openstack
/
tricircle
Code Issues Proposed changes
tricircle/doc/source/install/installation-cell.rst

7.2 KiB
Raw Blame History

Work with Nova cell v2(experiment)

Note

Multi-cell support of Nova cell v2 is under development. DevStack doesn't support multi-cell deployment currently, so the steps discussed in this document may seem not that elegant. We will keep updating this document according to the progress of multi-cell development by Nova team.

Setup

  • 1 Follow "Multi-pod Installation with DevStack" document to prepare your local.conf for both nodes, and set TRICIRCLE_DEPLOY_WITH_CELL to True for both nodes. Start DevStack in node1, then node2.

Note

After running DevStack in both nodes, a multi-cell environment will be prepared: there is one CentralRegion, where Nova API and central Neutron will be registered. Nova has two cells, node1 belongs to cell1, node2 belongs to cell2, and each cell will be configured to use a dedicated local Neutron. For cell1, it's RegionOne Neutron in node1; for cell2, it's RegionTwo Neutron in node2(you can set the region name in local.conf to make the name more friendly). End user can access CentralRegion endpoint of Nova and Neutron to experience the integration of Nova cell v2 and Tricircle.

  • 2 Stop the following services in node2:

    systemctl stop devstack@n-sch.service
systemctl stop devstack@n-super-cond.service
systemctl stop devstack@n-api.service

    if the service of devstack@n-api-meta.service exists, stop it:

    systemctl stop devstack@n-api-meta.service

Note

Actually for cell v2, only one Nova API is required. We enable n-api in node2 because we need DevStack to help us create the necessary cell database. If n-api is disabled, neither API database nor cell database will be created.

  • 3 In node2, run the following command:

    mysql -u$user -p$password -Dnova_cell1 -e 'select host, mapped from compute_nodes'

    you can see that this command returns you one row showing the host of node2 is already mapped:

    +-----------+--------+
| host      | mapped |
+-----------+--------+
| zhiyuan-2 |      1 |
+-----------+--------+

    This host is registered to Nova API in node2, which is already stopped by us, We need to update this row to set "mapped" to 0:

    mysql -u$user -p$password -Dnova_cell1 -e 'update compute_nodes set mapped = 0 where host = "zhiyuan-2"'

    then we can register this host again in step4.

  • 4 In node1, run the following commands to register the new cell:

    nova-manage cell_v2 create_cell --name cell2 \
  --transport-url rabbit://$rabbit_user:$rabbit_passwd@$node2_ip:5672/nova_cell1 \
  --database_connection mysql+pymysql://$db_user:$db_passwd@$node2_ip/nova_cell1?charset=utf8

nova-manage cell_v2 discover_hosts

    then you can see the new cell and host are added in the database:

    mysql -u$user -p$password -Dnova_api -e 'select cell_id, host from host_mappings'

+---------+-----------+
| cell_id | host      |
+---------+-----------+
|       2 | zhiyuan-1 |
|       3 | zhiyuan-2 |
+---------+-----------+

mysql -u$user -p$password -Dnova_api -e 'select id, name from cell_mappings'

+----+-------+
| id | name  |
+----+-------+
|  1 | cell0 |
|  2 | cell1 |
|  3 | cell2 |
+----+-------+

  • 5 In node1, run the following commands:

    systemctl restart devstack@n-sch.service

  • 6 In node1, check if compute services in both hosts are registered:

    openstack --os-region-name CentralRegion compute service list

+----+------------------+-----------+----------+---------+-------+----------------------------+
| ID | Binary           | Host      | Zone     | Status  | State | Updated At                 |
+----+------------------+-----------+----------+---------+-------+----------------------------+
|  5 | nova-scheduler   | zhiyuan-1 | internal | enabled | up    | 2017-09-20T06:56:02.000000 |
|  6 | nova-conductor   | zhiyuan-1 | internal | enabled | up    | 2017-09-20T06:56:09.000000 |
|  8 | nova-consoleauth | zhiyuan-1 | internal | enabled | up    | 2017-09-20T06:56:01.000000 |
|  1 | nova-conductor   | zhiyuan-1 | internal | enabled | up    | 2017-09-20T06:56:07.000000 |
|  3 | nova-compute     | zhiyuan-1 | nova     | enabled | up    | 2017-09-20T06:56:10.000000 |
|  1 | nova-conductor   | zhiyuan-2 | internal | enabled | up    | 2017-09-20T06:56:07.000000 |
|  3 | nova-compute     | zhiyuan-2 | nova     | enabled | up    | 2017-09-20T06:56:09.000000 |
+----+------------------+-----------+----------+---------+-------+----------------------------+

zhiyuan-1 has two nova-conductor services, because one of them is a super
conductor service.

  • 7 Create two aggregates and put the two hosts in each aggregate:

    nova --os-region-name CentralRegion aggregate-create ag1 az1
nova --os-region-name CentralRegion aggregate-create ag2 az2
nova --os-region-name CentralRegion aggregate-add-host ag1 zhiyuan-1
nova --os-region-name CentralRegion aggregate-add-host ag2 zhiyuan-2

  • 8 Create pods, tricircle client is used:

    openstack --os-region-name CentralRegion multiregion networking pod create --region-name CentralRegion
openstack --os-region-name CentralRegion multiregion networking pod create --region-name RegionOne --availability-zone az1
openstack --os-region-name CentralRegion multiregion networking pod create --region-name RegionTwo --availability-zone az2

  • 9 Create network and boot virtual machines:

    net_id=$(openstack --os-region-name CentralRegion network create --provider-network-type vxlan net1 -c id -f value)
openstack --os-region-name CentralRegion subnet create --subnet-range 10.0.1.0/24 --network net1 subnet1
image_id=$(openstack --os-region-name CentralRegion image list -c ID -f value)

openstack --os-region-name CentralRegion server create --flavor 1 --image $image_id --nic net-id=$net_id --availability-zone az1 vm1
openstack --os-region-name CentralRegion server create --flavor 1 --image $image_id --nic net-id=$net_id --availability-zone az2 vm2

Trouble Shooting

  • 1 After you run "compute service list" in step5, you only see services in node1, like:

    +----+------------------+-----------+----------+---------+-------+----------------------------+
| ID | Binary           | Host      | Zone     | Status  | State | Updated At                 |
+----+------------------+-----------+----------+---------+-------+----------------------------+
|  5 | nova-scheduler   | zhiyuan-1 | internal | enabled | up    | 2017-09-20T06:55:52.000000 |
|  6 | nova-conductor   | zhiyuan-1 | internal | enabled | up    | 2017-09-20T06:55:59.000000 |
|  8 | nova-consoleauth | zhiyuan-1 | internal | enabled | up    | 2017-09-20T06:56:01.000000 |
|  1 | nova-conductor   | zhiyuan-1 | internal | enabled | up    | 2017-09-20T06:55:57.000000 |
|  3 | nova-compute     | zhiyuan-1 | nova     | enabled | up    | 2017-09-20T06:56:00.000000 |
+----+------------------+-----------+----------+---------+-------+----------------------------+

    Though new cell has been registered in the database, the running n-api process in node1 may not recognize it. We find that restarting n-api can solve this problem.