Implement hacky/experimental traefik support for reverse proxy

zoe_api/api_endpoint.py

@@ -207,12 +207,14 @@ class APIEndpoint:
         endpoints = []
         for service in execution.services:
             services_info.append(self.service_by_id(user, service.id))
-            for port in service.description['ports']:
-                port_key = str(port['port_number']) + "/" + port['protocol']
-                backend_port = self.sql.ports.select(only_one=True, service_id=service.id, internal_name=port_key)
-                if backend_port is not None and backend_port.external_ip is not None:
-                    endpoint = port['url_template'].format(**{"ip_port": backend_port.external_ip + ":" + str(backend_port.external_port)})
-                    endpoints.append((port['name'], endpoint))
+            for port in service.ports:
+                if port.external_ip is not None:
+                    endpoint = port.url_template.format(**{"ip_port": port.external_ip + ":" + str(port.external_port)})
+                    if zoe_lib.config.get_conf().traefik_zk_ips is None:
+                        endpoint_ext = None
+                    else:
+                        endpoint_ext = '/{}/{}'.format(zoe_lib.config.get_conf().traefik_base_url, port.proxy_key())
+                    endpoints.append((port.name, endpoint, endpoint_ext))
 
         return services_info, endpoints
 

zoe_api/web/templates/execution_inspect.jinja2

@@ -64,7 +64,11 @@
 {% if endpoints|length > 0 %}
 <ul>
 {% for endp in endpoints|sort %}
+    {% if endp[2] != None %}
+        <li><a href="{{ endp[2] }}" target="_blank">{{ endp[0] }}</a> (<a href="{{ endp[1] }}">internal link</a>)</li>
+    {% else %}
     <li><a href="{{ endp[1] }}" target="_blank">{{ endp[0] }}</a> ({{ endp[1] }})</li>
+    {% endif %}
 {% endfor %}
 </ul>
 {% else %}

zoe_lib/config.py

@@ -99,6 +99,8 @@ def load_configuration(test_conf=None):
         argparser.add_argument('--proxy-path', help='Proxy base path', default='127.0.0.1')
         argparser.add_argument('--reverse-proxy-path', help='Base path in case Zoe is behind a reverse proxy under a path', default='')
         argparser.add_argument('--websocket_base', help='Base URL for websocket connections, you need to change it only when running Zoe behind a reverse proxy', default='ws://{{ server_address }}')
+        argparser.add_argument('--traefik-zk-ips', help='ZooKeeper address storing dynamic configuration for træfik', default=None)
+        argparser.add_argument('--traefik-base-url', help='Base path used in reverse proxy URLs generated for træfik', default='/zoe/proxy/')
 
         # Scheduler
         argparser.add_argument('--scheduler-class', help='Scheduler class to use for scheduling ZApps', choices=['ZoeElasticScheduler'], default='ZoeElasticScheduler')

zoe_lib/state/execution.py

@@ -21,7 +21,13 @@ import functools
 
 import psycopg2
 
+try:
+    from kazoo.client import KazooClient
+except ImportError:
+    KazooClient = None
+
 from zoe_lib.state.base import BaseRecord, BaseTable
+import zoe_lib.config
 
 log = logging.getLogger(__name__)
 
@@ -110,11 +116,36 @@ class Execution(BaseRecord):
         self._status = self.RUNNING_STATUS
         self.time_start = datetime.datetime.utcnow()
         self.sql_manager.executions.update(self.id, status=self._status, time_start=self.time_start)
+        #  This hackish code is here to support dynamic reverse proxying of web interfaces running in Zoe executions (e.g. jupyter)
+        #  The idea is to use Træfik to do the reverse proxying, configured to use zookeeper to store dynamic configuration
+        #  Zoe updates ZooKeeper whenever an execution runs or is terminated and Træfik craetes or deletes the route automatically
+        if zoe_lib.config.get_conf().traefik_zk_ips is not None:
+            zk = KazooClient(hosts=zoe_lib.config.get_conf().traefik_zk_ips)
+            zk.start()
+            for service in self.services:
+                for port in service.ports:
+                    endpoint = port.url_template.format(**{"ip_port": port.external_ip + ":" + str(port.external_port)})
+                    traefik_name = 'zoe_exec_{}_{}'.format(self.id, port.id)
+                    zk.create('/traefik/backends/{}/servers/server/url'.format(traefik_name), endpoint, makepath=True)
+                    zk.create('/traefik/frontends/{}/routes/path/rule'.format(traefik_name), 'PathPrefix:{}/{}'.format(zoe_lib.config.get_conf().traefik_base_url, port.proxy_key()), makepath=True)
+                    zk.create('/traefik/frontends/{}/backend'.format(traefik_name), traefik_name, makepath=True)
+            zk.create('/traefik/alias')
+            zk.delete('/traefik/alias')
+            zk.stop()
 
     def set_cleaning_up(self):
         """The services of the execution are being terminated."""
         self._status = self.CLEANING_UP_STATUS
         self.sql_manager.executions.update(self.id, status=self._status)
+        #  See comment in method above
+        if zoe_lib.config.get_conf().traefik_zk_ips is not None:
+            zk = KazooClient(hosts=zoe_lib.config.get_conf().traefik_zk_ips)
+            zk.start()
+            zk.delete('/traefik/backends/zoe_exec_{}', recursive=True)
+            zk.delete('/traefik/frontends/zoetest', recursive=True)
+            zk.create('/traefik/alias')
+            zk.delete('/traefik/alias')
+            zk.stop()
 
     def set_terminated(self, reason=None):
         """The execution is not running."""

zoe_lib/state/port.py

@@ -15,9 +15,11 @@
 
 """Interface to PostgresQL for Zoe state."""
 
+import hashlib
 import logging
 
 from zoe_lib.state.base import BaseRecord, BaseTable
+import zoe_lib.config
 
 log = logging.getLogger(__name__)
 
@@ -62,6 +64,10 @@ class Port(BaseRecord):
         self.external_port = None
         self.external_ip = None
 
+    def proxy_key(self):
+        """Return unique identifier that can be used ti generate proxy URLs."""
+        return hashlib.sha256("{}:{}".format(self.id, zoe_lib.config.get_conf().cookie_secret)).hexdigest()
+
 
 class PortTable(BaseTable):
     """Abstraction for the port table in the database."""

zoe_master/backends/common.py

@@ -43,6 +43,9 @@ def gen_environment(execution: Execution, service: Service, env_subst_dict: Dict
     env_list.append(('ZOE_GID', get_conf().fs_group_id))
     env_list.append(('ZOE_USER', execution.owner.username))
     env_list.append(('SERVICE_NAME', service.name))
+    if get_conf().traefik_zk_ips is not None:
+        for port in service.ports:
+            env_list.append(('REVERSE_PROXY_PATH_{}'.format(port.internal_number), '/{}/{}'.format(get_conf().traefik_base_url, port.proxy_key())))
 
     wk_vol = ZoeFSWorkspace().get(execution.owner)
     env_list.append(('ZOE_WORKSPACE', wk_vol.mount_point))