Clusters¶
Concept¶
A Cluster is a group of logical objects, each of which is called a
Node in Senlin’s terminology. A cluster can contain zero or more
nodes. A cluster has a profile_id
property that specifies which default
Profile to use when new nodes are created/scaled as members of the
cluster. It is valid for nodes in a cluster to reference different profile
objects because Senlin only mandates that all nodes in a cluster having the
same profile type.
Senlin provides APIs and command line supports to manage the cluster membership. Please refer to Cluster Membership for details. Senlin also supports attaching Policy objects to a cluster, customizing the policy properties when attaching a policy to a cluster. Please refer to Cluster-Policy Bindings for details.
Listing Clusters¶
The following command shows the clusters managed by the Senlin service:
$ openstack cluster list
+----------+------+--------+----------------------+------------+
| id | name | status | created_at | updated_at |
+----------+------+--------+----------------------+------------+
| 2959122e | c1 | ACTIVE | 2015-05-05T13:27:28Z | None |
| 092d0955 | c2 | ACTIVE | 2015-05-05T13:27:48Z | None |
+----------+------+--------+----------------------+------------+
Note that the first column in the output table is a short ID of a cluster
object. Senlin command line use short IDs to save real estate on screen so
that more useful information can be shown on a single line. To show the full
ID in the list, you can add the --full-id
option to the command:
$ openstack cluster list --full-id
+--------------------+------+--------+--------------------- +------------+
| id | name | status | created_at | updated_at |
+--------------------+------+--------+----------------------+------------+
| 2959122e-11c7-.... | c1 | ACTIVE | 2015-05-05T13:27:28Z | None |
| 092d0955-2645-.... | c2 | ACTIVE | 2015-05-05T13:27:48Z | None |
+--------------------+------+--------+----------------------+------------+
Sorting the List¶
You can specify the sorting keys and sorting direction when list clusters,
using the option --sort
. The --sort
option accepts a
string of format key1[:dir1],key2[:dir2],key3[:dir3]
, where the keys used
are cluster properties and the dirs can be one of asc
and desc
. When
omitted, Senlin sorts a given key using asc
as the default direction.
For example, the following command sorts the clusters using the name
property in descending order:
$ openstack cluster list --sort name:desc
When sorting the list of clusters, you can use one of name
, status
,
init_at
, created_at
and updated_at
.
Filtering the List¶
The openstack cluster list command also provides options for
filtering the cluster list at the server side. The option --filters
can be used for this purpose. For example, the following command filters the
clusters by the status
field:
$ openstack cluster list --filters status=ACTIVE
+----------+------+--------+----------------------+------------+
| id | name | status | created_at | updated_at |
+----------+------+--------+----------------------+------------+
| 2959122e | c1 | ACTIVE | 2015-05-05T13:27:28Z | None |
| 092d0955 | c2 | ACTIVE | 2015-05-05T13:27:48Z | None |
+----------+------+--------+----------------------+------------+
The option --filters
accepts a list of key-value pairs separated by
semicolon (;
), where each key-value pair is expected to be of format
<key>=<value>
. The valid keys for filtering include: status
, name
,
project
and user
.
Paginating the Query Results¶
In case you have a huge collection of clusters, you can limit the number of
clusters returned from Senlin server each time, using the option
--limit
. For example:
$ openstack cluster list --limit 1
+----------+------+--------+----------------------+------------+
| id | name | status | created_at | updated_at |
+----------+------+--------+----------------------+------------+
| 2959122e | c1 | ACTIVE | 2015-05-05T13:27:28Z | None |
+----------+------+--------+----------------------+------------+
Another option you can specify is the ID of a cluster after which you want to
see the returned list starts. In other words, you don’t want to see those
clusters with IDs that is or come before the one you specify. You can use the
option --marker
for this purpose. For example:
$ openstack cluster list --limit 1 \
--marker 2959122e-11c7-4e82-b12f-f49dc5dac270
+----------+------+--------+----------------------+------------+
| id | name | status | created_at | updated_at |
+----------+------+--------+----------------------+------------+
| 092d0955 | c2 | ACTIVE | 2015-05-05T13:27:48Z | None |
+----------+------+--------+----------------------+------------+
Only 1 cluster record is returned in this example and its UUID comes after the one specified from the command line.
