I think I found out that my PPPoE client become unstable because of ether1 spike translate to sfp-sfpplus1 after these update:
ROS 7.19
*) ethernet - fixed flow-control for RB5009;
ROS 7.19.1
*) switch - fixed advertise and speed settings for ether1 on RB5009 (introduced in v7.19.1);
my internet connection: pppoe-out1
vlan500
sfp-sfpplus1 and has no spike
sfp-sfpplus1 not part of bridge member. ether1 connected to 2.5G switch.
As you can see, some traffic from ether1 get send directly to sfp-sfpplus1 despite not part of bridge1 and it flood the 2.5G ONT which cause pppoe-out1 Tx Drop (sessions disconnect)

