Description

Customer reports intermittent connection between clients connected to an EVPN VXLAN setup, traffic drops are seen intermittently for a specific device while others work.

 

Customer may have exhausted its resources for type 5 tunnel NH entries on LEAF devices:

> show pfe vxlan nh-usage
 
Slot 0
 
------------------OVERLAY NH------------------
Overlay bank size                  : 8192. <<<<
 
Current usage                      :
Reserved entries                   : 1
ARP NDP/Type-5 tunnel NH entries   : 8190  <<<<
Available entries                  : 1     <<<<                                     
 
Expected usage                     :
Reserved entries                   : 1
ARP NDP/Type-5 tunnel NH entries   : 5637
Available entries                  : 2554
 
-----------------UNDERLAY NH-----------------
Underlay bank size                 : 57344
 
Current usage                      :
Reserved entries                   : 5
Underlay NH entries                : 182
Available entries                  : 57157
 
Expected usage                     :
Reserved entries                   : 5
NHs to CPU (IRB + L3 intf) entries : 26
NH ARPs of underlay L3 intfs       : 52
Total tagged and access ports      : 21
VTEP Flood NHs                     : 26
Available entries                  : 57214

And logs may be filled with these messages:

%PFE-3: fpc0 BRCM-VIRTUAL,brcm_virtual_tunnel_nh_install(),2561:Tunnel_nh_hw install failed for nh(9412)

%PFE-3: fpc0 BRCM-VIRTUAL,brcm_virtual_tunnel_create_overlay_nh(),1966:Failed egress if create (Table full) nh-id = 0x14341 gport = 0x80001eb7

Solution

The issue is related to default overlay NH scale limitation of 8192 NHs. This can be easily fixed by increasing the NH entries available with this command:

 

# set forwarding-options vxlan-routing next-hop <integer> <<<<<<<<< Where integer can go from 0 through 61440. The specified value must be a multiple of 4096—for example, 4096, 8192, 12.288, and so on.

NOTE: A restart of the PFE process is required for this change to take precedence.

 

Those 2 steps (configuration and PFE restart) should be performed across all LEAF devices for consistency and avoiding future events.

 

After this, traffic is stable and drops stopped.

Modification History

2024-05-09 : Article Created