yahoo-eng-team team mailing list archive

[Aleksandr Shaposhnikov <ashaposhnikov@xxxxxxxxxxxx>]

Public bug reported:

Basically in case of rescheduling dhcp agent (or restarting existing
one) it took a lot of time to create everything for servicing dhcp
requests in tenant networks (namespaces,dnsmasq,iptables,flows).
Basically starting of 708 networks on one dhcp-agent will took around 1
hour 15 minutes on very fast server because everything will be done in
single thread, one by one and so on. Probably some
multithreading/workers should be implemented here.

Steps to reproduce:
1. Create one public/external network with enought amount of IP addresses for 512-1024 connected networks. 
2. Create 1 Network, 1Router, 1 Subnetwork. Connect using router to the external network. Spawn 1 VM using newly created subnetwork.
3. Repeat step 2 500-700 times.
4. Restart dhcp agent and take a look on progress looking on namespaces, debug, logs etc. 

Expected behavior:

dhcp agent should restart in reasonable time (< 5 min)
  
Observed behavior:

dhcp agent restart will took around 1 hour or 600 networks. If you have
pretty dynamic environment (with lot of spawning/destroying of vm's, new
networks, proactive tenants ,etc) it may took longer and also wouldn't
process any updates for already created networks.

** Affects: neutron
     Importance: Undecided
         Status: New

-- 
You received this bug notification because you are a member of Yahoo!
Engineering Team, which is subscribed to neutron.
https://bugs.launchpad.net/bugs/1408164

Title:
  DHCP agent slowness

Status in OpenStack Neutron (virtual network service):
  New

Bug description:
  Basically in case of rescheduling dhcp agent (or restarting existing
  one) it took a lot of time to create everything for servicing dhcp
  requests in tenant networks (namespaces,dnsmasq,iptables,flows).
  Basically starting of 708 networks on one dhcp-agent will took around
  1 hour 15 minutes on very fast server because everything will be done
  in single thread, one by one and so on. Probably some
  multithreading/workers should be implemented here.

  Steps to reproduce:
  1. Create one public/external network with enought amount of IP addresses for 512-1024 connected networks. 
  2. Create 1 Network, 1Router, 1 Subnetwork. Connect using router to the external network. Spawn 1 VM using newly created subnetwork.
  3. Repeat step 2 500-700 times.
  4. Restart dhcp agent and take a look on progress looking on namespaces, debug, logs etc. 

  Expected behavior:

  dhcp agent should restart in reasonable time (< 5 min)
    
  Observed behavior:

  dhcp agent restart will took around 1 hour or 600 networks. If you
  have pretty dynamic environment (with lot of spawning/destroying of
  vm's, new networks, proactive tenants ,etc) it may took longer and
  also wouldn't process any updates for already created networks.

To manage notifications about this bug go to:
https://bugs.launchpad.net/neutron/+bug/1408164/+subscriptions