[202503] Backport exabgp fix to 202503 branch (#922)

This PR is to backport exabgp changes from master branch to 202503.
PR list
- sonic-net/sonic-mgmt#16744
- sonic-net/sonic-mgmt#17427
- sonic-net/sonic-mgmt#16834
- sonic-net/sonic-mgmt#18181
- sonic-net/sonic-mgmt#18476


These changes are required for running test `bgp/test_traffic_shift.py`

Author: bingwang-ms

ansible/library/exabgp.py

@@ -47,58 +47,51 @@
 DEFAULT_BGP_LISTEN_PORT = 179
 
 http_api_py = '''\
-from flask import Flask, request
+from __future__ import print_function
+import tornado.ioloop
+import tornado.web
 import sys
-import six
-
-#Disable banner msg from app.run, or the output might be caught by exabgp and run as command
-cli = sys.modules['flask.cli']
-cli.show_server_banner = lambda *x: None
-
-app = Flask(__name__)
-
-# Setup a command route to listen for prefix advertisements
-@app.route('/', methods=['POST'])
-def run_command():
-    # code made compatible to run in Py2 or Py3 environment
-    # to support back-porting
-    request_has_commands = False
-    if six.PY2:
-        request_has_commands = request.form.has_key('commands')
-    else:
-        request_has_commands = 'commands' in request.form
-
-    if request_has_commands:
-        cmds = request.form['commands'].split(';')
-    else:
-        cmds = [ request.form['command'] ]
-    for cmd in cmds:
-        sys.stdout.write("%s\\n" % cmd)
-    sys.stdout.flush()
-    return "OK\\n"
 
-if __name__ == '__main__':
-    # with werkzeug 3.x the default size of max_form_memory_size
-    # is 500K. Routes reach a bit beyond that and the client
-    # receives HTTP 413.
-    # Configure the max size to 4 MB to be safe.
-    if not six.PY2:
-        from werkzeug import Request
-        max_content_length = 4 * 1024 * 1024
-        Request.max_content_length = max_content_length
-        Request.max_form_memory_size = max_content_length
-        Request.max_form_parts = max_content_length
-    app.run(host='0.0.0.0', port=sys.argv[1])
+class route_handler(tornado.web.RequestHandler):
+    def post(self):
+        # Read the form data
+        command = self.get_body_argument("command", None)
+        commands = self.get_body_argument("commands", None)
+
+        # Process and print the command values
+        if command:
+            out_str = "{}\\n".format(command)
+            sys.stdout.write(out_str)
+        if commands:
+            values = commands.split(';')
+            for value in values:
+                out_str = "{}\\n".format(value)
+                sys.stdout.write(out_str)
+
+        sys.stdout.flush()
+        self.write("OK\\n")
+
+def make_app():
+    return tornado.web.Application([
+        (r"/upload", UploadHandler),
+    ])
+
+if __name__ == "__main__":
+    app = tornado.web.Application([
+        ("/", route_handler),
+    ])
+    app.listen(int(sys.argv[1]))
+    tornado.ioloop.IOLoop.current().start()
 '''
 
-dump_config_tmpl = '''\
+exabgp3_dump_config_tmpl = '''\
     process dump {
+        run /usr/bin/python {{ dump_script }};
         encoder json;
         receive {
             parsed;
             update;
         }
-        run /usr/bin/python {{ dump_script }};
     }
 '''
 
@@ -132,6 +125,14 @@ def run_command():
 # Example configs are available here
 # https://github.com/Exa-Networks/exabgp/tree/master/etc/exabgp
 # Look for sample for a given section for details
+
+exabgp4_dump_config_tmpl = '''\
+    process dump {
+        run /usr/bin/python {{ dump_script }};
+        encoder json;
+    }
+'''
+
 exabgp4_config_template = '''\
 {{ dump_config }}
 process http-api {
@@ -149,15 +150,34 @@ def run_command():
    passive;
    listen {{ listen_port }};
    {%- endif %}
-   api {
+   api http_api{
        processes [ http-api ];
    }
+   {%- if dump_config %}
+   api dumper {
+       processes [ dump ];
+       receive {
+           parsed;
+           update;
+       }
+   }
+   {%- endif %}
 }
 '''
 
