Try to make smoke less flakey (#1663)

.github/workflows/smoke/build-relay.sh

@@ -16,8 +16,10 @@ relay:
   am_relay: true
 EOF
 
-    export LIGHTHOUSES="192.168.100.1 172.17.0.2:4242"
-    export REMOTE_ALLOW_LIST='{"172.17.0.4/32": false, "172.17.0.5/32": false}'
+    # TEST-NET-3 placeholder IPs; smoke-relay.sh seds them to real container IPs.
+    # Mapping: .2 lighthouse1, .3 host2, .4 host3, .5 host4.
+    export LIGHTHOUSES="192.168.100.1 203.0.113.2:4242"
+    export REMOTE_ALLOW_LIST='{"203.0.113.4/32": false, "203.0.113.5/32": false}'
 
     HOST="host2" ../genconfig.sh >host2.yml <<EOF
 relay:
@@ -25,7 +27,7 @@ relay:
     - 192.168.100.1
 EOF
 
-    export REMOTE_ALLOW_LIST='{"172.17.0.3/32": false}'
+    export REMOTE_ALLOW_LIST='{"203.0.113.3/32": false}'
 
     HOST="host3" ../genconfig.sh >host3.yml
 

.github/workflows/smoke/build.sh

@@ -5,9 +5,15 @@ set -e -x
 rm -rf ./build
 mkdir ./build
 
-# TODO: Assumes your docker bridge network is a /24, and the first container that launches will be .1
-# - We could make this better by launching the lighthouse first and then fetching what IP it is.
-NET="$(docker network inspect bridge -f '{{ range .IPAM.Config }}{{ .Subnet }}{{ end }}' | cut -d. -f1-3)"
+# Smoke containers run on a dedicated docker network whose subnet is allocated
+# at smoke time, not known at build time. Configs are written with TEST-NET-3
+# placeholder IPs (RFC 5737) and smoke.sh / smoke-vagrant.sh / smoke-relay.sh
+# sed the real container IPs in before starting nebula.
+#
+# Placeholder mapping (last octet == fixed container slot):
+#   203.0.113.2 -> lighthouse1, 203.0.113.3 -> host2,
+#   203.0.113.4 -> host3,       203.0.113.5 -> host4.
+LIGHTHOUSE_IP="203.0.113.2"
 
 (
     cd build
@@ -25,16 +31,16 @@ NET="$(docker network inspect bridge -f '{{ range .IPAM.Config }}{{ .Subnet }}{{
         ../genconfig.sh >lighthouse1.yml
 
     HOST="host2" \
-        LIGHTHOUSES="192.168.100.1 $NET.2:4242" \
+        LIGHTHOUSES="192.168.100.1 $LIGHTHOUSE_IP:4242" \
         ../genconfig.sh >host2.yml
 
     HOST="host3" \
-        LIGHTHOUSES="192.168.100.1 $NET.2:4242" \
+        LIGHTHOUSES="192.168.100.1 $LIGHTHOUSE_IP:4242" \
         INBOUND='[{"port": "any", "proto": "icmp", "group": "lighthouse"}]' \
         ../genconfig.sh >host3.yml
 
     HOST="host4" \
-        LIGHTHOUSES="192.168.100.1 $NET.2:4242" \
+        LIGHTHOUSES="192.168.100.1 $LIGHTHOUSE_IP:4242" \
         OUTBOUND='[{"port": "any", "proto": "icmp", "group": "lighthouse"}]' \
         ../genconfig.sh >host4.yml
 

.github/workflows/smoke/smoke-relay.sh

@@ -6,6 +6,8 @@ set -o pipefail
 
 mkdir -p logs
 
+NETWORK="nebula-smoke-relay"
+
 cleanup() {
     echo
     echo " *** cleanup"
@@ -16,22 +18,53 @@ cleanup() {
     then
         docker kill lighthouse1 host2 host3 host4
     fi
+    docker network rm "$NETWORK" >/dev/null 2>&1
 }
 
