's name is in the file and not on this page.

Policies
18
Blocking
18
In shadow
0
Accepted
18
Namespaces
4
HTTPRoutes
23
First policies applied, in shadow
2026-09-26
Last policy switched to blocking
2026-10-24
Edge pods
2
Edge pod restarts
0
Days since the pods last rolled
19
Requests through the edge since then
24,920,379
Edge image, short id
d1d389f5
Nodes
6
Kubernetes
1.36.5
Envoy Gateway
1.8.3
Snapshot taken
2026-11-17

Shadow to blocking

28 days from shadow to blocking.

The first policies went on in shadow mode on 26 September 2026. Shadow mode logs what a policy would have blocked and lets the request through. We read those logs, cleared the false positives we found, and switched the last route to blocking on 24 October 2026. Nothing has gone back to shadow since.

First policies, in shadow2026-09-26All 18 blocking2026-10-24This snapshot2026-11-17First policies, in shadow2026-09-26All 18 blocking2026-10-24This snapshot2026-11-17
Fig. 2Policies in shadow from 26 September 2026, all blocking from 24 October 2026, snapshot taken 17 November 2026. Each tick is one day.

Method

The command behind every number.

We took the snapshot by hand on 17 November 2026 and committed it to the site's source as one JSON file. The pages are built from that file. Nothing is fetched when you load them. When the figures change, the file changes, the date changes with it, and the site is rebuilt. If a number here is wrong, the file is wrong, and you can hold us to that.

Policy counts and dates

As recorded in the snapshot

kubectl get revetmentpolicies -A -o json
mode, status.conditions[Accepted], lastTransitionTime

What it counts: every RevetmentPolicy in every namespace of the cluster. The mode field says whether a policy is blocking or in shadow. Accepted is the controller confirming the policy is attached to its HTTPRoute. The two dates are read from lastTransitionTime: the earliest for when the first policy went on, the latest for when the last one switched to blocking.

policies.total · policies.blocking · policies.shadow · policies.accepted · policies.policied_since · policies.blocking_since

Requests, restarts and days up

As recorded in the snapshot

Envoy admin /stats/prometheus on both edge pods, envoy_http_downstream_rq_total summed
kubectl get pods -n ingress-nginx for restarts and start time

What it counts: envoy_http_downstream_rq_total, read from the Envoy admin endpoint on each of the two edge pods and added together. That is every request that reached the edge on any route, policied or not, since the pods last started. It is not the number of requests Revetment inspected, and it is not a count of blocks. The restart count and the start time come from the pod list; the days-up figure is the time since the pods last rolled. A roll is a deploy, not a crash.

edge.pods · edge.restarts · edge.days_up_since_last_roll · edge.downstream_requests_since_last_roll · edge.image_short

Recorded without a command

The HTTPRoute total, the namespace count, the node count and the two versions are in the snapshot but have no command written beside them. They came from the same session at the same cluster. The next snapshot should record how.

policies.httproutes_total · policies.namespaces · cluster.nodes · cluster.kubernetes · cluster.envoy_gateway

REQUESTSevery route,policied or notEDGE POD 1EnvoyRevetmentmoduleEDGE POD 2EnvoyRevetmentmoduleCOUNTER READ HERE23 HTTPROUTES18 carry a policy18 blocking0 in shadow5 without a policypass through untouched8 WORKLOADSour own services,hostnames left out24,920,379 requests since the pods last rolledREQUESTSevery route, policied or notCOUNTER READ HEREEDGE POD 1EnvoyRevetmentmoduleEDGE POD 2EnvoyRevetmentmodule24,920,379 requestssince the pods last rolled23 HTTPROUTES18 carry a policy18 blocking0 in shadow5 without a policypass through untouched8 WORKLOADSour own services, hostnames left out
Fig. 3Where the counter sits. Both edge pods count every request they receive, on all 23 routes, before any policy is consulted.

What this doesn't show

What this page doesn't show.

  • No latency figures. There are none anywhere on this site. That is a choice, not an oversight. A benchmark we ran on our own rig, judged by rules we wrote, is not something you can check, so we leave it out.
  • No counts of blocked requests. The snapshot has no count of requests blocked or that would have been blocked, so this page shows none. The request total above is traffic through the edge, not a verdict count.
  • One small estate. Six nodes and 23 routes is small. Your routes, your traffic and your false positives are different, and the only count worth your time is what shadow mode logs on your cluster. The way to get it is the same way we did: install, watch, then switch on.

What a WAF at this position cannot do is a separate page: What Revetment can't stop.