Creating a Cluster¶
To create a cluster, you need to provide the ID or name of the profile to be associated with the cluster. For example:
$ openstack cluster create --profile qstack c3
+------------------+--------------------------------------+
| Property | Value |
+------------------+--------------------------------------+
| config | {} |
| created_at | None |
| data | {} |
| dependents | {} |
| desired_capacity | 0 |
| domain_id | None |
| id | 60424eb3-6adf-4fc3-b9a1-4a035bf171ac |
| init_at | 2015-05-05T13:35:47Z |
| location | None |
| max_size | -1 |
| metadata | {} |
| min_size | 0 |
| name | c3 |
| node_ids | |
| profile_id | bf38dc9f-d204-46c9-b515-79caf1e45c4d |
| profile_name | qstack |
| project_id | 333acb15a43242f4a609a27cb097a8f2 |
| status | INIT |
| status_reason | Initializing |
| timeout | 3600 |
| updated_at | None |
| user_id | 0b82043b57014cd58add97a2ef79dac3 |
+------------------+--------------------------------------+
From the output you can see that a new cluster object created and put to
INIT
status. Senlin will verify if profile specified using the option
--profile
does exist. The server allows the <PROFILE>
value to be a profile name, a profile ID or the short ID of a profile object.
If the profile is not found or multiple profiles found matching the value, you
will receive an error message.
Controlling Cluster Capacity¶
When creating a cluster, by default senlin will create a cluster
with no nodes, i.e. the desired_capacity
will be set to 0. However, you
can specify the desired capacity of the cluster, the maximum size and/or the
minimum size of the cluster. The default value for min_size
is 0 and the
default value for max_size
is -1, meaning that there is no upper bound for
the cluster size.
The following command creates a cluster named “test_cluster
”, with its
desired capacity set to 2, its minimum size set to 1 and its maximum size set
to 3:
$ openstack cluster create --desired-capacity 2 \
--min-size 1 --max-size 3 \
--profile myprofile \
test_cluster
Senlin API and Senlin engine will validate the settings for these capacity arguments when receiving this request. An error message will be returned if the arguments fail to pass this validation, or else the cluster creation request will be queued as an action for execution.
When desired_capacity
is not specified and min_size
is not specified,
Senlin engine will create an empty cluster. When either desired_capacity
or min_size
is specified, Senlin will start the process of creating nodes
immediately after the cluster object is created.
Other Properties¶
You can use the option --metadata
(or -M
) to associate
some key-value pairs to the cluster to be created. These data are referred to
as the “metadata” for the cluster.
Since cluster operations may take some time to finish when being executed and
Senlin interacts with the backend services to make it happen, there needs a
way to verify whether an operation has timed out. When creating a cluster
using the openstack cluster create command line, you can use the
option --timeout
to specify the default time out in number
of seconds. This value would be the global setting for the cluster.
You can use the option --config
to pass in key-value pairs to the
cluster to be created. The following config properties are supported:
node.name.format
: Specify how cluster nodes are automatically named. The value can contain placeholders like$nI
for node index padded with n number of zeros to the left, or$nR
for random string of length n.cluster.stop_node_before_delete
: If set to True, cluster operations that result in a node deletion (e.g. scale-in, resize, etc) will request a node stop first. Once the node has been successfully shutdown, the node is deleted. The default setting is False for which a cluster performs a node delete without stopping the node.cluster.stop_timeout_before_update
: Specifies the timeout value in seconds to wait for when stopping a node as part of a cluster update operation. If this value is not set, the value for default_nova_timeout in the configuration will be used.
Showing Details of a Cluster¶
When there are clusters in the Senlin database, you can request Senlin to show the details about a cluster you are interested in.
