radiusward authenticates Wi‑Fi, VPN and subscriber sessions against your users file, SQL database or Active Directory — and stores every session record on three servers before it answers, so a crash never loses a record or counts one twice.
$ docker compose -f deploy/compose/compose.yaml up -d --wait
✔ node-0 node-1 node-2 gateway-a gateway-b postgres healthy
$ deploy/compose/demo-failover.sh
load: 10,000 sessions, 2,000 requests/s for 25 s
kill -9 of the leader: node-2
new leader: node-0
verifying every acknowledged event against the replicated state...
sessions=10000
lost=0Your invoice is built on session records. A RADIUS server outage, a lost Stop, or a retransmission counted twice turns into a billing dispute. radiusward keeps every record on three servers and counts each session once — through outages and failovers.
Wi‑Fi and wired 802.1X, VPN: certificates or passwords against Active Directory, LDAP or SQL, a second factor if you want one. Logins keep working when a server is lost, and the records of who was online when are not.
One container image, a Helm chart tested every night under failures, Prometheus metrics with dashboards and alerts, a continuous backup — and no appliance. Runs on Kubernetes, plain hosts, or Amazon ECS.
A device sends its session records to the RADIUS server and retries for a while. If the server is down for longer than that, the records — the usage you bill — are gone.
radiusward: a record is confirmed only after it is written to disk on two of three servers. A crash of any one of them loses nothing that was confirmed.
A device that gets no answer in time sends the same record again — often to its secondary server. A server that does not recognise the retry stores the session twice and inflates the usage.
radiusward: retries are recognised by session, not by server — also after a failover, also through the other gateway — and counted once.
With a primary and a secondary server, the Start lands on one and the Stop on the other. Reconciling them afterwards is where the ghost sessions and the billing errors come from.
radiusward: all gateways write to the same replicated state, so a session is one session wherever its records arrived. A failed leader is replaced in about two seconds.
radiusward is young: a release candidate, automated tests on every change, and published results — not yet a production track record, device certifications or an independent security audit.What the evidence shows, and what it does not →
Apache-2.0. The code, every design decision, the tests and their results are public — and everything on this page can be reproduced on your laptop.