Make host_manager use scatter-gather and ignore down cells

[Surya Seetharaman]

Currently the "_get_computes_for_cells" function in the host_manager of scheduler runs sequentially and this affects the performance in case of large deployments (running a lot of cells) :

https://github.com/openstack/nova/blob/stable/pike/nova/scheduler/host_manager.py#L601

So it would be nice to use the scatter_gather_all_cells function to do this operation in parallel.

Also apart from the performance scaling point of view, in case connection to a particular cell fails, it would be nice to have sentinels returned which is done by the scatter_gather_all_cells. This helps when a cell is down.

[OpenStack Infra (hudson-openstack)]

Reviewed: https://review.openstack.org/539617
Committed: https://git.openstack.org/cgit/openstack/nova/commit/?id=fdea8b723ba5a25ea9dc0917401fbb1401e05ee3
Submitter: Zuul
Branch: master

commit fdea8b723ba5a25ea9dc0917401fbb1401e05ee3
Author: Dan Smith <email address hidden>
Date: Wed Jan 31 09:30:11 2018 -0800

    Make host_manager use scatter-gather and ignore down cells

    This makes the host_manager query for computes in parallel across all the
    cells. It also ignores cells that fail or time out so that scheduling can
    proceed.

    Closes-Bug: #1746561
    Change-Id: I48d8b763f475c010fa48ee1db232a6d3ae75f5e6

[OpenStack Infra (hudson-openstack)]

This issue was fixed in the openstack/nova 18.0.0.0b2 development milestone.

[OpenStack Infra (hudson-openstack)]

Fix proposed to branch: stable/queens
Review: https://review.openstack.org/637599

[OpenStack Infra (hudson-openstack)]

Reviewed: https://review.openstack.org/637599
Committed: https://git.openstack.org/cgit/openstack/nova/commit/?id=6933c880b8cea760c1e930b320b3d44b1e7e1b53
Submitter: Zuul
Branch: stable/queens

commit 6933c880b8cea760c1e930b320b3d44b1e7e1b53
Author: Dan Smith <email address hidden>
Date: Wed Jan 31 09:30:11 2018 -0800

    Make host_manager use scatter-gather and ignore down cells

    This makes the host_manager query for computes in parallel across all the
    cells. It also ignores cells that fail or time out so that scheduling can
    proceed.

    Closes-Bug: #1746561
    Change-Id: I48d8b763f475c010fa48ee1db232a6d3ae75f5e6
    (cherry picked from commit fdea8b723ba5a25ea9dc0917401fbb1401e05ee3)

[OpenStack Infra (hudson-openstack)]

This issue was fixed in the openstack/nova 17.0.10 release.