There is probably a better subject title, as this issue is not the problem but only a symtom we have had with ePMP from software versions 2.52 thru 2.61.
We use Rancid to pull the configs from the ePMP's on our network. Over time, more and more ePMPs will stop responding to cli command "config show dump". This is brought to my attention by the config diff's email which shows the entire config being deleted from CVS, and maybe a subsequent polling will retrieve the config.
If you ssh into the radio immediately after noticing it failing to supply the config data and reboot it, the unit will be fine for a while, anywhere from a week to a few weeks when the symtom will return.
If you ignore the symptom, eventually you will not be able to login at all, requiring a power cycle to restore proper operation.
So far, I have not determined if this affects the wireless connectivity or data flows.
My thoughts are something is corrupting memory, or causing the snmp interface to not respond, eventually affecting the SSH service as well. Currently, this morning, I have ten devices which do not respond to ssh but still are pingable.
Also, it seems to not be every device, but the same bunch over and over. This affects AP units and SM units, as well as a few setup as PTP links. As common as this is on our network, I have not seen or heard any talk of other operators with this issue.