When using SN2010 with debian and switchdev, please suggest a few ideas for buffer monitoring.
The mlxsw github wiki talks about ethtool statistics: tc_no_buffer_discard_uc_tc_X : The number of unicast packets with traffic class X dropped due to lack of shared buffer resources.
I also found the ‘Devlink exporter’ section which exports packet drop events to Prometheus.
But I can’t find a way to graph / query % of shared buffers in use.
I gave that a read. Do you have suggestions for creating some aggregate numbers to record and graph for monitoring reasons?
It seems there is a byte pool and a descriptor pool involved.
Can I query the ASIC for something like:
Of the byte pool 60% is used.
Of the descriptor pool 20% is used.
Doing this per port and per pool seems complicated and not needed. This page seems to be about QoS - we’re not really interested in QoS, just buffers used while doing standard routing.