Ubiquiti UDR: How to rollback to the previous image after a failed upgrade

Topology deep dive for a UDR rolling back to the previous good image after a failed push

I run a small ISP / WISP shop and own the Ubiquiti edge for a clutch of Tier-2 town clients between Coimbatore, Madurai and Trichy. The UDR sits in most of those racks, and a UDR rolling back to the previous good image after a failed push is the call I get from the Mysuru jewellery-shop chain backhaul when the morning shift opens the cafe shutters. This guide is what I actually do, not what the marketing PDF says.

In our typical SMB stack the UDR sits two rungs down from the BSNL or Airtel handoff. The fibre lands on an ONT, the ONT trunks into a USG or UDM, and the UDR carries the LAN VLANs to the access points, the IP CCTV NVR, the POS PCs, the Wi-Fi 6 APs in the hall, and the printer behind the billing desk. VLAN 10 is staff, VLAN 20 is POS, VLAN 30 is CCTV, VLAN 40 is guest. The trunk uplink runs LACP across ports 23 and 24, and the management VLAN is 99. Knowing the topology before you touch anything saves the call, every time.

Short version. Console. Power. Network. Firmware. Long version: I once spent four hours on a Hyderabad cantonment cafe roof in February because I skipped the power check and assumed a flaky uplink, only to find a Tier-2 town line that was dipping to 178 V at 14:00 IST when the compressor next door cut in. I now carry a Fluke 117 and a 1 kVA APC SMC1000I-IN line-interactive in the boot of the car. That investment paid for itself the second time.

Configuration walkthrough I actually ship

This is the trimmed-down config I push to a fresh UDR the moment it lands on the SMB rack. Nothing fancy. The point is to get the device adopted, the VLANs trunked, the management isolated and the syslog flowing to the Wazuh box in our Bengaluru NOC before I leave the site. If you cannot hand the cafe owner a one-page paper diagram with the WhatsApp number of our NOC on the bottom, you are not done.

# UniFi OS shell on the gateway / Console role
info
show version
show hardware
show system fan
show system temperature
show system uptime
show interfaces
show interfaces switching
swctrl port show

# Controlled upgrade via the UniFi Network controller API
curl -k -b cookies.txt -c cookies.txt \
  -X POST https://unifi-controller.local:8443/api/login \
  -d '{"username":"netops","password":"..."}'
curl -k -b cookies.txt \
  -X POST https://unifi-controller.local:8443/api/s/default/cmd/devmgr \
  -d '{"cmd":"upgrade","mac":"e0:63:da:11:22:33","url":"https://...bin"}'

# Verify SHA-256 before activating
sha256sum BZ.mt7621_ubnt_unifi.bin
# Compare against the value on https://ui.com/download

# Rollback from controller UI or CLI
set system image rollback
commit
save

Troubleshooting commands by platform

When I open a call on this category I run a fixed sequence so I do not waste site-visit hours. Console first, then power, then network, then firmware. The console cable is a USB-C-to-RJ45 rollover I keep wrapped on the side of my Decathlon backpack. Power is a clamp meter on the C13 rail, then a quick AC sanity check at the wall. Network is a five-minute mirror of port 24 to port 23 on a laptop with Wireshark. Firmware is checked against the controller and the SHA-256 file on ui.com. Every call where I skip a step I regret it within an hour.

UniFi OS exposes most of what you need through info, show version, show hardware, show system fan, show system temperature and swctrl port show. EdgeOS hangs onto the older Vyatta-style show interfaces, show ethernet eth0 phy and the helpful show queueing fair-queue interfaces when you suspect QoS shaping issues on the BSNL uplink. For PoE budget arithmetic I lean on swctrl poe budget and the controller dashboard side by side.

Real codes and errors you will see on the UDR

• show version banner mismatch with the controller is normal during a staged push.
• commit failed: config validation on UniFi OS means a VLAN ID outside 1-4094.

India compliance and deployment notes

MeitY DPDP rules and the RBI Master Direction on outsourcing both want our SMB clients to log who changed what on the network gear, and to keep those logs for ninety days. For our Ubiquiti estate we ship syslog to a Wazuh box in our Bengaluru DC over TLS, retain it on a 1 TB NVMe for ninety days, then push to cold storage on a Hetzner box in Helsinki for the seven-year tail. The controller writes admin audit events as JSON, which makes the Wazuh decoder trivial. For our BFSI clients we add a separate jump-server with TOTP MFA in front of the controller, because their internal IS-audit asks for it. None of this is exotic; it is just the cost of selling network infrastructure to anyone regulated in India in 2026.

