sr-iov device scanning takes significant time

[Kevin Benton]

The excessive calls to 'ip link show' can consume significant amounts of CPU cycles on every scan_devices iteration on densely packed SR-IOV systems.

Snip from https://bugs.launchpad.net/mos/+bug/1653131 :

On a node with SR-IOV interfaces neutron-sriov-agent seems to call 'ip link show' every 2 second for each interface. This leads to CPU utilization up to 60+% (I've seen 32% for neutron-rootwrap and 32% for neutron-sriov-nic-agent)

Since output of `ip link show' is the same for all interfaces, it is enough to call it only once per 2 sec interval

root@node-7:~# grep '15:10:' /var/log/neutron/neutron-sriov-agent.log | grep "'ip', 'link', 'show'" | wc -l
5783

[OpenStack Infra (hudson-openstack)]

Fix proposed to branch: stable/newton
Review: https://review.openstack.org/416523

[OpenStack Infra (hudson-openstack)]

Fix proposed to branch: stable/mitaka
Review: https://review.openstack.org/416524

[OpenStack Infra (hudson-openstack)]

Reviewed: https://review.openstack.org/416400
Committed: https://git.openstack.org/cgit/openstack/neutron/commit/?id=9a022e7d7b85b7c21cf26698fe59c818c4577194
Submitter: Jenkins
Branch: master

commit 9a022e7d7b85b7c21cf26698fe59c818c4577194
Author: Kevin Benton <email address hidden>
Date: Tue Jan 3 18:10:36 2017 -0800

    Reduce IP link show calls for SR-IOV scan loop

    The device scanning logic was calling 'ip link show' for every
    vf_index on the system, which was taking an excessive amount of
    time an CPU cycles on very dense SR-IOV servers (>64 interfaces).
    This was being executed every 2 seconds so it would cause the
    agent to take a large percentage of the CPU time.

    This patch adjusts the logic to use two 'ip link show' commands for
    checking all of the operations in a single get_assigned_devices_info
    call.

    Closes-Bug: #1653880
    Change-Id: I361caaf7f70580ee8c7b48a4a9c9477bdefd8b4b

[OpenStack Infra (hudson-openstack)]

Reviewed: https://review.openstack.org/416524
Committed: https://git.openstack.org/cgit/openstack/neutron/commit/?id=67e334b81c4038f2cf210e1410b774f9b19fe10e
Submitter: Jenkins
Branch: stable/mitaka

commit 67e334b81c4038f2cf210e1410b774f9b19fe10e
Author: Kevin Benton <email address hidden>
Date: Tue Jan 3 18:10:36 2017 -0800

    Reduce IP link show calls for SR-IOV scan loop

    The device scanning logic was calling 'ip link show' for every
    vf_index on the system, which was taking an excessive amount of
    time an CPU cycles on very dense SR-IOV servers (>64 interfaces).
    This was being executed every 2 seconds so it would cause the
    agent to take a large percentage of the CPU time.

    This patch adjusts the logic to use two 'ip link show' commands for
    checking all of the operations in a single get_assigned_devices_info
    call.

    Conflicts:
      neutron/tests/unit/plugins/ml2/drivers/mech_sriov/agent/test_eswitch_manager.py

    Closes-Bug: #1653880
    Change-Id: I361caaf7f70580ee8c7b48a4a9c9477bdefd8b4b
    (cherry picked from commit 9a022e7d7b85b7c21cf26698fe59c818c4577194)

[OpenStack Infra (hudson-openstack)]

Reviewed: https://review.openstack.org/416523
Committed: https://git.openstack.org/cgit/openstack/neutron/commit/?id=8f98999b8a13bafd78962262b3c51332f3ee50c8
Submitter: Jenkins
Branch: stable/newton

commit 8f98999b8a13bafd78962262b3c51332f3ee50c8
Author: Kevin Benton <email address hidden>
Date: Tue Jan 3 18:10:36 2017 -0800

    Reduce IP link show calls for SR-IOV scan loop

    The device scanning logic was calling 'ip link show' for every
    vf_index on the system, which was taking an excessive amount of
    time an CPU cycles on very dense SR-IOV servers (>64 interfaces).
    This was being executed every 2 seconds so it would cause the
    agent to take a large percentage of the CPU time.

    This patch adjusts the logic to use two 'ip link show' commands for
    checking all of the operations in a single get_assigned_devices_info
    call.

    Conflicts:
      neutron/tests/unit/plugins/ml2/drivers/mech_sriov/agent/test_eswitch_manager.py

    Closes-Bug: #1653880
    Change-Id: I361caaf7f70580ee8c7b48a4a9c9477bdefd8b4b
    (cherry picked from commit 9a022e7d7b85b7c21cf26698fe59c818c4577194)

[OpenStack Infra (hudson-openstack)]

This issue was fixed in the openstack/neutron 10.0.0.0b3 development milestone.

[OpenStack Infra (hudson-openstack)]

This issue was fixed in the openstack/neutron 9.2.0 release.

[OpenStack Infra (hudson-openstack)]

This issue was fixed in the openstack/neutron 8.4.0 release.