When the VPLS instance is on ACX, with one out of 2 child interface in detached state & a single packet is sent, the MAC is relearned from Egress interface in VPLS instance instead of ingress.

We are not seeing this issue when core is on MX.

Customer has a setup wherein they have below design.

Switch1 <-> Core (with VPLS) <-> Switch2

The switch ae bundle will have one out of 2 links as backup & hence the same is in detached state at core.

When switch sent a BPDU packet, it get forwarded to peer switch & then ACX put the MAC as learned from Egress interface, which cause traffic blackholing.

If we sent single ICMP packet, we can see same behaviour & with a 2nd icmp packet, its re-learned correctly from ingress.

It also get cleared after default mac flush timer.

If we remove the detached link from ACX bundle, the issue is not seen.

Lab:
MX80 <-> ACX7348 <-> MX80 (2nd link)

*Lab outputs*

ACX:
labroot@ACX7348-C-EG-re0> show lacp interfaces 
Aggregated interface: ae1
    LACP state:           Role   Exp   Def  Dist  Col  Syn  Aggr  Timeout  Activity
      et-0/0/0           Actor    No    No   Yes  Yes  Yes   Yes     Fast    Active
      et-0/0/0         Partner    No    No   Yes  Yes  Yes   Yes     Fast    Active
      et-0/0/1           Actor    No   Yes    No   No   No   Yes     Fast    Active
      et-0/0/1         Partner    No   Yes    No   No   No   Yes     Fast   Passive
    LACP protocol:        Receive State  Transmit State          Mux State 
      et-0/0/0                  Current   Fast periodic Collecting distributing
      et-0/0/1                Defaulted   Fast periodic           Detached

labroot@ACX7348-C-EG-re0> clear ethernet-switching table 

{master}
labroot@ACX7348-C-EG-re0> clear ethernet-switching mac-learning-log 

{master}
labroot@ACX7348-C-EG-re0> show ethernet-switching table 

{master}
labroot@ACX7348-C-EG-re0> show ethernet-switching mac-learning-log 

{master}

MX:
labroot@MX80> show configuration interfaces irb | display set 
set interfaces irb unit 101 family inet address 10.10.101.1/24

labroot@MX80> show configuration routing-instances test | display set 
set routing-instances test instance-type virtual-router
set routing-instances test interface ge-1/0/2.101

labroot@MX80> show configuration interfaces ge-1/0/2 | display set 
set interfaces ge-1/0/2 vlan-tagging
set interfaces ge-1/0/2 mac ca:fe:00:11:22:33
set interfaces ge-1/0/2 unit 101 vlan-id 101
set interfaces ge-1/0/2 unit 101 family inet address 10.10.101.20/24

labroot@MX80> clear arp 
10.85.161.3      deleted
10.85.161.2      deleted
10.85.161.5      deleted
10.85.161.37     deleted
10.10.101.1      deleted
10.10.101.20     deleted

labroot@MX80> clear arp vpn test 

labroot@MX80>

ACX:
labroot@ACX7348-C-EG-re0> show ethernet-switching table               

MAC flags (S - static MAC, D - dynamic MAC, L - locally learned, P - Persistent static, C - Control MAC
           SE - statistics enabled, NM - non configured MAC, R - remote PE MAC, O - ovsdb MAC
           GBP - group based policy, B - Blocked MAC)


Ethernet switching table : 2 entries, 2 learned
Routing instance : VPLS
    Vlan                MAC                 MAC         Age   GBP     Logical                NH        MAC        RTR
    name                address             flags             Tag     interface              Index     property   ID
    v101                08:81:f4:82:47:e0   D             -           ae1.101                0                    0       
    v101                ca:fe:00:11:22:33   D             -           ae1.101                0                    0       

{master}
labroot@ACX7348-C-EG-re0> show ethernet-switching mac-learning-log    
Fri Jan 30 14:09:29 vlan_name v101+101 mac ca:fe:00:11:22:33 was learned on et-0/0/2.101 with flags: 0x2001f
Fri Jan 30 14:09:29 vlan_name v101+101 mac ca:fe:00:11:22:33 was moved from et-0/0/2.101 to ae1.101 with flags: 0x2001f
Fri Jan 30 14:09:29 vlan_name v101+101 mac 08:81:f4:82:47:e0 was learned on ae1.101 with flags: 0x2001f

Cleared after mac flush:
Fri Jan 30 14:14:06 vlan_name v101+101 mac ca:fe:00:11:22:33 was deleted from ae1.101 with flags: 0x1000
Fri Jan 30 14:14:29 vlan_name v101+101 mac ca:fe:00:11:22:33 was learned on et-0/0/2.101 with flags: 0x2001f

labroot@ACX7348-C-EG-re0> show ethernet-switching table               

MAC flags (S - static MAC, D - dynamic MAC, L - locally learned, P - Persistent static, C - Control MAC
           SE - statistics enabled, NM - non configured MAC, R - remote PE MAC, O - ovsdb MAC
           GBP - group based policy, B - Blocked MAC)


Ethernet switching table : 2 entries, 2 learned
Routing instance : VPLS
    Vlan                MAC                 MAC         Age   GBP     Logical                NH        MAC        RTR
    name                address             flags             Tag     interface              Index     property   ID
    v101                08:81:f4:82:47:e0   D             -           ae1.101                0                    0       
    v101                ca:fe:00:11:22:33   D             -           et-0/0/2.101           0                    0   

This was replicated in lab & was analysed by Engineering via internal PR 1934724.

The behaviour flagged in this PR is expected by current design from both Broadcom and PFE perspective and this is day-1 behaviour in ACX7k.

We are planning to change this behaviour to follow mx but this will be a behavioural change and code churn will be high.

According to ACX7K design, a LAG/Lag members with asymmetric LACP configuration (one side LACP, one side non-LACP) will not block the ingress/incoming traffic.

Broadcom has a limitation in which it can block only egress traffic for a lag member. So, only egress traffic is blocked. This will be applicable for all of the ACX varieties.

So for the customer suggestion in this case is: to have symmetric LACP configuration on both of the peers.

We can resolve this issue by deleting the members from lag bundle(only in relevant cases), but this will take good amount of design and test cycles, we are planning commit this only in master branch.