In ussuri, configuring the neutron.conf bulk_reload_interval enables bulk reload mode. The current code looks to be incompatible with python3.
With the current ussuri code, which looks unchanged on master, I get the following error in docker logs:
---
Running command: 'neutron-dhcp-agent --config-file /etc/neutron/neutron.conf --config-file /etc/neutron/dhcp_agent.ini'
+ exec neutron-dhcp-agent --config-file /etc/neutron/neutron.conf --config-file /etc/neutron/dhcp_agent.ini
Traceback (most recent call last):
File "/usr/lib/python3.6/site-packages/eventlet/hubs/hub.py", line 461, in fire_timers
timer()
File "/usr/lib/python3.6/site-packages/eventlet/hubs/timer.py", line 59, in __call__
cb(*args, **kw)
File "/usr/lib/python3.6/site-packages/eventlet/semaphore.py", line 147, in _do_acquire
waiter.switch()
File "/usr/lib/python3.6/site-packages/neutron/agent/dhcp/agent.py", line 154, in _reload_bulk_allocations
for network_id in self._network_bulk_allocations.keys():
RuntimeError: dictionary changed size during iteration
---
After this, I see no further updates to the dnsmasq hosts file.
The current code looks like this:
https://github.com/openstack/neutron/blob/56bb42fcbc43b619c3c07897c7de88f29158e4b8/neutron/agent/dhcp/agent.py#L157
---
def _reload_bulk_allocations(self):
while True:
for network_id in self._network_bulk_allocations.keys():
network = self.cache.get_network_by_id(network_id)
self.call_driver('bulk_reload_allocations', network)
del self._network_bulk_allocations[network_id]
eventlet.greenthread.sleep(self.conf.bulk_reload_interval)
---
I think the problem is the "del" statement .. code like this works in python2 but not in python3. However, I also wonder if there's some race hazard here with new IDs being appended.
I suspect something like this might work better:
---
def _reload_bulk_allocations(self):
while True:
deleted = self._network_bulk_allocations.copy()
self._network_bulk_allocations = {}
for network_id in deleted:
network = self.cache.get_network_by_id(network_id)
self.call_driver('bulk_reload_allocations', network)
eventlet.greenthread.sleep(self.conf.bulk_reload_interval)
---
However, even this is probably susceptible to a race hazard. Probably need a mutex around any update to self._network_bulk_allocations.
Strange we haven't seen this before, thanks for filing the bug.
I think doing something like you propose would work, and it shouldn't need the copy()
while True:
to_reload = self._network_
self.
for network_id in to_reload:
network = self.cache.
eventlet.