yahoo-eng-team team mailing list archive

[OpenStack Infra <1975674@xxxxxxxxxxxxxxxxxx>]

Reviewed:  https://review.opendev.org/c/openstack/neutron/+/843253
Committed: https://opendev.org/openstack/neutron/commit/e09b128f416a809cd7734aba8ab52220ea01b2e2
Submitter: "Zuul (22348)"
Branch:    master

commit e09b128f416a809cd7734aba8ab52220ea01b2e2
Author: Henning Eggers <henning@xxxxxxxxxxx>
Date:   Wed May 25 11:17:43 2022 +0200

    Defer flow deletion in openvswitch firewall
    
    Reduces the deletion time of conjunction flows on hypervisors
    where virtual machines reside which are part of a security
    group that has remote security groups as target which contain
    thousands of ports.
    
    Without deferred deletion the agent will call ovs-ofctl several
    hundred times in succession, during this time the agent will
    block any new vm creation or neutron port modifications on this
    hypervisor.
    
    This patch has been tested using a single network with a single
    vm with a security group that points to a remote security group
    with 2000 ports.
    
    During testing without the patch, the iteration time for deletion
    was at around 500 seconds. After adding the patch to the l2 agent
    on the test environment the same deletion time went down to
    4 seconds.
    
    Closes-Bug: #1975674
    Change-Id: I46b1fe94b2e358f7f4b2cd4943a74ebaf84f51b8


** Changed in: neutron
       Status: In Progress => Fix Released

-- 
You received this bug notification because you are a member of Yahoo!
Engineering Team, which is subscribed to neutron.
https://bugs.launchpad.net/bugs/1975674

Title:
  Neutron agent blocks during VM deletion when a remote security group
  is involved

Status in neutron:
  Fix Released

Bug description:
  When deleting a VM that has a security group referring to a remote
  security group, the neutron agent will block for as long as it takes
  to remove the respective flows. This happens when the remote security
  group contains many (thousands) ports referring to other VMs.

  Steps to reproduce:
    - Create a VM with security group A
    - Add a rule to security group A allowing access from a remote security group B
    - Add a large number or ports to security group B (e.g. 2000)
      - The respective ovs flows will be added
    - Delete the VM
      - The ovs flows will be removed

  Expected:
    - VM and flow to be deleted within seconds
    - No impact to other VMs on the same hypervisor

  Actual:
    - Flow deletion takes a long time, sometimes up to 10 minutes
    - While flows are being deleted, no VMs can be created on the same hypervisor

  The reason for this behavior is that under the hood the agent calls
  ovs-ofctl (via execve()) once for each port in the remote security
  group. These calls quickly add up to minutes if there are many ports.

  The proposed solution would be to use deferred execution for the flow
  deletion. In that case it becomes a bulk operation and around 400
  flows are deleted in one call. In addition it runs in the background
  and does not block the agent for other operations.

To manage notifications about this bug go to:
https://bugs.launchpad.net/neutron/+bug/1975674/+subscriptions