network.append({})
# allocate in resource resource model and compute resource limits for new container
+ cpu_limit = mem_limit = disk_limit = -1
+ cpu_period = cpu_quota = None
if self._resource_model is not None:
- # TODO pass resource limits to new container (cf. Dockernet API) Issue #47
+ # call allocate in resource model to calculate resource limit for this container
(cpu_limit, mem_limit, disk_limit) = alloc = self._resource_model.allocate(name, flavor_name)
LOG.debug("Allocation result: %r" % str(alloc))
+ # check if we have a cpu_limit given by the used resource model
+ if cpu_limit > 0:
+ # calculate cpu period and quota for CFS
+ # (see: https://www.kernel.org/doc/Documentation/scheduler/sched-bwc.txt)
+ # TODO consider multi core machines etc! non trivial!
+ # Attention minimum cpu_quota is 1ms (micro)
+ cpu_period = 100000 # lets consider a fixed period of 100000 microseconds for now
+ cpu_quota = cpu_period * cpu_limit # calculate the fraction of cpu time for this container
+ LOG.debug(
+ "CPU limit: cpu_quota = cpu_period * cpu_limit = %f * %f = %f" % (cpu_period, cpu_limit, cpu_quota))
+ # ATTENTION >= 1000 to avoid a invalid argument system error ... no idea why
+ if cpu_quota < 1000:
+ cpu_quota = 1000
+ LOG.warning("Increased CPU quota for %d to avoid system error." % name)
+ # TODO add memory and disc limitations
# create the container
d = self.net.addDocker(
"%s" % (name),
dimage=image,
dcmd=command,
datacenter=self,
- flavor_name=flavor_name)
+ flavor_name=flavor_name,
+ cpu_period=int(cpu_period) if cpu_limit > 0 else None, # set cpu limits if needed
+ cpu_quota=int(cpu_quota) if cpu_limit > 0 else None,
+ )
# connect all given networks
for nw in network:
# TODO we cannot use TCLink here (see: https://github.com/mpeuster/dockernet/issues/3)