Changelog
Changelog
One entry per release, written for the people who run the panel.
-
0.2.32
Card series simplification
- Card generation works from the series alone. The separate card-profile layer is gone, and the settings it held — plan, validity, price and design — now live on the series itself. One object to create, one object to look at.
- Counter sales mint cards directly into the clerk's own series, created on the first sale of the shift.
-
0.2.31
Maintenance state is stamped from the push
- The time a maintenance change was applied is recorded from the push itself rather than from the next status poll, so Control shows the instance's real state immediately instead of up to one poll late.
-
0.2.30
Maintenance mode applies as soon as it is pushed
- Maintenance mode takes effect the moment Control pushes it, rather than at the instance's next poll.
- The panels check whether the instance is back before reloading, so a maintenance window ends with the page returning rather than with an error the operator has to refresh past.
-
0.2.29
Management network follow-ups
- The instance list reports the path a call actually took, rather than the path it expected to take before dialling. A prediction and a result are not the same fact, and only one of them is worth showing.
- Disabling the overlay for several instances at once is never refused part way through; the operation reports per-instance outcomes instead.
-
0.2.28
A private management network between Control and instances
- An opt-in private network between X-Radius Control and each instance. Calls from Control prefer that path and fall back to the public HTTPS endpoint automatically when it is unavailable, so a broken overlay degrades rather than cuts management off.
- The overlay's state is visible per instance, including which path the last call actually used.
-
0.2.27
Counter-sale slips, five portal designs, instance backups
- Counter sales print a slip whose design you choose, with a separate default per card use, and every printed slip carries the identity of the card it belongs to.
- Scheduled, encrypted instance backups to a Google Drive you connect, managed from Control, with a recovery drill you can run.
- Maintenance mode per instance, set from Control.
- Five redrawn subscriber-portal designs replace the previous four. The default is the X-Radius design; the others are picked per tenant.
-
0.2.26
ZeroTier as an operator VPN mode
- ZeroTier as a tunnel mode for the operator VPN, with the controller running on your own instance rather than on a third-party service. It covers the routers where the other modes are not available on the hardware, notably RouterOS 7 on ARM boards.
- The panel prepares the package the router needs and reports when the device is missing it, instead of failing at the end of the onboarding script.
-
0.2.25
Remote access over the relay port the instance already uses
- Remote access to a router from the panel can share the relay port the instance already listens on, rather than needing a port of its own. For an operator behind a restrictive firewall that is one fewer opening to request and one fewer thing to explain.
-
0.2.24
Two operator VPN modes on the same network
- The operator VPN is chosen per network device instead of once per instance. A network can now run WireGuard on the routers that support it and OpenVPN on the ones that do not, at the same time, with both terminating on the same instance.
- The generated onboarding script picks the tunnel that matches the device's RouterOS version.
-
0.2.23
Licence-weighted resource share
- A resource share on the licence and on the package it came from, editable in Control. It scales the per-tenant database gate and request budgets introduced in the previous release, so a larger customer on a shared instance gets a proportionally larger slice rather than the same slice as everyone else.
- The share is applied live: changing it in Control takes effect on the instance without a restart.
-
0.2.22
Resource isolation between tenants on one instance
- A per-tenant limit on concurrent database work, applied per request rather than per connection.
- A deadline on every request and a statement timeout on every query, so a slow query fails instead of queueing behind itself.
- A maximum page size on every list, so a client cannot ask for a million rows.
- Per-tenant ceilings on background job depth and on the live RADIUS log's event rate.
-
0.2.21
Correctness pass across billing, sessions and the API
- Card reversals always leave a balanced ledger. Two paths that could credit a wallet without a matching debit are closed.
- Sessions that move a very large amount of traffic are now metered from the full counters the router reports, so a long or fast session is billed for what it actually used.
- Session correlation is bound to the network device that was verified for the request, so an accounting packet can no longer be attributed to the wrong tenant.
- Permission checks around administrator roles were tightened in two places.