-exabgp_supervisord_conf_tmpl = '''\
+# Unlike in ExaBGP V3.x, in V4+ the process API is expected to acknowledge
+# with 'done' or 'error' string back to ExaBGP. Else the pipe becomes blocked
+# and ExaBGP will hang. Alternatively the acknowledgement can be disabled.
+# https://github.com/Exa-Networks/exabgp/wiki/Migration-from-3.4-to-4.x#api
+exabgp_v4_env_tmpl = '''\
+[exabgp.api]
+ack = false
+'''
+
+exabgp_supervisord_conf_tmpl_p1 = '''\
 [program:exabgp-{{ name }}]
-command=/usr/local/bin/exabgp /etc/exabgp/{{ name }}.conf
+'''
+exabgp_supervisord_conf_tmpl_p3 = '''\
 stdout_logfile=/tmp/exabgp-{{ name }}.out.log
 stderr_logfile=/tmp/exabgp-{{ name }}.err.log
 stdout_logfile_maxbytes=10000000
@@ -170,6 +190,18 @@ def run_command():
 startsecs=1
 numprocs=1
 '''
+exabgp_supervisord_conf_tmpl_p2_v3 = '''\
+command=/usr/local/bin/exabgp /etc/exabgp/{{ name }}.conf
+'''
+exabgp_supervisord_conf_tmpl_p2_v3_debug = '''\
+command=/usr/local/bin/exabgp --debug /etc/exabgp/{{ name }}.conf
+'''
+exabgp_supervisord_conf_tmpl_p2_v4 = '''\
+command=/usr/local/bin/exabgp --env /etc/exabgp/exabgp.env /etc/exabgp/{{ name }}.conf
+'''
+exabgp_supervisord_conf_tmpl_p2_v4_debug = '''\
+command=/usr/local/bin/exabgp --debug --env /etc/exabgp/exabgp.env /etc/exabgp/{{ name }}.conf
+'''
 
 
 def exec_command(module, cmd, ignore_error=False, msg="executing command"):
@@ -234,8 +266,12 @@ def setup_exabgp_conf(name, router_id, local_ip, peer_ip, local_asn, peer_asn, p
 
     dump_config = ""
     if dump_script:
-        dump_config = jinja2.Template(
-            dump_config_tmpl).render(dump_script=dump_script)
+        if six.PY2:
+            dump_config = jinja2.Template(
+                exabgp3_dump_config_tmpl).render(dump_script=dump_script)
+        else:
+            dump_config = jinja2.Template(
+                exabgp4_dump_config_tmpl).render(dump_script=dump_script)
 
     # backport friendly checking; not required if everything is Py3
     t = None
@@ -259,14 +295,42 @@ def setup_exabgp_conf(name, router_id, local_ip, peer_ip, local_asn, peer_asn, p
         out_file.write(data)
 
 
+def setup_exabgp_env():
+    try:
+        os.mkdir("/etc/exabgp", 0o755)
+    except OSError:
+        pass
+    with open("/etc/exabgp/exabgp.env", 'w') as out_file:
+        out_file.write(exabgp_v4_env_tmpl)
+
+
 def remove_exabgp_conf(name):
     try:
         os.remove("/etc/exabgp/%s.conf" % name)
     except Exception:
         pass
 
