Maybe it is a combination of issues:
-
Not seeing any traffic when the gateway came back online: the devices might have given up due to not receiving any confirmations (if that’s true: yet another reason to not use confirmed uplinks), or when trying to re-join a lot during the two weeks that the gateway did not work?
-
Not being able to join after the reset: are these LoRaWAN 1.1 devices? LoRaWAN 1.1 uses an incremental value for the DevNonce, and devices should always store the last value they used. If the reset made them (erroneously) start at zero again, then TTN would ignore the first join attempts, until at some point a value is used that TTN has not seen earlier.† However: (earlier) TTN would show an error then; see OTAA shows "Activation DevNonce not valid: already used".
The first bullet point might not be relevant. How much time did you wait to see if data came in after the gateway was back online? At some point a proper device will back off in its join attempts, so a long time might pass between two tries.
But re-registering a device does remove the history of DevNonces, so if the issue at hand is related to that, then I’d assume a 1.1 device (or a bad 1.0.x device). You could easily verify by resetting a device again, and/or see the DevNonce using an online decoder.
As an aside: in the first post you wrote “some devices”, in the last “the device”. This applied to multiple devices, right?
† The community network does not support 1.1 yet, but the 1.0.x and 1.1 Join Requests are the same, except that in 1.0.x the DevNonce is a random value which makes the chances for using the same value much smaller. After receiving a 1.0.x Join Accept, a 1.1 device will fall back to 1.0.x.