I observe a strange situation with the learning of mac addresses.
Scheme is attached.
Fact:
If any host from internet (like 2.2.2.2) try setup tcp session with my host (1.1.1.1) - my switch (QFX10002-1 and QFX10002-2) does not learn the address from the downstream switch EX4550.
In result "switch" downgrade to "hub" and "broadcast" traffic 2.2.2.2 -> 1.1.1.1 to all ports.
If I in my router MX480 try:
1) ping 1.1.1.1 from router
or
2) clear arp hostname 1.1.1.1
admin@QFX10002-nl-1> show ethernet-switching table | grep 90:b1:1c:30:3b:1e
vlan_350 90:b1:1c:30:3b:1e DR - ae3.0 0 0
admin@QFX10002-nl-2> show ethernet-switching table | grep 90:b1:1c:30:3b:1e
vlan_350 90:b1:1c:30:3b:1e DL - ae3.0 0 0
Ok. mac here.
admin@QFX10002-nl-2> show configuration protocols l2-learning
{master:0}
admin@QFX10002-nl-1> show configuration protocols l2-learning
{master:0}
MX480:
start shell
% sysctl -a | grep arp_cache
net.link.ether.inet.arp_cache_size: 2565
net.link.ether.inet.arp_cache_perm_size: -365
net.link.ether.inet.arp_cache_size_threshold: 0
net.link.ether.inet.arp_cache_timeout_size: 2034
net.link.ether.inet.arp_cache_rearp_size: 0
net.link.ether.inet.arp_cache_retry_size: 170
%
If any other details are needed, then I will write. How to properly troubleshouting and fix it.