 
-def setup_exabgp_supervisord_conf(name):
+def setup_exabgp_supervisord_conf(name, debug=False):
+    exabgp_supervisord_conf_tmpl = None
+    if six.PY2:
+        if debug:
+            exabgp_supervisord_conf_tmpl = exabgp_supervisord_conf_tmpl_p1 + \
+                exabgp_supervisord_conf_tmpl_p2_v3_debug + \
+                exabgp_supervisord_conf_tmpl_p3
+        else:
+            exabgp_supervisord_conf_tmpl = exabgp_supervisord_conf_tmpl_p1 + \
+                exabgp_supervisord_conf_tmpl_p2_v3 + \
+                exabgp_supervisord_conf_tmpl_p3
+    else:
+        if debug:
+            exabgp_supervisord_conf_tmpl = exabgp_supervisord_conf_tmpl_p1 + \
+                exabgp_supervisord_conf_tmpl_p2_v4_debug + \
+                exabgp_supervisord_conf_tmpl_p3
+        else:
+            exabgp_supervisord_conf_tmpl = exabgp_supervisord_conf_tmpl_p1 + \
+                exabgp_supervisord_conf_tmpl_p2_v4 + \
+                exabgp_supervisord_conf_tmpl_p3
     t = jinja2.Template(exabgp_supervisord_conf_tmpl)
     data = t.render(name=name)
     with open("/etc/supervisor/conf.d/exabgp-%s.conf" % name, 'w') as out_file:
@@ -302,7 +366,8 @@ def main():
             peer_asn=dict(required=False, type='int'),
             port=dict(required=False, type='int', default=5000),
             dump_script=dict(required=False, type='str', default=None),
-            passive=dict(required=False, type='bool', default=False)
+            passive=dict(required=False, type='bool', default=False),
+            debug=dict(required=False, type='bool', default=False)
         ),
         supports_check_mode=False)
 
@@ -316,32 +381,35 @@ def main():
     port = module.params['port']
     dump_script = module.params['dump_script']
     passive = module.params['passive']
+    debug = module.params['debug']
 
     setup_exabgp_processor()
+    if not six.PY2:
+        setup_exabgp_env()
 
     result = {}
     try:
         if state == 'started':
             setup_exabgp_conf(name, router_id, local_ip, peer_ip, local_asn,
                               peer_asn, port, dump_script=dump_script, passive=passive)
-            setup_exabgp_supervisord_conf(name)
+            setup_exabgp_supervisord_conf(name, debug=debug)
             refresh_supervisord(module)
             start_exabgp(module, name)
         elif state == 'restarted':
             setup_exabgp_conf(name, router_id, local_ip, peer_ip, local_asn,
                               peer_asn, port, dump_script=dump_script, passive=passive)
-            setup_exabgp_supervisord_conf(name)
+            setup_exabgp_supervisord_conf(name, debug=debug)
             refresh_supervisord(module)
             restart_exabgp(module, name)
         elif state == 'present':
             setup_exabgp_conf(name, router_id, local_ip, peer_ip, local_asn,
                               peer_asn, port, dump_script=dump_script, passive=passive)
-            setup_exabgp_supervisord_conf(name)
+            setup_exabgp_supervisord_conf(name, debug=debug)
             refresh_supervisord(module)
         elif state == 'configure':
             setup_exabgp_conf(name, router_id, local_ip, peer_ip, local_asn,
                               peer_asn, port, dump_script=dump_script, passive=passive)
-            setup_exabgp_supervisord_conf(name)
+            setup_exabgp_supervisord_conf(name, debug=debug)
         elif state == 'stopped':
             stop_exabgp(module, name)
         elif state == 'absent':

tests/bgp/bgp_monitor_dump.py

@@ -2,19 +2,36 @@
 
 from sys import stdin
 import json
-
 DUMP_FILE = "/tmp/bgp_monitor_dump.log"
 
+# The announcement format is different between exabgp v3 and v4
+# The default value is set to 'v3' if the version cannot be determined
+# from the message
+exabgp_version = 'v3'
+ver_found = False
+
 while True:
     with open(DUMP_FILE, "a") as f:
         line = stdin.readline()
         obj = json.loads(line)
+        ver = obj.get('exabgp')
+        if ver and not ver_found:
+            ver_found = True
+            if ver.startswith('4'):
+                exabgp_version = 'v4'
         if 'update' not in obj['neighbor']['message']:
             continue
         announce = obj['neighbor']['message']['update']['announce']
         keys = ('ipv4 unicast', 'ipv6 unicast')
         for key in keys:
             if key in announce:
                 for _, route in list(announce[key].items()):
-                    for ip, _ in list(route.items()):
-                        f.write(ip + "\n")
+                    if exabgp_version == 'v3':
+                        for ip, _ in list(route.items()):
+                            f.write(ip + "\n")
+                    elif exabgp_version == 'v4':
+                        route_list = route
+                        for r in route_list:
+                            for msg_type, ip in r.items():
+                                if msg_type == 'nlri':
+                                    f.write(ip + "\n")

