Jenkins Redeployment changes (#674)

scripts/busybox-ping-test/alcor_services.ini

@@ -52,29 +52,29 @@ network_config_manager = {"name":"ncm",    "port":9009}
 gateway_manager        = {"name":"gm",     "port":9009}
 
 [AlcorControlAgents]
-node1 = 172.31.22.156
-node2 = 172.31.31.42
+node1 = 172.31.19.133
+node2 = 172.31.21.202
 
 [test_setup]
 vpc_id            = "9192a4d4-ffff-4ece-b3f0-8d36e3d88001"
 project_id        = "3dda2801-d675-4688-a63f-dcda8d327f50"
 tenant_id         = "3dda2801-d675-4688-a63f-dcda8d327f50"
 network_id        = "9192a4d4-ffff-4ece-b3f0-8d36e3d88001"
-cidr              = "10.0.0.0/16"
+cidr              = "172.16.0.0/16"
 node_id           = ["1112a4d4-ffff-4ece-b3f0-8d36e3d85001", "1112a4d4-ffff-4ece-b3f0-8d36e3d85002"]
 node_name         = ["node1", "node2"]
 subnet_id         = "8182a4d4-ffff-4ece-b3f0-8d36e3d88001"
 security_group_id = "3dda2801-d675-4688-a63f-dcda8d111111"
 device_id         = ["8182a4d4-ffff-4ece-b3f0-8d36e3d00001", "8182a4d4-ffff-4ece-b3f0-8d36e3d00002"]
 port_name         = ["port101", "port102"]
 port_id           = ["7122a4d4-ffff-5eee-b3f0-8d36e3d01101", "7122a4d4-ffff-5eee-b3f0-8d36e3d02201"]
-ip_addrs          = ["10.0.1.101", "10.0.1.102"]
+ip_addrs          = ["172.16.1.101", "172.16.1.102"]
 container_names   = ["con1", "con2"]
 
 [L3_AttachRouter_then_CreatePorts]
 subnet_ids   = ["8182a4d4-ffff-4ece-b3f0-8d36e3d88001", "8182a4d4-ffff-4ece-b3f0-8d36e3d88002"]
-cidrs        = ["10.0.1.0/24", "10.0.2.0/24"]
-ip_addrs     = ["10.0.1.101", "10.0.2.201"]
+cidrs        = ["172.16.1.0/24", "172.16.2.0/24"]
+ip_addrs     = ["172.16.1.101", "172.16.2.201"]
 subnet_names = ["subnet1", "subnet2"]
 device_ids   = ["8182a4d4-ffff-4ece-b3f0-8d36e3d00001", "8182a4d4-ffff-4ece-b3f0-8d36e3d00002"]
 
@@ -84,7 +84,7 @@ sg_names= ["sg1","sg2"]
 device_ids   = ["8182a4d4-ffff-4ece-b3f0-8d36e3d00001", "8182a4d4-ffff-4ece-b3f0-8d36e3d00002"]
 
 [gateways]
-gateway_info = [{"gw" : "10.0.1.1" , "ips" : ["10.0.1.101", "10.0.1.102"]}, {"gw" : "10.0.2.1", "ips" : ["10.0.2.201"]}]
+gateway_info = [{"gw" : "172.16.1.1" , "ips" : ["172.16.1.101", "172.16.1.102"]}, {"gw" : "172.16.2.1", "ips" : ["172.16.2.201"]}]
 
 [vpc_info]
 vpc_info = {"cidr":${test_setup:cidr},"id":${test_setup:vpc_id}, "project_id":${test_setup:project_id}}
@@ -93,7 +93,7 @@ vpc_info = {"cidr":${test_setup:cidr},"id":${test_setup:vpc_id}, "project_id":${
 node_info = {"node_id":${test_setup:node_id}, "node_name":${test_setup:node_name}, "server_port":8080, "veth":"eth0"}
 
 [subnet_info]
-subnet_info = {"cidr":${test_setup:cidr},"id":${test_setup:subnet_id},"ip_version":4,"network_id":${test_setup:network_id},"name":"subnet1","host_routes":[{"destination":"10.0.1.0/24","nexthop":"10.0.1.1"}]}
+subnet_info = {"cidr":${test_setup:cidr},"id":${test_setup:subnet_id},"ip_version":4,"network_id":${test_setup:network_id},"name":"subnet1","host_routes":[{"destination":"172.16.1.0/24","nexthop":"172.16.1.1"}]}
 
 [security_groups]
 security_group_info = {"create_at":"string","description":"string","id":${test_setup:security_group_id},"name":"sg1","project_id":${test_setup:project_id},"security_group_rules":[],"tenant_id":${test_setup:tenant_id},"update_at":"string"}

scripts/busybox-ping-test/create_test_cases.py

@@ -116,7 +116,7 @@ def prepare_test_L3_AttachRouter_then_CreatePorts(ip_mac, ser_port):
     serv = read_config_file_section("services")
     create_default_segment_table(ser_port["sgs"])
     create_node(ser_port["nm"], ip_mac)
-    change = {'change':'cidr','cidr':"10.0.0.0/16"}
+    change = {'change':'cidr','cidr':"172.16.0.0/16"}
     create_vpc(ser_port["vpm"], change)
     create_security_group(ser_port["sgm"])
 
@@ -157,7 +157,7 @@ def prepare_test_L3_CreatePorts_then_AttachRouter(ip_mac, ser_port):
     serv = read_config_file_section("services")
     create_default_segment_table(ser_port["sgs"])
     create_node(ser_port["nm"], ip_mac)
-    change = {'change':'cidr','cidr':"10.0.0.0/16"}
+    change = {'change':'cidr','cidr':"172.16.0.0/16"}
     create_vpc(ser_port["vpm"], change)
     get_vpcs(ser_port["vpm"])
     create_security_group(ser_port["sgm"])

scripts/busybox-ping-test/helper_functions.py

@@ -148,6 +148,39 @@ def print_output(output):
       print(line)
 