You can use the name, the ID or the “short ID” of a cluster to name a cluster for show. Senlin API and engine will verify if the identifier you specified can uniquely identify a cluster. An error message will be returned if there is no cluster matching the identifier or if more than one cluster matching it.
An example is shown below:
$ openstack cluster show c3
+------------------+--------------------------------------+
| Field | Value |
+------------------+--------------------------------------+
| config | {} |
| created_at | 2015-07-07T03:30:53Z |
| data | {} |
| dependents | {} |
| desired_capacity | 2 |
| domain_id | None |
| id | 2b7e9294-b5cd-470f-b191-b18f7e672495 |
| init_at | 2015-05-07T03:30:52Z |
| location | None |
| max_size | -1 |
| metadata | {} |
| min_size | 0 |
| name | c3 |
| node_ids | b28692a5-2536-4921-985b-1142d6045e1f |
| | 4be10a88-e340-4518-a9e1-d742c53ac37f |
| profile_id | bf38dc9f-d204-46c9-b515-79caf1e45c4d |
| profile_name | qstack |
| project_id | 333acb15a43242f4a609a27cb097a8f2 |
| status | ACTIVE |
| status_reason | Node stack2: Creation succeeded |
| timeout | 3600 |
| updated_at | None |
| user_id | 0b82043b57014cd58add97a2ef79dac3 |
+------------------+--------------------------------------+
From the result, you can examine the list of nodes (if any) that are members of this cluster.
Updating a Cluster¶
Once a cluster has been created, you change its properties using the openstack cluster update command. For example, to change the name of a cluster, you can use the following command:
$ openstack cluster update --name web_bak web_servers
You can change the timeout
property using option --timeout
.
You can change the metadata associated with cluster using option
--metadata
.
Using the openstack cluster update command, you can change the profile used by the cluster and its member nodes. The following example launches a global update on the cluster for switching to a different profile:
$ openstack cluster update --profile fedora21_server web_cluster
Suppose the cluster web_cluster
is now using a profile of type
os.nova.server
where a Fedora 20 image is used, the command above will
initiate a global upgrade to a new profile where a Fedora 21 image is used.
Senlin engine will verify whether the new profile has the same profile type
with that of the existing one and whether the new profile has a well-formed
spec
property. If everything is fine, the engine will start a node level
profile update process. The node level update operation is subject to policy
checkings/enforcements when there is an update policy attached to the cluster.
Please refer to Policies and Cluster-Policy Bindings for more
information.
Resizing a Cluster¶
The openstack cluster command line supports several different sub-commands to resize a cluster.
openstack cluster resize
¶
The command openstack cluster resize takes several arguments that allow you to resize a cluster in various ways:
you can change the size of a cluster to a specified number;
you can add a specified number of nodes to a cluster or remove a specified number of nodes from a cluster;
you can instruct openstack cluster resize to resize a cluster by a specified percentage;
you can tune the
min_size
and/ormax_size
property of a cluster when resizing it;you can request a size change made on a best-effort basis, if the resize operation cannot be fully realized due to some restrictions, this argument tells Senlin engine whether it is still expected to partially realize the resize operation.
You can specify one and only one of the following options for the openstack cluster resize command:
use
--capacity
to specify the exact value of the new cluster size;use
--adjustment
to specify the relative number of nodes to add/remove;use
--percentage
to specify the percentage of cluster size change.
The following command resizes the cluster test_cluster
to 2 nodes,
provided that the min_size
is less than or equal to 2 and the max_size
is either no less than 2 or equal to -1 (indicating that there is no upper
bound for the cluster size). This command makes use of the option
--capacity
, where <CAPACITY>
is the new size of the
cluster:
$ openstack cluster resize --capacity 2 test_cluster
Another way to resize a cluster is by specifying the --adjustment
option, where <ADJUSTMENT>
can be a positive or a negative
integer giving the number of nodes to add or remove respectively. For example,
the following command adds two nodes to the specified cluster:
$ openstack cluster resize --adjustment 2 test_cluster
The following command removes two nodes from the specified cluster:
$ openstack cluster resize --adjustment -2 test_cluster
Yet another way to resize a cluster is by specifying the size change in
percentage. You will use the option --percentage
for
this purpose. The <PERCENTAGE>
value can be either a positive float value
or a negative float value giving the percentage of cluster size. For example,
the following command increases the cluster size by 30%:
$ openstack cluster resize --percentage 30 test_cluster
The following command decreases the cluster size by 25%:
$ openstack cluster resize --percentage -25 test_cluster
Senlin engine computes the actual number of nodes to add or to remove based on
the current size of the cluster, the specified percentage value, the
constraints (i.e. the min_size
and the max_size
properties).