tests/common/helpers/bgp.py

@@ -59,7 +59,7 @@ class BGPNeighbor(object):
     def __init__(self, duthost, ptfhost, name,
                  neighbor_ip, neighbor_asn,
                  dut_ip, dut_asn, port, neigh_type=None,
-                 namespace=None, is_multihop=False, is_passive=False):
+                 namespace=None, is_multihop=False, is_passive=False, debug=False):
         self.duthost = duthost
         self.ptfhost = ptfhost
         self.ptfip = ptfhost.mgmt_ip
@@ -73,6 +73,7 @@ def __init__(self, duthost, ptfhost, name,
         self.namespace = namespace
         self.is_passive = is_passive
         self.is_multihop = not is_passive and is_multihop
+        self.debug = debug
 
     def start_session(self):
         """Start the BGP session."""
@@ -111,7 +112,8 @@ def start_session(self):
             peer_ip=self.peer_ip,
             local_asn=self.asn,
             peer_asn=self.peer_asn,
-            port=self.port
+            port=self.port,
+            debug=self.debug
         )
         if not wait_tcp_connection(self.ptfhost, self.ptfip, self.port, timeout_s=60):
             raise RuntimeError("Failed to start BGP neighbor %s" % self.name)

tests/dualtor/test_orchagent_slb.py

@@ -1,4 +1,6 @@
 import ipaddress
+import logging
+import os.path
 import pytest
 import random
 import time
@@ -144,6 +146,9 @@ def _find_ipv6_vlan(mg_facts):
             ptfhost.shell("ip route add %s via %s" % (conn["local_addr"], vlan_intf_addr))
         yield connections
     finally:
+        upper_tor_host.shell("show arp")
+        lower_tor_host.shell("show arp")
+        ptfhost.shell("ip route show")
         for conn in list(connections.values()):
             ptfhost.shell("ifconfig %s 0.0.0.0" % conn["neighbor_intf"], module_ignore_errors=True)
             ptfhost.shell("ip route del %s" % conn["local_addr"], module_ignore_errors=True)
@@ -165,11 +170,33 @@ def bgp_neighbors(ptfhost, setup_interfaces):
             conn["local_addr"].split("/")[0],
             conn["local_asn"],
             conn["exabgp_port"],
-            is_passive=True
+            is_passive=True,
+            debug=True
         )
     return neighbors
 
 
+@pytest.fixture(scope="module", autouse=True)
+def save_slb_exabgp_logfiles(ptfhost, pytestconfig, request):
+    """Save slb exabgp log files to the log directory."""
+    # remove log files before test
+    log_files_before = ptfhost.shell("ls /tmp/exabgp-slb_*.log")["stdout"].split()
+    for log_file in log_files_before:
+        ptfhost.file(path=log_file, state="absent")
+
+    yield
+
+    test_log_file = pytestconfig.getoption("log_file", None)
+    if test_log_file:
+        log_dir = os.path.dirname(os.path.abspath(test_log_file))
+        log_files = ptfhost.shell("ls /tmp/exabgp-slb_*.log")["stdout"].split()
+        for log_file in log_files:
+            logging.debug("Save slb exabgp log %s to %s", log_file, log_dir)
+            ptfhost.fetch(src=log_file, dest=log_dir + os.path.sep, fail_on_missing=False, flat=True)
+    else:
+        logging.info("Skip saving slb exabgp log files to log directory as log directory not set.")
+
+
 @pytest.fixture(params=['ipv4', 'ipv6'])
 def ip_version(request):
     """Traffic IP version to test."""