Merge pull request #16236 from opensourcerouting/fix/recursive_weighted_ecmp

zebra: Set the weight for non-recursive next-hop

lib/zclient.c

@@ -2134,6 +2134,7 @@ struct nexthop *nexthop_from_zapi_nexthop(const struct zapi_nexthop *znh)
 	n->ifindex = znh->ifindex;
 	n->gate = znh->gate;
 	n->srte_color = znh->srte_color;
+	n->weight = znh->weight;
 
 	/*
 	 * This function currently handles labels

tests/topotests/bgp_weighted_ecmp_recursive/r1/frr.conf

@@ -0,0 +1,16 @@
+!
+int r1-eth0
+ ip address 192.168.12.1/24
+!
+int r1-eth1
+ ip address 192.168.13.1/24
+!
+router bgp 65000
+ no bgp ebgp-requires-policy
+ neighbor 192.168.12.2 remote-as internal
+ neighbor 192.168.12.2 timers 1 3
+ neighbor 192.168.12.2 timers connect 1
+ neighbor 192.168.13.3 remote-as internal
+ neighbor 192.168.13.3 timers 1 3
+ neighbor 192.168.13.3 timers connect 1
+!

tests/topotests/bgp_weighted_ecmp_recursive/r2/frr.conf

@@ -0,0 +1,20 @@
+!
+int r2-eth0
+ ip address 192.168.12.2/24
+!
+int r2-eth1
+ ip address 192.168.24.2/24
+!
+router bgp 65000
+ no bgp ebgp-requires-policy
+ neighbor 192.168.12.1 remote-as internal
+ neighbor 192.168.12.1 timers 1 3
+ neighbor 192.168.12.1 timers connect 1
+ neighbor 192.168.24.4 remote-as internal
+ neighbor 192.168.24.4 timers 1 3
+ neighbor 192.168.24.4 timers connect 1
+ address-family ipv4 unicast
+  redistribute connected
+  neighbor 192.168.12.1 route-reflector-client
+ exit-address-family
+!

tests/topotests/bgp_weighted_ecmp_recursive/r3/frr.conf

@@ -0,0 +1,20 @@
+!
+int r3-eth0
+ ip address 192.168.13.3/24
+!
+int r3-eth1
+ ip address 192.168.35.3/24
+!
+router bgp 65000
+ no bgp ebgp-requires-policy
+ neighbor 192.168.13.1 remote-as internal
+ neighbor 192.168.13.1 timers 1 3
+ neighbor 192.168.13.1 timers connect 1
+ neighbor 192.168.35.5 remote-as internal
+ neighbor 192.168.35.5 timers 1 3
+ neighbor 192.168.35.5 timers connect 1
+ address-family ipv4 unicast
+  redistribute connected
+  neighbor 192.168.13.1 route-reflector-client
+ exit-address-family
+!

tests/topotests/bgp_weighted_ecmp_recursive/r4/frr.conf

@@ -0,0 +1,17 @@
+!
+int r4-eth0
+ ip address 192.168.24.4/24
+!
+router bgp 65000
+ no bgp ebgp-requires-policy
+ no bgp network import-check
+ neighbor 192.168.24.2 remote-as internal
+ neighbor 192.168.24.2 timers 1 3
+ neighbor 192.168.24.2 timers connect 1
+ address-family ipv4 unicast
+  network 10.10.10.10/32 route-map rmap
+ exit-address-family
+!
+route-map rmap permit 10
+ set extcommunity bandwidth 4000
+exit

tests/topotests/bgp_weighted_ecmp_recursive/r5/frr.conf

@@ -0,0 +1,17 @@
+!
+int r5-eth0
+ ip address 192.168.35.5/24
+!
+router bgp 65000
+ no bgp ebgp-requires-policy
+ no bgp network import-check
+ neighbor 192.168.35.3 remote-as internal
+ neighbor 192.168.35.3 timers 1 3
+ neighbor 192.168.35.3 timers connect 1
+ address-family ipv4 unicast
+  network 10.10.10.10/32 route-map rmap
+ exit-address-family
+!
+route-map rmap permit 10
+ set extcommunity bandwidth 5000
+exit

