MikroTik monitoring from GEN

The cable started to fail at 10:12.
You replaced it at 17:30.

Oversight reads RouterOS routers and switches through REST and SNMP: every port's traffic and errors, the hardware's health, and anything else the API can report. When a link starts to go bad, you get one message that names the port, before anyone rings to say the network is down.

ALARMNetwork warning · 2 new 10:13 UTC
Head office                       WARN
└ crs354-core                     WARN  1m
  ├ Errors 1-48  ether14 +1420    WARN
  └ Link flaps  ether14  2        WARN
Before it becomes an outage

What 10:12 looks like.

10:12A damaged patch lead on ether14 starts throwing bad frames and drops its link twice. Traffic carries on, so nobody rings.
10:13The next read finds 1,420 more errors on ether14 than the last. The rule is set at 100, so the port goes amber.
10:13One message names the switch, the port and the count, with the link flaps beside it.
10:17The syslog sender's next upload carries RouterOS's own line, ether14 link up (speed 100M, full duplex), from a port that normally runs at 1G.
10:20An engineer opens the chart, sees the errors begin at 10:12 and books a swap for after hours.
17:30The lead is replaced. The errors stop, the port is back at 1G, and the recovery goes to the people who were told.
An illustrative day, on one-minute reads. At five minutes the alarm could be up to five minutes later. The limits are settings: up by 100 for amber, one link flap for amber.
What Oversight reads

Every port, and anything else RouterOS can tell you.

48 ports, two sensors

REST answers for many ports in one request, so one sensor reads traffic on up to 48 ports and another reads their errors, each port with its own chart and rules. Errors follow the factory port name, so renaming a port cannot quietly read the wrong one.

The MikroTik guide

Health

CPU and board temperature, voltage, power supplies as ok or fail, CPU load, fan state, PoE ports, SFP optics and LTE signal. The MikroTik MIB is already loaded, so SNMP objects can be picked by name.

Which route for what

Anything the API can return

A sensor can ask for any REST path, extract what it wants from the reply and put rules on it: BGP sessions established, OSPF neighbours Full, a WireGuard peer's last handshake, an IPsec peer. If RouterOS can report it over REST, it can alarm.

Extraction and rules

A probe on the router

A capable router can host the probe itself, in a RouterOS container, so a site needs no separate machine. It takes arm64 or CHR, RouterOS 7.21 or later and 1 GB of RAM: CCR2004, CCR2116, CCR2216 and RB5009 are candidates, CRS switches and small routers are not. Setting up takes two reboots, and switching on container mode needs a button press or power cycle at the router, or from the hypervisor on CHR, so it cannot be done entirely remotely. It goes down with the router, so that router and its internet connection still need a probe outside.

What it needs: REST is RouterOS 7 only, and SwOS has no REST API. RouterOS 6 is watched through SNMP. MikroTik's binary API on ports 8728 and 8729 is not used.

Early warning

The signs come before the outage.

A link rarely fails without notice. These all run on the same platform, with the same rules, messages and escalation.

Links about to fail

Bad checksums, runts, jabber and collisions climbing on a port, a link that keeps flapping, and an SFP's received signal falling away from the level it was commissioned at.

Sensors worth having

Adverse traffic

A rule can catch an uplink falling by 90% from one read to the next, or a port running at its limit. Traffic is a one-second sample at each read, which is right for those and wrong for capacity planning or billing.

How rules work

A slow drift

Switched on for a sensor, behaviour analysis compares it with its own history and flags one going somewhere new, or somewhere odd for the time of week. It flags and never alarms.

Behaviour analysis

Syslog, offloaded

RouterOS logs to a syslog sender at the site. Rules there work from its topics, drop the firewall chatter and pass on link changes, reboots without a clean shutdown and probably a loop warnings, so only those leave. Lines can be counted per device and trended.

The syslog sender

One alarm, not forty

Every view of a sensor votes, from probe groups in different places. Make what sits behind an uplink depend on it, and when the uplink fails it alarms once, while the checks that can no longer reach their target are held rather than alarmed.

How dependencies work

Actions

Email, Matrix, a ticket, or a webhook to any URL, so your own script can fail over to another uplink. If nobody acknowledges, the alarm climbs after the minutes you set, as far as a phone call read aloud.

Rules and channels and escalation

Pay for what runs

A 48-port switch, every port and its health watched, from about £1.28 a month (roughly $2 or €2*), with ports polled every five minutes in working hours.

No per-sensor licence, no tiers and no prepayment. Invoiced monthly in arrears, in GBP. *Estimated: other currencies vary with the exchange rate.

From a probe on site, for a monthCost
Traffic on all 48 ports, one REST sensor, every 5 minutes in working hours£0.41
Errors on all 48 ports, one REST sensor, every 5 minutes in working hours£0.41
Two health sensors, REST, every 15 minutes round the clock£0.46
  • Working hours here are 10 hours a day, 5 days a week. A schedule stops the polling, so nothing is charged outside it.
  • To see a cable failing at 22:00 by 22:01, poll every minute round the clock: the two 48-port sensors are then £13.82 a month.
  • In Oversight, by SNMP at one value per sensor, the same 48 ports with traffic and errors each way would be 192 sensors and about £332 a month at that rate, which is why REST is the route to use here.
  • Each further sensor every 15 minutes, round the clock, is £0.23 by REST or £0.12 by SNMP.
  • Rules, escalation, email, webhooks and tickets are included. SMS and voice are metered separately.
  • Rather GEN set it up? See services, from £49 an hour.

Find out from Oversight, not from your users.

Register as a client and an account manager will walk you through the early steps, with no commitment at all. Or talk to an expert first.

The same platform watches the rest of the estate too, from Proxmox and Ceph to Linux and Windows servers and email. See what else it covers.