When computing the new capacity for the cluster, senlin engine will determine the value based on the following rules:
If the value of new capacity is greater than 1.0 or less than -1.0, it will be rounded to the integer part of the value. For example, 3.4 will be rounded to 3, -1.9 will be rounded to -1;
If the value of the new capacity is between 0 and 1, Senlin will round it up to 1;
If the value of the new capacity is between 0 and -1, Senlin will round it down to -1;
The new capacity should be in the range of
min_size
andmax_size
, inclusively, unless option--strict
is specified;The range checking will be performed against the current size constraints if no new value for
min_size
and/ormax_size
is given, or else Senlin will first verify the new size constraints and perform range checking against the new constraints;If option
--min-step
is specified, the<MIN_STEP>
value will be used if the absolute value of the new capacity value is less than<MIN_STEP>
.
If option --strict`
is specified, Senlin will strictly conform to
the cluster size constraints. If the capacity value falls out of the range,
the request will be rejected. When --strict
is set to False
,
Senlin engine will do a resize on a best-effort basis.
Suppose we have a cluster A with min_size
set to 5 and its current size is
7. If the new capacity value is 4 and option --strict
is set to
True
, the request will be rejected with an error message. If the new
capacity value is 4 and the option --strict
is not set, Senlin will
try resize the cluster to 5 nodes.
Along with the openstack cluster resize command, you can specify
the new size constraints using either the option --min-size
or
the option --max-size
or both.
openstack cluster shrink
and openstack cluster expand
¶
The openstack cluster shrink command and the
openstack cluster expand command are provided for convenience when
you want to remove a specific number of nodes from a cluster or add a specific
number of nodes to a cluster, respectively. These two commands both take an
argument <COUNT>
which is a positive integer representing the number of
nodes to add or remove. For example, the following command adds two nodes to
the web_servers
cluster:
$ openstack cluster expand --count 2 web_servers
The following command removes two nodes from the web_servers
cluster:
$ openstack cluster shrink --count 2 web_servers
The option --count
is optional. If this option is specified,
Senlin will use it for cluster size change, even when there are scaling
policies attached to the cluster. If this option is omitted, however, Senlin
will treat it as implicitly set to value 1.
Checking a Cluster¶
A cluster can be checked using the openstack cluster check command, for example:
$ openstack cluster check mycluster
All nodes belonging to the specified cluster will perform the check operation. If a node’s physical resource is not ACTIVE, the node status will be changed as part of the check operation.
Recovering a Cluster¶
A cluster can be recovered using the openstack cluster recover command, for example:
$ openstack cluster recover mycluster --check true
The option --check
is optional. If this option is set,
the cluster will perform check operation before doing recovery. The restore
operation will delete nodes from the specified cluster and recreate it.
Deleting a Cluster¶
A cluster can be deleted using the openstack cluster delete command, for example:
$ openstack cluster delete mycluster
Note that in this command you can use the name, the ID or the “short ID” to
specify the cluster object you want to delete. If the specified criteria
cannot match any clusters, you will get a ResourceNotFound
exception. If
more than one cluster matches the criteria, you will get a MultipleChoices
exception.
When there are nodes in the cluster, the Senlin engine will launch a process to delete all nodes from the cluster and destroy them before deleting the cluster object itself.
See Also¶
There are other operations related to clusters. Please refer to the following links for operations related to cluster membership management and the creation and management of cluster-policy bindings: