I have CRS 320-8P-8B-4S+RM switch connected to Eaton 5E1200UF UPS. As soon as the UPS starts providing power from the battery, the switch starts logging the following warnings:
(content after // is not a value from the log, it’s my comment on the log line)
PSU1 power input not detected
etherX detected poe-out status: controller_error // (X is port number); repeats for all ports
etherX detected poe-out status: controller_init // (X is port number); repeats for all ports
PSU2 returned to state OK
PSU1 returned to state OK
etherX detected poe-out status: controller_error
...
And this repeats until I restore power to UPS.
I tried disabling POE on all ports, nothing changed. I monitored UPS’s load and it’s never above 6%. Besides the switch, UPS also powers a NAS and a TP Link poe injector, which then powers cAP AX. The injector was introduced to the structure because POE-powering cAP AX through the switch resulted in “sometimes” facing “short_circuit” errors on the POE-powered port (and that also happened when the switch was directly powered from the wall outlet, not the UPS - I tried all combinations). I suspected PSU issues with the switch, because of which I introduced the POE injector and disabled POE. But then I faced this issue with the UPS and I think this shouldn’t be expected.
That error makes me suspect a zero-crossing problem with the UPS’s on-battery output voltage. I go into considerable detail on that problem here, but the bottom line is that yes, some PSU designs do care about this, and no, it doesn’t constitute a design error in either the UPS or the powered device. It’s simply a consideration to take into account when choosing the right UPS for your application.
Mind, this is speculation since I don’t have either that switch or that UPS here to test. My linked article gives you the details you need to check the guess yourself.
The Eaton 5E1200UF is a line-interactive UPS, meaning it has a small transfer delay (around 2–6 ms) when switching from AC power to battery power. During this brief period, the power supply is not fully uninterrupted.
Because of this, the CRS320 might briefly lose power on the affected PSU input. When that happens, PoE-Out will reset—this is normal behavior. The PoE output depends on constant input power, so any interruption, even for a millisecond, will cause a PoE reset.
If your CRS320 has dual power supplies and each is connected to a separate, independent power source, the device will keep running without any issues. As long as at least one PSU remains powered through the switchover (for example, if one UPS switches to battery and the other remains on), the PoE output will not reset.
For setups where uninterrupted power is critical, especially if you’re relying on PoE-powered devices, we recommend using a true online (double-conversion) UPS, which ensures continuous power without any transfer delay.
While that is true, most PSUs have enough rail capacitance that this delay results in a small dip in the DC output voltage, not a true interruption. You would have to test this under load to be certain, but few devices will even see this dip, much less respond improperly to it.
Contrast the 0 AC crossing gap I brought up: it is a real thing and can cause real issues, because it happens 100 or 120 times per second, depending on whether you’re on 50 or 60 Hz power.
You’re right — many PSUs, including those used in the CRS320, have internal capacitance that helps them ride through very short power dips. As a result, the CRS320 itself does not shut off or reboot during the typical switchover delay of a line-interactive UPS.
However, the PoE-Out functionality is intentionally reset during such transitions, as a safeguard to ensure stable operation. This is expected behavior by design.
Thank you all for your help! (Sorry about the delay in answering – email notifications were disabled by default!)
Awesome technical explanation of the problem, tangent! I must be honest that I don’t [yet] fully understand everything outlined in the comment you linked on the SuperUser forum, but I’m getting there.
As inra also wrote, I think the conclusion here is that a true online (double-conversion) UPS is needed. The switchover likely is causing some problems, but I’d assume those would be solved shortly after the switchover is complete, while my issues continue until power is restored (and a pure sine wave is back).
I’ll have to look into getting a smaller (cheaper) online UPS just for the switch now. Not sure when I’ll get my hands on one, but I’ll make sure to post here once I test it.
Note that true-sine and double-conversion are independent features. You can have all four possible combinations. This means you still have an experiment to run here before you know which of us has guessed correctly about the source of your problem.