Cost-wise, the UDR lands in the SMB Tier-2 zone of the market. GeM tender pricing for a BSNL franchise touched INR 28,900 last quarter; private resellers in Lamington Road ask INR 35-37k. AMC from a Lamington Road reseller usually sits around INR 3,800-5,200 per year for spare-on-shelf and one-week carry-in. Compared with a Cisco Catalyst 1000-24FP-4G-L at roughly INR 1.95 lakh landed for the same port count, or an HPE Aruba 6000-24G-CL4-4SFP at INR 1.42 lakh, the Ubiquiti comes in well under half the BoQ. The trade-off is the support model: with Ubiquiti you are the TAC for your client. With Cisco SmartNet at INR 85,000 to INR 2,00,000 a year, someone else is.

Real-world deployment I did

Last September a Coimbatore textile-trader I support over WhatsApp opened the shop at 09:18 IST and pinged me: the UDR above the billing counter was on a red light, the POS PCs were islanded and the CCTV NVR was screaming because the SIP doorbell could not reach the gate. I asked him to send a photo of the front LEDs. The system LED was solid red, port LED 12 was rapid amber, the rest were dark. That LED pattern, with the Tier-2 town power conditions we have, almost always points to a brown-out event during the morning compressor cut-in. I had him pull the C13 cable for sixty seconds, swap to a spare APC Back-UPS, and bring it up. UniFi OS came up clean, the controller adopted within 35 seconds, and we were back in business before the first jeans-print order shipped. Total down time: 19 minutes. Total cost: zero. I logged it as a brown-out event and quoted him a INR 4,200 line-interactive APC for the rack. He paid for that within the week.

On the UDR the upgrade story is the same every time. A junior tech in the Madurai office pushed an upgrade at 11:40 IST on a Friday without my sign-off, the controller threw a SHA-256 mismatch, and the switch went into a reload loop. I drove back from a Trichy site visit, popped the console cable, caught the U-Boot prompt at 115200 8N1, and TFTP-recovered the stable BZ.mt7621_ubnt_unifi.bin we keep on a USB. Eighteen minutes from console plug-in to controller-adopted. The lesson is the same one I teach every new hire: stage the image, verify the SHA-256, push during a maintenance window, and keep a recovery USB in your bag. Always.

Extended FAQs from real SMB and WISP calls

Will the UDR fit in a 600 mm depth rack at a Tier-2 town SMB?

Yes. The UDR chassis is 290 mm deep, so it sits comfortably in the 600 mm wall-mount cabinets the Lamington Road and SP Road resellers ship. Leave 1U above for the patch tray and 1U below for the P-DU; that gives you cable bend radius for your Belden Cat6 STP.

Does the UniFi controller need to be on a public IP?

No. We run ours on a self-hosted UDM Pro behind a NAT, and bring the remote UDR devices in over WireGuard tunnels back to the Bengaluru NOC. That keeps the controller off the public internet and satisfies the BFSI auditors who refuse to greenlight a SaaS controller.

What is the realistic PoE budget for a real SMB load?

On the USW-24-PoE the total PoE budget is 250 W. In an SMB cafe rack we typically see eight Wi-Fi 6 APs at 13 W average, four IP cameras at 6 W each, one VoIP doorbell at 4 W and one G-IoT sensor at 2 W. That is 134 W of real load, well inside the 250 W budget, with headroom for a future scale.

How long does the UDR actually last in a Tier-2 town power environment?

From my own asset register: median 4.2 years to first fault, longest run 6.7 years on a Mysuru jewellery shop. The two killers are dust ingress on the fan tray and brown-out cycling on the PSU. Line-interactive UPS on the rack adds an honest two years to the MTBF.

Can I run mixed UniFi OS and Cisco IOS in the same SMB rack?

Yes, and we do this all the time. Trunk the VLANs across an LACP between the UDR and the Cisco edge, keep VLAN numbering identical on both, and put the management VLAN in its own SVI on the Cisco side. Just remember that LLDP-MED behaves slightly different across the two, so test your VoIP phone adoption before the cutover.

What is the right escalation path for a hardware fault on the UDR?

Step one is to confirm via console and a known-good PSU. Step two is to file a UI RMA at ui.com/support with the MAC, serial, and a clear photo of the symptom LED pattern. Step three is to buy a spot replacement from a Lamington Road or SP Road dealer if your SMB client cannot tolerate the two-week RMA window. Charge the spot replacement to AMC if you priced AMC at INR 4,200 or higher per year.

Is the Ubiquiti AMC worth it for a small ISP?

For our shop, yes. We pay INR 4,200 per device per year for a Mumbai-based reseller and that buys us same-day swap of a known-good UDR when a client is down. The math works out at about three swaps per year across the estate of forty switches, so the cost per swap lands near INR 56,000, well below the cost of a day-long downtime call for a BFSI franchisee.