R80.20 T5 ARP table critical issue

[George_Liu]

I have a customer install R80.20 T5 with HPE Gen10 server, recently customer report that user randomly cannot connect to network, after research that I found 3.1 kernel not perform a well arp garbage recycling.

All incomplete arp record were kept in arp table cause new arp entry cannot insert.

I reprocedure this issue on my lab, the gateway only install, set nothing and not intialize, just ping and issue occure (at lease 1 mintues arp incomplete exist)

I had raise a TAC case, any good news will update here.

[Timothy_Hall]

The arp driver will most certainly have been updated for kernel version 3.10; this ARP overflow issue is covered in my book so I'm definitely interested in any behavior changes here.  Can you please provide output from the following commands on the 3.10 system with the problem:

ls -l /proc/sys/net/ipv4/neigh/default

cat /proc/sys/net/ipv4/neigh/default/*

[George_Liu]

The setting are all default withou change manually

[George_Liu]

The is capture from customer, befroe I change arp table cache size from 4096 to 16384, there is a lot of log in /var/log/message show below, few seconds a time.

net_ratelimit: N callbacks suppressed

[George_Liu]

I try to use some linux command to delete the arp entry, but it still stay in incomplete state.

I think when delete arp entry is set the record to incomplete state and waiting for remove, it's should be the mechanism of arp recycling.

[George_Liu]

I do the lab tonight, T11 (kernel 3.1) also have the same arp issue, but T101 (kernel 2.6) won't.