tests/topotests/bgp_weighted_ecmp_recursive/test_bgp_weighted_ecmp_recursive.py

@@ -0,0 +1,109 @@
+#!/usr/bin/env python
+# SPDX-License-Identifier: ISC
+
+# Copyright (c) 2024 by
+# Donatas Abraitis <donatas@opensourcerouting.org>
+#
+
+"""
+Test if weighted ECMP works and recursive weight (link bandwidth) is
+inherited to non-recursive next-hops.
+"""
+
+import os
+import re
+import sys
+import json
+import pytest
+import functools
+
+pytestmark = [pytest.mark.bgpd]
+
+CWD = os.path.dirname(os.path.realpath(__file__))
+sys.path.append(os.path.join(CWD, "../"))
+
+# pylint: disable=C0413
+from lib import topotest
+from lib.topogen import Topogen, TopoRouter, get_topogen
+from lib.common_config import step
+
+
+def setup_module(mod):
+    topodef = {
+        "s1": ("r1", "r2"),
+        "s2": ("r1", "r3"),
+        "s3": ("r2", "r4"),
+        "s4": ("r3", "r5"),
+    }
+    tgen = Topogen(topodef, mod.__name__)
+    tgen.start_topology()
+
+    router_list = tgen.routers()
+
+    for _, (rname, router) in enumerate(router_list.items(), 1):
+        router.load_frr_config(os.path.join(CWD, "{}/frr.conf".format(rname)))
+
+    tgen.start_router()
+
+
+def teardown_module(mod):
+    tgen = get_topogen()
+    tgen.stop_topology()
+
+
+def test_bgp_weighted_ecmp_recursive():
+    tgen = get_topogen()
+
+    if tgen.routers_have_failure():
+        pytest.skip(tgen.errors)
+
+    r1 = tgen.gears["r1"]
+
+    def _bgp_converge():
+        output = json.loads(r1.vtysh_cmd("show ip route 10.10.10.10/32 json"))
+        expected = {
+            "10.10.10.10/32": [
+                {
+                    "selected": True,
+                    "installed": True,
+                    "nexthops": [
+                        {
+                            "ip": "192.168.24.4",
+                            "active": True,
+                            "recursive": True,
+                            "weight": 204,
+                        },
+                        {
+                            "ip": "192.168.12.2",
+                            "active": True,
+                            "resolver": True,
+                            "weight": 204,
+                        },
+                        {
+                            "ip": "192.168.35.5",
+                            "active": True,
+                            "recursive": True,
+                            "weight": 255,
+                        },
+                        {
+                            "ip": "192.168.13.3",
+                            "active": True,
+                            "resolver": True,
+                            "weight": 255,
+                        },
+                    ],
+                }
+            ]
+        }
+        return topotest.json_cmp(output, expected)
+
+    test_func = functools.partial(
+        _bgp_converge,
+    )
+    _, result = topotest.run_and_expect(test_func, None, count=30, wait=1)
+    assert result is None, "Can't converge"
+
+
+if __name__ == "__main__":
+    args = ["-s"] + sys.argv[1:]
+    sys.exit(pytest.main(args))

zebra/zebra_nhg.c

@@ -1747,6 +1747,12 @@ static struct nexthop *nexthop_set_resolved(afi_t afi,
 	SET_FLAG(resolved_hop->flags, NEXTHOP_FLAG_ACTIVE);
 
 	resolved_hop->vrf_id = nexthop->vrf_id;
+
+	/* Using weighted ECMP, we should respect the weight and use
+	 * the same value for non-recursive next-hop.
+	 */
+	resolved_hop->weight = nexthop->weight;
+
 	switch (newhop->type) {
 	case NEXTHOP_TYPE_IPV4:
 	case NEXTHOP_TYPE_IPV4_IFINDEX: