Slot reboot on BD8K due to Async Queue growing with CustomType 42 messages
IPARP dynamic entries have reach maximum default (8192) size and there are pending entries.
-> show iparp 0.0.0.0 Dynamic Entries : 8192 Static Entries : 0 Pending Entries : 10 In Request : 779164479 In Response : 330759766 Out Request : 384991566 Out Response : 310592387 Failed Requests : 9035043 Proxy Answered : 310588656 Rx Error : 0 Dup IP Addr : x.x.x.x Rejected Count : 98129646 Rejected IP : x.x.x.x Rejected Port : 8:14 Rejected I/F : Vlan-x Max ARP entries : 8192 Max ARP pending entries : 256
I/O modules continuously rebooting with Async Queue messages showing the customType details
<Erro:Kern.Card.Error> Slot-8: async queue is growing (400019) type is 17481770, customType is 42 <Erro:Kern.Card.Error> Slot-7: async queue is growing (400055) type is 12435498, customType is 42
IPv4 Multicast hardware L3 table full messages are logging every hour
<Warn:Kern.IPv4Mc.Warning> Slot-7: IPv4 multicast entry not added. Hardware L3 Table full. (Logged at most once per hour.) <Warn:Kern.IPv4Mc.Warning> Slot-8: IPv4 multicast entry not added. Hardware L3 Table full. (Logged at most once per hour.)
iproute reserved-entries statistics show high number of IPv4 multicast entries
Large number of MAC moves will trigger FDB flush and IPARP re-learning. During the re-learning process, pending ARP entries could cause async queue to grow and I/O module could experience a reboot as a result.
Module reboot will stop by increasing the IPARP table size, by changing the multicast learning mode from (S,G,V) to (*,G,V), and by allocating more reserved entries for impacted slots.
Reduce the multicast table size by changing the multicast learning mode [ configure igmp snooping forwarding-mode group-vlan ]
Allocate more space from HW IP ROUTE table for IPARP or Host entries [ configure iproute reserved-entries <num-of-routes> slot <slot-no> ]
It is important to note that multicast learning mode should only be changed when there is no L3 multicast (PIM) configured on the switch. Changing the learning mode from (S,G,V) to (*,G,V) is recommended only with L2 multicast traffic.