On 16-05-20 01:50 AM, Eric Dumazet wrote:
It wont be inaccurate (no drift), as the writers hold the qdisc
spinlock.
It it the same with say IP/TCP SNMP counters :
When an incoming frame is handled, it might change many SNMP counters,
but an SNMP agent might fetch the whole set of SNMP values in the middle
of the changes.
IpInReceives
IpInDelivers
TcpInSegs
IpExtInOctets
IpExtInNoECTPkts
...
The only 'problem' can be a off-by-one (or off-by-bytes-in-the-packet)
transient error, that all SNMP agents are normally handling just fine.
I think qdisc stats being off by a bit are ok - hence the rcu
suggestion. I would have problems with action stats being off
(where they are used for billing).
cheers,
jamal