 trap cleanup EXIT
 
+# Create a dedicated smoke network with an explicit subnet (required for --ip
+# below). Probe a short list of candidates so a locally-used range doesn't
+# fail the whole test — we only need one to be free.
+docker network rm "$NETWORK" >/dev/null 2>&1 || true
+for candidate in 172.30.0.0/24 172.31.0.0/24 10.98.0.0/24 10.99.0.0/24 192.168.230.0/24; do
+    if docker network create --subnet "$candidate" "$NETWORK" >/dev/null 2>&1; then
+        break
+    fi
+done
+if ! docker network inspect "$NETWORK" >/dev/null 2>&1; then
+    echo "failed to create $NETWORK: every candidate subnet is in use" >&2
+    exit 1
+fi
+
+# Derive container IPs from the network's assigned subnet. Slots: .2 lighthouse1,
+# .3 host2, .4 host3, .5 host4 — matches the placeholders in build-relay.sh.
+SUBNET="$(docker network inspect -f '{{(index .IPAM.Config 0).Subnet}}' "$NETWORK")"
+PREFIX="${SUBNET%/*}"
+PREFIX="${PREFIX%.*}"
+LIGHTHOUSE_IP="$PREFIX.2"
+HOST2_IP="$PREFIX.3"
+HOST3_IP="$PREFIX.4"
+HOST4_IP="$PREFIX.5"
+
+# Sed the placeholder TEST-NET-3 IPs in the host configs to the real ones.
+for f in build/host2.yml build/host3.yml build/host4.yml; do
+    sed "s|203\.0\.113\.|$PREFIX.|g" "$f" >"$f.tmp"
+    mv "$f.tmp" "$f"
+done
+
 docker run --name lighthouse1 --rm nebula:smoke-relay -config lighthouse1.yml -test
-docker run --name host2 --rm nebula:smoke-relay -config host2.yml -test
-docker run --name host3 --rm nebula:smoke-relay -config host3.yml -test
-docker run --name host4 --rm nebula:smoke-relay -config host4.yml -test
+docker run --name host2 --rm -v "$PWD/build/host2.yml:/nebula/host2.yml:ro" nebula:smoke-relay -config host2.yml -test
+docker run --name host3 --rm -v "$PWD/build/host3.yml:/nebula/host3.yml:ro" nebula:smoke-relay -config host3.yml -test
+docker run --name host4 --rm -v "$PWD/build/host4.yml:/nebula/host4.yml:ro" nebula:smoke-relay -config host4.yml -test
 
-docker run --name lighthouse1 --device /dev/net/tun:/dev/net/tun --cap-add NET_ADMIN --rm nebula:smoke-relay -config lighthouse1.yml 2>&1 | tee logs/lighthouse1 | sed -u 's/^/  [lighthouse1]  /' &
+docker run --name lighthouse1 --network "$NETWORK" --ip "$LIGHTHOUSE_IP" --device /dev/net/tun:/dev/net/tun --cap-add NET_ADMIN --rm nebula:smoke-relay -config lighthouse1.yml 2>&1 | tee logs/lighthouse1 | sed -u 's/^/  [lighthouse1]  /' &
 sleep 1
-docker run --name host2 --device /dev/net/tun:/dev/net/tun --cap-add NET_ADMIN --rm nebula:smoke-relay -config host2.yml 2>&1 | tee logs/host2 | sed -u 's/^/  [host2]  /' &
+docker run --name host2 --network "$NETWORK" --ip "$HOST2_IP" -v "$PWD/build/host2.yml:/nebula/host2.yml:ro" --device /dev/net/tun:/dev/net/tun --cap-add NET_ADMIN --rm nebula:smoke-relay -config host2.yml 2>&1 | tee logs/host2 | sed -u 's/^/  [host2]  /' &
 sleep 1
-docker run --name host3 --device /dev/net/tun:/dev/net/tun --cap-add NET_ADMIN --rm nebula:smoke-relay -config host3.yml 2>&1 | tee logs/host3 | sed -u 's/^/  [host3]  /' &
+docker run --name host3 --network "$NETWORK" --ip "$HOST3_IP" -v "$PWD/build/host3.yml:/nebula/host3.yml:ro" --device /dev/net/tun:/dev/net/tun --cap-add NET_ADMIN --rm nebula:smoke-relay -config host3.yml 2>&1 | tee logs/host3 | sed -u 's/^/  [host3]  /' &
 sleep 1
-docker run --name host4 --device /dev/net/tun:/dev/net/tun --cap-add NET_ADMIN --rm nebula:smoke-relay -config host4.yml 2>&1 | tee logs/host4 | sed -u 's/^/  [host4]  /' &
+docker run --name host4 --network "$NETWORK" --ip "$HOST4_IP" -v "$PWD/build/host4.yml:/nebula/host4.yml:ro" --device /dev/net/tun:/dev/net/tun --cap-add NET_ADMIN --rm nebula:smoke-relay -config host4.yml 2>&1 | tee logs/host4 | sed -u 's/^/  [host4]  /' &
 sleep 1
 
