Customer experienced a MAC/IP issue between two QFX leaf switches. An ARP entry was missing from one leaf but present on the other leaf. We had to clear the ARP entry off of the leaf that had the entry to get it to readvertise the ARP entry and cause the leaf without the ARP entry to trigger an update event, which fixed the issue. We believe interface flaps or downstream reboots are triggering an issue with learning at times.
============================Leaf A=======================================
root@switch> show log messages | match ae6 | match updown
Apr 22 04:52:04 switch rpd[11077]: RPD_IFL_NOTIFICATION: EVENT [UpDown] ae6.2 index 939 [Broadcast Multicast] address #0 10.2.f.ff.0.0
Apr 22 04:52:04 switch rpd[11077]: RPD_IFA_NOTIFICATION: EVENT ae6.2 index 939
root@switch> show interfaces ae6
Physical interface: ae6, Enabled, Physical link is Up
Interface index: 663, SNMP ifIndex: 631
Link-level type: Flexible-Ethernet, MTU: 9216, Speed: 10Gbps, BPDU Error: None, Ethernet-Switching Error: None, MAC-REWRITE Error: None, Loopback: Disabled, Source filtering: Disabled,
Flow control: Disabled, Minimum links needed: 1, Minimum bandwidth needed: 1bps
Device flags : Present Running
Interface flags: SNMP-Traps Internal: 0x4000
Ethernet segment value: 01:00:00:00:00:01:01:01:01:00, Mode: all-active
Last flapped : 2026-04-22 21:38:06 UTC (00:16:19 ago) <=========== it flapped.
============================Leaf B=======================================
Didn't flapped.
A logical interface flap was observed on one of the uplinks between the Cisco access switch and the Juniper VXLAN leaf switch.
While the physical link remained up, the logical protocol state dropped, causing the Juniper switch to detect an LACP timeout and temporarily bring the aggregated interface down. The event automatically recovered within a few seconds, and no hardware failures or physical‑layer errors were observed on the link.
However, this event appears to have triggered a MAC/IP learning issue between the leaf switches.
It appears that the connectivity to the server was restored after the server‑facing physical switch port went down and came back up, which likely forced Leaf‑A to relearn the server's MAC and ARP entries.
This is related to PR:
https://prsearch.juniper.net/problemreport/PR1816344
That states: " When there is quick flap of core facing interface on scaled setup, the remote MAC and MAC-IP deletion and re-addition is still enqueued.
As a result, you may observe traffic failure due to ARP resolution failure. This involves timing for issue to happen. "
The solution is to upgrade to a code where the PR is fixed.