I took all the "masquerade" and put the netmap but the problem still remains, when they disconnect more than 500 pppoes it seems that the cpu processing increases, I forgive access to CCR1036 and when these 500 pppoe are connecting again at the same time the cpu increases and the CCR1036 is unavailable until all pppoes go up, I removed all the "masquerade" rules, but the problem remains.

Traffic also decreases a lot, traffic in v4 is zero.

Set MPLS to "no" in the PPP profile

This should resolve your issue.