 set +x
@@ -76,7 +109,13 @@ docker exec host4 sh -c 'kill 1'
 docker exec host3 sh -c 'kill 1'
 docker exec host2 sh -c 'kill 1'
 docker exec lighthouse1 sh -c 'kill 1'
-sleep 5
+
+# Wait up to 30s for all backgrounded jobs to exit rather than relying on a
+# fixed sleep.
+for _ in $(seq 1 30); do
+    [ -z "$(jobs -r)" ] && break
+    sleep 1
+done
 
 if [ "$(jobs -r)" ]
 then

.github/workflows/smoke/smoke-vagrant.sh

@@ -8,6 +8,8 @@ export VAGRANT_CWD="$PWD/vagrant-$1"
 
 mkdir -p logs
 
+NETWORK="nebula-smoke"
+
 cleanup() {
     echo
     echo " *** cleanup"
@@ -19,21 +21,51 @@ cleanup() {
         docker kill lighthouse1 host2
     fi
     vagrant destroy -f
+    docker network rm "$NETWORK" >/dev/null 2>&1
 }
 
 trap cleanup EXIT
 
+# Create a dedicated smoke network with an explicit subnet (required for --ip
+# below). Probe a short list of candidates so a locally-used range doesn't
+# fail the whole test — we only need one to be free.
+docker network rm "$NETWORK" >/dev/null 2>&1 || true
+for candidate in 172.30.0.0/24 172.31.0.0/24 10.98.0.0/24 10.99.0.0/24 192.168.230.0/24; do
+    if docker network create --subnet "$candidate" "$NETWORK" >/dev/null 2>&1; then
+        break
+    fi
+done
+if ! docker network inspect "$NETWORK" >/dev/null 2>&1; then
+    echo "failed to create $NETWORK: every candidate subnet is in use" >&2
+    exit 1
+fi
+
+# Derive container IPs from the network's assigned subnet. Slots: .2 lighthouse1,
+# .3 host2 — matches the placeholders in build.sh.
+SUBNET="$(docker network inspect -f '{{(index .IPAM.Config 0).Subnet}}' "$NETWORK")"
+PREFIX="${SUBNET%/*}"
+PREFIX="${PREFIX%.*}"
+LIGHTHOUSE_IP="$PREFIX.2"
+HOST2_IP="$PREFIX.3"
+
+# Sed the placeholder TEST-NET-3 IPs in the host configs to the real ones.
+# This must happen before `vagrant up` rsyncs build/ into the VM for host3.
+for f in build/host2.yml build/host3.yml; do
+    sed "s|203\.0\.113\.|$PREFIX.|g" "$f" >"$f.tmp"
+    mv "$f.tmp" "$f"
+done
+
 CONTAINER="nebula:${NAME:-smoke}"
 
 docker run --name lighthouse1 --rm "$CONTAINER" -config lighthouse1.yml -test
-docker run --name host2 --rm "$CONTAINER" -config host2.yml -test
+docker run --name host2 --rm -v "$PWD/build/host2.yml:/nebula/host2.yml:ro" "$CONTAINER" -config host2.yml -test
 
 vagrant up
 vagrant ssh -c "cd /nebula && /nebula/$1-nebula -config host3.yml -test" -- -T
 
