Symptom

C9300X Standalone or in a half-ring stack running IOS XE 17.6 to 17.9 has been observed to silently drop packets if the internal forwarding path of the flow traverses the 2 ASICs on the system No observable interface counters or statistics will reflect this packet loss You can confirm the ASIC to port mappings via 'show platform software fed switch active ifm mappings' The instance column will show you the ASIC that the corresponding port maps to: 9300X-24Y#show platform software fed switch active ifm mappings Interface IF_ID Inst Asic Core Port SubPort Mac Cntx LPN GPN Type Active TwentyFiveGigE1/0/1 0x9 1 1 0 0 0 7 8 1 1 NRU Y TwentyFiveGigE1/0/2 0xa 1 1 0 1 0 6 9 2 2 NRU Y TwentyFiveGigE1/0/3 0x41 1 1 0 2 0 5 10 3 3 NRU Y TwentyFiveGigE1/0/4 0xc 1 1 0 3 0 4 11 4 4 NRU Y TwentyFiveGigE1/0/5 0xd 1 1 0 4 0 3 12 5 5 NRU Y TwentyFiveGigE1/0/6 0xe 1 1 0 5 0 2 13 6 6 NRU Y TwentyFiveGigE1/0/7 0xf 1 1 0 6 0 1 14 7 7 NRU Y TwentyFiveGigE1/0/8 0x10 1 1 0 7 0 0 15 8 8 NRU Y TwentyFiveGigE1/0/9 0x11 1 1 0 8 0 7 16 9 9 NIF Y TwentyFiveGigE1/0/10 0x12 1 1 0 9 0 6 17 10 10 NIF Y TwentyFiveGigE1/0/11 0x13 1 1 0 10 0 5 18 11 11 NIF Y TwentyFiveGigE1/0/12 0x14 1 1 0 11 0 4 19 12 12 NIF Y TwentyFiveGigE1/0/13 0x44 0 0 0 0 0 7 8 13 13 NRU Y TwentyFiveGigE1/0/14 0x16 0 0 0 1 0 6 9 14 14 NRU Y TwentyFiveGigE1/0/15 0x17 0 0 0 2 0 5 10 15 15 NRU Y TwentyFiveGigE1/0/16 0x18 0 0 0 3 0 4 11 16 16 NRU Y TwentyFiveGigE1/0/17 0x19 0 0 0 4 0 3 12 17 17 NRU Y TwentyFiveGigE1/0/18 0x1a 0 0 0 5 0 2 13 18 18 NRU Y TwentyFiveGigE1/0/19 0x1b 0 0 0 6 0 1 14 19 19 NRU Y TwentyFiveGigE1/0/20 0x1c 0 0 0 7 0 0 15 20 20 NRU Y TwentyFiveGigE1/0/21 0x1d 0 0 0 8 0 7 16 21 21 NIF Y TwentyFiveGigE1/0/22 0x1e 0 0 0 9 0 6 17 22 22 NIF Y TwentyFiveGigE1/0/23 0x1f 0 0 0 10 0 5 18 23 23 NIF Y

Conditions

Traffic flow ingresses port on one ASIC and egresses a port on a different ASIC This issue has been observed for both TCP and UDP traffic flows Following PIDs are subject of this problem: C9300X-24Y, C9300X-48HX, C9300X-48TX

Workaround

There are 3 workarounds: 1. connect a stack cable to the switch stack ports. One cable is required that connects stack port 1 to stack port 2 1a. complete the ring in the stack 2. Manually move links to where the flows will ingress and egress the same ASIC and the loss will not be observed 3. Disable the diagnostic test #5 that is not actively used on the system but still runs in the background. This test impact is contributing to the traffic loss. From global configuration, issue "no diagnostic monitor switch test 5" This will require a save and reload to take effect.

Further Problem Description