gateway allocation cleanup (#900)

config/system/all.conf

@@ -7,7 +7,7 @@ include=default.yaml
 site_description="Multi-Device All-Tests Configuration"
 
 # Upstream dataplane port from the external (secondary) switch.
-# Not strictly necessary, but included for illustrative purposes.
+switch_setup.of_dpid=2
 switch_setup.uplink_port=4
 
 # Switch interfaces

config/system/alt.yaml

@@ -11,10 +11,11 @@ switch_setup:
   data_intf: alt-intf
   alt_of_port: 6669
   alt_varz_port: 9305
-  uplink_port: 10
   ext_br: alt-switch
   model: EXT_STACK
   native: True
+  of_dpid: 2
+  uplink_port: 10
 
 # Faux device connection for testing.
 interfaces:

config/system/ata.yaml

@@ -11,6 +11,8 @@ switch_setup:
   model: EXT_STACK
   data_intf: alt-intf
   data_mac: '22:22:22:22:22:22'
+  of_dpid: 2
+  uplink_port: 7
 
 run_trigger:
   native_vlan: 122

config/system/base.yaml

@@ -10,6 +10,11 @@ include: default.yaml
 # Description for dashboard.
 site_description: "Baseline Configuration"
 
+# Default switch configuration.
+switch_setup:
+  of_dpid: 2
+  uplink_port: 7
+
 # Faux interface devices
 interfaces:
   faux:

config/system/default.yaml

@@ -7,11 +7,6 @@
 # Description for dashboard.
 site_description: "Default Configuration"
 
-# Default switch configuration.
-switch_setup:
-  of_dpid: 2
-  uplink_port: 7
-
 # Default time to monitor before starting tests.
 monitor_scan_sec: 30
 

config/system/dts.yaml

@@ -12,7 +12,6 @@ switch_setup:
   varz_port: 5678
   model: EXT_STACK
   native: True
-  of_dpid: '0'
 
 run_trigger:
   vlan_start: 272

config/system/muddy.conf

@@ -7,7 +7,7 @@ include=default.yaml
 site_description="Multi-Device Configuration"
 
 # Upstream dataplane port from the external (secondary) switch.
-# Not strictly necessary, but included for illustrative purposes.
+switch_setup.of_dpid=2
 switch_setup.uplink_port=4
 
 # Switch interfaces

config/system/multi.conf

@@ -7,7 +7,7 @@ include=default.yaml
 site_description="Multi-Device Configuration"
 
 # Upstream dataplane port from the external (secondary) switch.
-# Not strictly necessary, but included for illustrative purposes.
+switch_setup.of_dpid=2
 switch_setup.uplink_port=4
 
 # Switch interfaces

daq/runner.py

@@ -148,7 +148,7 @@ class DAQRunner:
     faucet events, connected hosts (to test), and gcp for logging. This
     class owns the main event loop and shards out work to subclasses."""
 
-    MAX_GATEWAYS = 9
+    _DEFAULT_MAX_GATEWAYS = 9
     _DEFAULT_RETENTION_DAYS = 30
     _SITE_CONFIG = 'site_config.json'
     _RUNNER_CONFIG_PATH = 'runner/setup'
@@ -157,8 +157,14 @@ class owns the main event loop and shards out work to subclasses."""
 
     def __init__(self, config):
         self.configurator = configurator.Configurator()
-        self.gateway_sets = set(range(1, self.MAX_GATEWAYS+1))
         self.config = config
+        switch_setup = self.config.get('switch_setup', {})
+        max_devices = float(switch_setup.get('uplink_port', 'inf')) - 1
+        self.gateway_sets = set(range(1, int(min(self._DEFAULT_MAX_GATEWAYS, max_devices) + 1)))
+        # TODO: uplink port should not be required for base topology
+        # Uplink port is used to configure device ports on the pri switch.
+        switch_setup['uplink_port'] = switch_setup.get('uplink_port', len(self.gateway_sets))
+        self.config['switch_setup'] = switch_setup
         self._result_sets = {}
         self._devices = Devices()
         self._ports = {}
@@ -588,9 +594,10 @@ def main_loop(self):
 
         self._terminate()
 