 
+# check if all alcor services are up and running.
+# Success: all ports from 9001 through 9016 should show up in netstat output
+# try for 5 minutes, waiting 10 seconds each time
+# NOTE: Update the port list when new services are added.
+# 9001 9002 9003 9004 9005 9006 9007 9008 9009 9010 9011 9012 9015 9016
+def check_alcor_services():
+    wait_limit = 300
+    sleep_time = 10
+    wait_time = 0
+    try:
+        command = "netstat -ant | awk -F: '/90[0-9][0-9]/ {print $4}' | sed 's/[\t ]*$//' | sort -n | tr '[\n]' ' ' | sed 's/[\t ]*$//'"
+        iter = 1
+        while wait_time < wait_limit:
+            pipe = sp.Popen(command, stdout=sp.PIPE, stderr=sp.PIPE, shell=True)
+            res = pipe.communicate()
+            retcode = pipe.returncode
+            if retcode > 0:
+                print("Failed to execute command", repr(str(command)))
+                print_output(res[1])
+            elif "9001 9002 9003 9004 9005 9006 9007 9008 9009 9010 9011 9012 9015 9016" in str(res):
+                print("SUCCESS for: ", command, "\n")
+                return True
+            iter = iter + 1
+            wait_time = wait_time + sleep_time
+            time.sleep(sleep_time)
+
+        return False
+    except:
+        print("ERROR", sys.exc_info()[0])
+    return False
+
+
+
 def execute_command(command):
     try:
       pipe = sp.Popen(command, stdout=sp.PIPE, stderr=sp.PIPE, shell=True)

scripts/busybox-ping-test/ping_test.py

@@ -41,6 +41,8 @@ def build_containers(services_dict):
 
     print("building container images")
     services_list = get_file_list(ALCOR_SERVICES)
+    # segment_service is bogus service, only used to pick up the
+    # internal port number of vpc_manager, which is 9001
     for service_name in services_dict.keys():
       if service_name == "segment_service":
          continue
@@ -61,6 +63,7 @@ def build_containers(services_dict):
 def start_containers(serv):
     start_containers = []
     for service_name in serv.keys():
+        extra_args = ""
         if service_name == "segment_service":
             continue
         service_info = json.loads(serv[service_name])
@@ -69,10 +72,17 @@ def start_containers(serv):
 
         if service_name == "ignite":
             ports = "-p 10800:10800 -p 10801:10801 -p 47100:47100 -p 47500:47500 "
-            start_cmd = run_container + ports + mnt_and_image + "sh"
+            extra_args = "sh"
+        elif service_name == "vpc_manager":
+            # expose internal and external ports in VPM
+            vpm_info = json.loads(serv["segment_service"])
+            ports = "--net=host -p {}:{} -p {}:{} ".format(service_info["port"], service_info["port"], vpm_info["port"], vpm_info["port"])
         else:
             ports = "--net=host -p {}:{} ".format(service_info["port"], service_info["port"])
-            start_cmd = run_container + ports + mnt_and_image
+
+        start_cmd = run_container + ports + mnt_and_image
+        if not extra_args:
+            start_cmd = start_cmd + " " + extra_args
 
         start_containers.append(start_cmd)
 
@@ -85,20 +95,21 @@ def start_containers(serv):
 def stop_containers(service_list):
     command = "sudo docker container stop "
     for service in service_list:
-      if service == "segment_service":
+      if service == "sgs":
          continue
       execute_command(command + service)
 
 
 def remove_containers(service_list):
     command = "sudo docker container rm "
     for service in service_list:
-      if service == "segment_service":
+      if service == "sgs":
          continue
       execute_command(command + service)
 
 
 def main():
+    extra_wait_time = 120
     config_file =  "{}/alcor_services.ini".format(ALCOR_TEST_DIR)
     config_file_object = read_config_file(config_file)
     services_dict = dict(config_file_object.items("services"))
@@ -160,8 +171,14 @@ def main():
       print("ERROR: Quitting test\n")
       sys.exit(1)
 
-    print("Wait for 60 seconds until all services are started...")
-    time.sleep(60)
+    print("Waiting for Alcor services to be up and running...\n")
+    alcor_status = check_alcor_services()
+    if alcor_status == False:
+        print("ERROR: Alcor Services failed to start ...\n")
+        sys.exit(1)
+
+    print("Alcor Services are up and running, but waiting for {} seconds more...\n".format(extra_wait_time))
+    time.sleep(extra_wait_time)
 
     if args.testcase:
       if (args.testcase == 1):

services/data_plane_manager/src/main/java/com/futurewei/alcor/dataplane/service/impl/DpmServiceImpl.java

@@ -509,6 +509,9 @@ private List<String> processRouterConfiguration(NetworkConfiguration networkConf
                             unicastGoalStateMap.put(hostIp, unicastGoalState);
                             for (Neighbor.NeighborState neighbor : neighbors)
                             {
+                                // neighbor can be NULL, at least in S5, so skip it
+                                if (neighbor == null)
+                                    continue;
                                 unicastGoalState.getGoalStateBuilder().addNeighborStates(neighbor);
                                 for (Neighbor.NeighborConfiguration.FixedIp fixIp : neighbor.getConfiguration().getFixedIpsList())
                                 {
@@ -606,4 +609,4 @@ public InternalDPMResultList updateNetworkConfiguration(NetworkConfiguration net
     public InternalDPMResultList deleteNetworkConfiguration(NetworkConfiguration networkConfig) throws Exception {
         return processNetworkConfiguration(networkConfig);
     }
-}
+}