-docker run --name lighthouse1 --device /dev/net/tun:/dev/net/tun --cap-add NET_ADMIN --rm "$CONTAINER" -config lighthouse1.yml 2>&1 | tee logs/lighthouse1 | sed -u 's/^/  [lighthouse1]  /' &
+docker run --name lighthouse1 --network "$NETWORK" --ip "$LIGHTHOUSE_IP" --device /dev/net/tun:/dev/net/tun --cap-add NET_ADMIN --rm "$CONTAINER" -config lighthouse1.yml 2>&1 | tee logs/lighthouse1 | sed -u 's/^/  [lighthouse1]  /' &
 sleep 1
-docker run --name host2 --device /dev/net/tun:/dev/net/tun --cap-add NET_ADMIN --rm "$CONTAINER" -config host2.yml 2>&1 | tee logs/host2 | sed -u 's/^/  [host2]  /' &
+docker run --name host2 --network "$NETWORK" --ip "$HOST2_IP" -v "$PWD/build/host2.yml:/nebula/host2.yml:ro" --device /dev/net/tun:/dev/net/tun --cap-add NET_ADMIN --rm "$CONTAINER" -config host2.yml 2>&1 | tee logs/host2 | sed -u 's/^/  [host2]  /' &
 sleep 1
 vagrant ssh -c "cd /nebula && sudo sh -c 'echo \$\$ >/nebula/pid && exec /nebula/$1-nebula -config host3.yml'" 2>&1 -- -T | tee logs/host3 | sed -u 's/^/  [host3]  /' &
 sleep 15
@@ -96,7 +128,14 @@ vagrant ssh -c "ping -c1 192.168.100.2" -- -T
 vagrant ssh -c "sudo xargs kill </nebula/pid" -- -T
 docker exec host2 sh -c 'kill 1'
 docker exec lighthouse1 sh -c 'kill 1'
-sleep 1
+
+# Wait up to 30s for all backgrounded jobs to exit. vagrant ssh in particular
+# takes a beat to tear down after nebula exits on the VM, so a fixed sleep is
+# racy.
+for _ in $(seq 1 30); do
+    [ -z "$(jobs -r)" ] && break
+    sleep 1
+done
 
 if [ "$(jobs -r)" ]
 then

.github/workflows/smoke/smoke.sh

@@ -6,6 +6,8 @@ set -o pipefail
 
 mkdir -p logs
 
+NETWORK="nebula-smoke"
+
 cleanup() {
     echo
     echo " *** cleanup"
@@ -16,24 +18,56 @@ cleanup() {
     then
         docker kill lighthouse1 host2 host3 host4
     fi
+    docker network rm "$NETWORK" >/dev/null 2>&1
 }
 
 trap cleanup EXIT
 
+# Create a dedicated smoke network with an explicit subnet (required for --ip
+# below). Probe a short list of candidates so a locally-used range doesn't
+# fail the whole test — we only need one to be free.
+docker network rm "$NETWORK" >/dev/null 2>&1 || true
+for candidate in 172.30.0.0/24 172.31.0.0/24 10.98.0.0/24 10.99.0.0/24 192.168.230.0/24; do
+    if docker network create --subnet "$candidate" "$NETWORK" >/dev/null 2>&1; then
+        break
+    fi
+done
+if ! docker network inspect "$NETWORK" >/dev/null 2>&1; then
+    echo "failed to create $NETWORK: every candidate subnet is in use" >&2
+    exit 1
+fi
+
+# Derive container IPs from the network's assigned subnet. Slots: .2 lighthouse1,
+# .3 host2, .4 host3, .5 host4 — matches the placeholders in build.sh.
+SUBNET="$(docker network inspect -f '{{(index .IPAM.Config 0).Subnet}}' "$NETWORK")"
+PREFIX="${SUBNET%/*}"
+PREFIX="${PREFIX%.*}"
+LIGHTHOUSE_IP="$PREFIX.2"
+HOST2_IP="$PREFIX.3"
+HOST3_IP="$PREFIX.4"
+HOST4_IP="$PREFIX.5"
+
+# Sed the placeholder TEST-NET-3 IPs in the host configs to the real ones.
+# build/lighthouse1.yml has no IPs to rewrite so it's skipped.
+for f in build/host2.yml build/host3.yml build/host4.yml; do
+    sed "s|203\.0\.113\.|$PREFIX.|g" "$f" >"$f.tmp"
+    mv "$f.tmp" "$f"
+done
+
 CONTAINER="nebula:${NAME:-smoke}"
 
