On 3/27/23 11:16, Ales Musil wrote:
> Currently, there is one ct.est flow per LB VIP,
> that was required to keep track if we need to
> pass the "skip_snat" or "force_snat" flags.
> However since c1d6b8ac ("northd: Store skip_snat and force_snat in
> ct_label/mark")
> the flags are carried in the ct entry and
> we can use match on them the same way we do
> for related traffic.
>
> Simplify the logic for established
> traffic through load balancers, by removing
> the requirement for one ct.est flow per VIP
> and replacing them with three generic ct.est flows:
> match=(ct.est && !ct.rel && !ct.new && ct_mark.natted), action=(next;)
> match=(ct.est && !ct.rel && !ct.new && ct_mark.natted && ct_mark.force_snat
> == 1), action=(flags.force_snat_for_lb = 1; next;)
> match=(ct.est && !ct.rel && !ct.new && ct_mark.natted && ct_mark.skip_snat ==
> 1), action=(flags.skip_snat_for_lb = 1; next;)
>
> This allows us avoiding of matching on L4
> in defrag stage by not storing the L3 and L4
> destination in registers. Match directly on
> L3 and L4 destination for ct.new in DNAT stage.
>
> Populate the registers in LB affinity check stage
> as they are needed for LB affinity learn.
>
> Reported-at: https://bugzilla.redhat.com/2172048
> Reported-at: https://bugzilla.redhat.com/2170885
> Signed-off-by: Ales Musil <[email protected]>
> Reviewed-by: Simon Horman <[email protected]>
> Signed-off-by: Dumitru Ceara <[email protected]>
> (cherry picked from commit ce46a1ba)
> ---
Thanks! Applied to 23.03.
Regards,
Dumitru
_______________________________________________
dev mailing list
[email protected]
https://mail.openvswitch.org/mailman/listinfo/ovs-dev
