[c-nsp] ME3600X 15.2S memory leak - Update!

Mark Tinka mark.tinka at seacom.mu
Sat Mar 5 09:56:01 EST 2016


So feedback from the ME3600X BU is that this issue is likely similar to
what Jason described earlier in this thread.

So the workaround is to disable egress ACL functionality.

Apparently, 15.5(3)S2 is supposed to be fixed, but given that our
upgrade to this code still led to a forwarding freeze, safe to say Cisco
need to put more work into this.

Just FYI, the ASR920 is not affected, or any other platform running this
code base.

The ME3600X/3800X really is an exciting box :-\...

Mark.

On 1/Mar/16 12:02, Mark Tinka wrote:
>
> On 29/Feb/16 12:17, Spyros Kakaroukas wrote:
>
>> This is interesting.
>>
>> I've had something that fits your description happen on a me3600x a few days
>> ago. I don't have OOB on this box yet ( Hello Murphy! ) but the local logs
>> indicate that IS-IS went down, the box stopped routing completely but it did
>> not crash or show any other indications of anything awry. Unfortunately the
>> local peeps power-cycled it without snapping a sh tech first , so I don't
>> have any other info.
> This is exactly our issue, Spyros.
>
> If it does happen again to you, please capture the below before
> rebooting the device (we got these from TAC):
>
> conf t
>  scheduler interval 500
> end
> !
> show interfaces (take this output 5x times)
> show spanning-tree detail
> show spanning-tree summary
> show spanning-tree detail | inc ieee|occurr|from|is exec
> show process cpu sorted
> show process cpu history
> show process memory sorted
> show process memory statistics
> show platform port-asic stats drop
> !
> conf t
>  service internal
> end
> !
>
> Run each of the below commands 5x times:
>
> show platform aspdma all_counters 0
> show platform aspdma all_counters 1
> show platform aspdma all_counters 0
> show platform aspdma all_counters 1
> show platform asic stats 0
> show platform asic stats 1
> show platform asic stats 0
> show platform asic stats 1
> show platform qos policer cpu 0 0
> show platform qos policer cpu 1 0
> show platform qos policer cpu 2 0
> show platform qos policer cpu 3 0
> show platform qos policer cpu 6 0
> show platform qos policer cpu 7 0
> show platform qos policer cpu 8 0
> show platform asic  ecc 0
> show platform asic  ecc 1
> !
> sdcli
> !
>
> Run each of the below commands 5x times:
>
>                 nile stats ipstats 0
>                 nile stats ipstats 1
>                 nile stats channel 0
>                 nile stats channel 1
>                 nile debug stats 0 ForwardProcessorStats
>                 nile debug stats 1 ForwardProcessorStats
>                 nile debug stats 0 EgressHeaderBufferProcessor
>                 nile debug stats 1 EgressHeaderBufferProcessor
>                 nile debug stats 0 ChannelCompact
>                 nile debug stats 1 ChannelCompact
>                 nile pp reg egrfixptokenin show 0 0
>                 nile pp reg egrfixptokenout show 0 0
>                 nile pp reg maceight0flowcontroldropcounter show 0 0
>                 nile pp reg maceight1flowcontroldropcounter show 0 0
>                 nile pp reg maceight2flowcontroldropcounter show 0 0
>                 nile pp reg maceight0flowcontroldropcounter show 1 0
>                 nile pp reg maceight1flowcontroldropcounter show 1 0
>                 nile pp reg maceight2flowcontroldropcounter show 1 0
>                 nile pp reg fixptokenin show 1 0
>                 nile pp reg fixptokenout show 1 0
>                 nile pp reg fixptokenin show 0 0
>                 nile pp reg fixptokenout show 0 0
>                 nile qm reg replicationdebugcounters show 1 0
>                 nile qm reg replicationdebugcounters show 0 0
>                 arsenic mmap i_read 0x40 0008010c
>                 arsenic mmap i_read 0x40 0009010c
>                 arsenic mmap i_read 0x40 0026010c
>                 arsenic mmap i_read 0x45 0008010c
>                 arsenic mmap i_read 0x45 0009010c
>                 arsenic mmap i_read 0x45 0026010c
>
> Mark.
>




More information about the cisco-nsp mailing list