-    def _target_set_has_capacity(self):
+    def _target_set_has_capacity(self, device):
         num_triggered = len(self._devices.get_triggered_devices())
-        return num_triggered < self._max_hosts
+        existing = self._get_existing_gateway(device)
+        return num_triggered < self._max_hosts and (existing or self.gateway_sets)
 
     def _target_set_trigger(self, device, remote_trigger=False):
         assert self._devices.contains(device), 'Target device %s is not expected' % device
@@ -617,7 +624,7 @@ def _target_set_trigger(self, device, remote_trigger=False):
 
         device.wait_remote = False
 
-        if self._target_set_has_capacity():
+        if self._target_set_has_capacity(device):
             LOGGER.info('Target device %s direct activate', device)
             self._target_set_activate(device)
         else:
@@ -626,11 +633,14 @@ def _target_set_trigger(self, device, remote_trigger=False):
                         device, len(self._target_set_queue))
 
     def _target_set_consider(self):
-        if self._target_set_queue and self._target_set_has_capacity():
-            device = self._target_set_queue.pop(0)
-            LOGGER.info('Target device %s pop activate (%s)',
-                        device, len(self._target_set_queue))
-            self._target_set_activate(device)
+        if self._target_set_queue:
+            for num, device in enumerate(self._target_set_queue):
+                if self._target_set_has_capacity(device):
+                    LOGGER.info('Target device %s pop activate (%s)',
+                                device, len(self._target_set_queue))
+                    self._target_set_activate(device)
+                    self._target_set_queue.pop(num)
+                    break
 
     def _target_set_activate(self, device):
         external_dhcp = device.dhcp_mode == DhcpMode.EXTERNAL
@@ -744,22 +754,28 @@ def _get_test_metadata(self, extension=".daqmodule", root=os.path.join(DAQ_LIB_D
                 }
         return metadata
 
-    def _create_gateway(self, device):
+    def _get_existing_gateway(self, device):
         is_native = device is None
-        assert not is_native or not self._native_gateway, 'native gateway already initialized'
         if self._native_gateway:
             return self._native_gateway
 
-        group_name = 'native' if is_native else device.group
         if not is_native:
-            group_devices = self._devices.get_by_group(group_name)
+            group_devices = self._devices.get_by_group(device.group)
             existing_gateways = {device.gateway for device in group_devices if device.gateway}
             assert len(existing_gateways) <= 1, 'only one existing gateway per group allowed'
             if existing_gateways:
                 existing = existing_gateways.pop()
-                LOGGER.debug('Gateway for existing device group %s is %s', group_name, existing)
+                LOGGER.debug('Gateway for existing device group %s is %s', device.group, existing)
                 return existing
+        return None
 
+    def _create_gateway(self, device):
+        is_native = device is None
+        assert not is_native or not self._native_gateway, 'native gateway already initialized'
+        existing = self._get_existing_gateway(device)
+        if existing:
+            return existing
+        group_name = 'native' if is_native else device.group
         set_num = 1 if is_native else self._find_gateway_set(device)
         LOGGER.info('Gateway for device group %s not found, creating set num %d',
                     group_name, set_num)

daq/session_server.py

@@ -146,7 +146,7 @@ def send_device_heartbeats(self):
             if delta < self._disconnect_timeout_sec:
                 self._send_reply(device_mac, SessionProgress())
             else:
-                LOGGER.warning('Disconnect timeout for %s after %s', device_mac, delta)
+                LOGGER.warning('Disconnect timeout for %s after %ss', device_mac, round(delta))
                 self.close_stream(device_mac)
 
 

daq/topology.py

@@ -54,7 +54,7 @@ def __init__(self, config):
         self.sec_name = 'sec'
         switch_setup = self.config.get('switch_setup', {})
         self.sec_port = int(switch_setup['uplink_port'])
-        self.sec_dpid = int(switch_setup['of_dpid'], 0)
+        self.sec_dpid = int(switch_setup.get('of_dpid', 0))
         self.ext_ofip = switch_setup.get('lo_addr')
         self.ext_intf = switch_setup.get('data_intf')
         self._native_faucet = switch_setup.get('native')

testing/test_dhcp.sh

@@ -7,6 +7,7 @@ echo DHCP Tests >> $TEST_RESULTS
 cat <<EOF > /tmp/daq.conf
 include=${DAQ_LIB}/config/system/default.yaml
 site_description="Multi-Device Configuration"
+switch_setup.of_dpid=2
 switch_setup.uplink_port=7
 interfaces.faux-1.opts=
 interfaces.faux-2.opts=xdhcp

testing/test_many.sh

@@ -17,6 +17,7 @@ echo Many Tests >> $TEST_RESULTS
 
 echo include=../config/system/default.yaml > local/system.conf
 echo monitor_scan_sec=5 >> local/system.conf
+echo switch_setup.of_dpid=2 >> local/system.conf
 echo switch_setup.uplink_port=$((NUM_DEVICES+1)) >> local/system.conf
 echo gcp_cred=$gcp_cred >> local/system.conf
 echo dhcp_lease_time=120s >> local/system.conf

testing/test_utils.sh

@@ -15,6 +15,7 @@ function generate {
   rm -rf inst/runtime_conf
 
   echo switch_setup.uplink_port=$((faux_num+1)) >> local/system.conf
+  echo switch_setup.of_dpid=2 >> local/system.conf
 
   # Create required number of faux devices
   for iface in $(seq 1 $faux_num); do