Revert current implementation of ECMP_Nexthop monitor feature since
it can trigger over-flush of CT entries committed by incoming traffic.
In particular, at the moment there is no way to link the next-hop IP
address (so the ct-entry ID) to the incoming traffic identified
by the ethernet source mac address and so to the ct_lablel committed
by the following logical flow:
table=10(lr_in_ecmp_stateful), priority=100 ,
match=(inport == "R1_ext" && ip4.dst == 10.0.0.0/24 && (ct.new || ct.est)),
action=(ct_commit { ct_label.ecmp_reply_eth = eth.src; ct_mark.ecmp_reply_port
= 2; ct_label.label = 1; }; next;)
Remove the feature while working on an improved approach.
Changes in v2:
- Bump ovn-sb.ovsschema version to 20.37.0.
Lorenzo Bianconi (3):
Revert "ofctrl: Introduce ecmp_nexthop_monitor."
Revert "northd: Add nexhop id in ct_label.label."
Revert "northd: Introduce ECMP_Nexthop table in SB db."
controller/ofctrl.c | 54 -------------------
controller/ofctrl.h | 2 -
controller/ovn-controller.c | 2 -
lib/ovn-util.h | 2 -
northd/en-lflow.c | 3 --
northd/en-northd.c | 35 ------------
northd/en-northd.h | 4 --
northd/inc-proc-northd.c | 8 +--
northd/northd.c | 105 ++++--------------------------------
northd/northd.h | 11 ----
ovn-sb.ovsschema | 18 +------
ovn-sb.xml | 31 -----------
tests/ovn-northd.at | 4 --
tests/ovn.at | 4 +-
tests/system-ovn.at | 79 ++++++++++-----------------
15 files changed, 44 insertions(+), 318 deletions(-)
--
2.46.0
_______________________________________________
dev mailing list
[email protected]
https://mail.openvswitch.org/mailman/listinfo/ovs-dev