 docker run --name lighthouse1 --rm "$CONTAINER" -config lighthouse1.yml -test
-docker run --name host2 --rm "$CONTAINER" -config host2.yml -test
-docker run --name host3 --rm "$CONTAINER" -config host3.yml -test
-docker run --name host4 --rm "$CONTAINER" -config host4.yml -test
+docker run --name host2 --rm -v "$PWD/build/host2.yml:/nebula/host2.yml:ro" "$CONTAINER" -config host2.yml -test
+docker run --name host3 --rm -v "$PWD/build/host3.yml:/nebula/host3.yml:ro" "$CONTAINER" -config host3.yml -test
+docker run --name host4 --rm -v "$PWD/build/host4.yml:/nebula/host4.yml:ro" "$CONTAINER" -config host4.yml -test
 
-docker run --name lighthouse1 --device /dev/net/tun:/dev/net/tun --cap-add NET_ADMIN --rm "$CONTAINER" -config lighthouse1.yml 2>&1 | tee logs/lighthouse1 | sed -u 's/^/  [lighthouse1]  /' &
+docker run --name lighthouse1 --network "$NETWORK" --ip "$LIGHTHOUSE_IP" --device /dev/net/tun:/dev/net/tun --cap-add NET_ADMIN --rm "$CONTAINER" -config lighthouse1.yml 2>&1 | tee logs/lighthouse1 | sed -u 's/^/  [lighthouse1]  /' &
 sleep 1
-docker run --name host2 --device /dev/net/tun:/dev/net/tun --cap-add NET_ADMIN --rm "$CONTAINER" -config host2.yml 2>&1 | tee logs/host2 | sed -u 's/^/  [host2]  /' &
+docker run --name host2 --network "$NETWORK" --ip "$HOST2_IP" -v "$PWD/build/host2.yml:/nebula/host2.yml:ro" --device /dev/net/tun:/dev/net/tun --cap-add NET_ADMIN --rm "$CONTAINER" -config host2.yml 2>&1 | tee logs/host2 | sed -u 's/^/  [host2]  /' &
 sleep 1
-docker run --name host3 --device /dev/net/tun:/dev/net/tun --cap-add NET_ADMIN --rm "$CONTAINER" -config host3.yml 2>&1 | tee logs/host3 | sed -u 's/^/  [host3]  /' &
+docker run --name host3 --network "$NETWORK" --ip "$HOST3_IP" -v "$PWD/build/host3.yml:/nebula/host3.yml:ro" --device /dev/net/tun:/dev/net/tun --cap-add NET_ADMIN --rm "$CONTAINER" -config host3.yml 2>&1 | tee logs/host3 | sed -u 's/^/  [host3]  /' &
 sleep 1
-docker run --name host4 --device /dev/net/tun:/dev/net/tun --cap-add NET_ADMIN --rm "$CONTAINER" -config host4.yml 2>&1 | tee logs/host4 | sed -u 's/^/  [host4]  /' &
+docker run --name host4 --network "$NETWORK" --ip "$HOST4_IP" -v "$PWD/build/host4.yml:/nebula/host4.yml:ro" --device /dev/net/tun:/dev/net/tun --cap-add NET_ADMIN --rm "$CONTAINER" -config host4.yml 2>&1 | tee logs/host4 | sed -u 's/^/  [host4]  /' &
 sleep 1
 
 # grab tcpdump pcaps for debugging
@@ -131,7 +165,13 @@ docker exec host4 sh -c 'kill 1'
 docker exec host3 sh -c 'kill 1'
 docker exec host2 sh -c 'kill 1'
 docker exec lighthouse1 sh -c 'kill 1'
-sleep 5
+
+# Wait up to 30s for all backgrounded jobs to exit rather than relying on a
+# fixed sleep.
+for _ in $(seq 1 30); do
+    [ -z "$(jobs -r)" ] && break
+    sleep 1
+done
 
 if [ "$(jobs -r)" ]
 then