openstack-manuals/doc/config-reference/compute/section_compute-scheduler.xml
thorking 076b7a8bec Enable retry for multiple force hosts
If there are multiple force hosts/nodes, retry filter will help
retry on the force hosts/nodes if there are VM boot failure.

Change-Id: I58c95f16e320a03d41988c860e18434412c475c3
Closes-Bug:1435156
2015-10-27 18:09:33 +08:00

1063 lines
55 KiB
XML

<?xml version="1.0" encoding="utf-8"?>
<!DOCTYPE section [
<!ENTITY % openstack SYSTEM "../../common/entities/openstack.ent">
%openstack;
]>
<section xmlns="http://docbook.org/ns/docbook"
xmlns:xi="http://www.w3.org/2001/XInclude"
xmlns:xlink="http://www.w3.org/1999/xlink"
version="5.0"
xml:id="section_compute-scheduler">
<?dbhtml stop-chunking?>
<title>Scheduling</title>
<para>Compute uses the <systemitem class="service"
>nova-scheduler</systemitem> service to determine how to
dispatch compute requests. For example, the
<systemitem class="service">nova-scheduler</systemitem>
service determines on which host a VM should launch. In the
context of filters, the term <firstterm>host</firstterm> means
a physical node that has a <systemitem class="service"
>nova-compute</systemitem> service running on it. You can
configure the scheduler through a variety of options.</para>
<para>Compute is configured with the following default scheduler
options in the <filename>/etc/nova/nova.conf</filename>
file:</para>
<programlisting language="ini">
scheduler_driver_task_period = 60
scheduler_driver = nova.scheduler.filter_scheduler.FilterScheduler
scheduler_available_filters = nova.scheduler.filters.all_filters
scheduler_default_filters = RetryFilter, AvailabilityZoneFilter, RamFilter, DiskFilter, ComputeFilter, ComputeCapabilitiesFilter, ImagePropertiesFilter, ServerGroupAntiAffinityFilter, ServerGroupAffinityFilter</programlisting>
<para>By default, the <option>scheduler_driver</option> is
configured as a filter scheduler, as described in the next
section. In the default configuration, this scheduler
considers hosts that meet all the following criteria:</para>
<itemizedlist>
<listitem>
<para>Have not been attempted for scheduling purposes
(<literal>RetryFilter</literal>).</para>
</listitem>
<listitem>
<para>Are in the requested availability zone
(<literal>AvailabilityZoneFilter</literal>).</para>
</listitem>
<listitem>
<para>Have sufficient RAM available
(<literal>RamFilter</literal>).</para>
</listitem>
<listitem>
<para>Have sufficient disk space available for root and ephemeral storage
(<literal>DiskFilter</literal>).</para>
</listitem>
<listitem>
<para>Can service the request
(<literal>ComputeFilter</literal>).</para>
</listitem>
<listitem>
<para>Satisfy the extra specs associated with the instance
type
(<literal>ComputeCapabilitiesFilter</literal>).</para>
</listitem>
<listitem>
<para>Satisfy any architecture, hypervisor type, or
virtual machine mode properties specified on the
instance's image properties
(<literal>ImagePropertiesFilter</literal>).</para>
</listitem>
<listitem>
<para>
Are on a different host than other instances of a group
(if requested)
(<literal>ServerGroupAntiAffinityFilter</literal>).</para>
</listitem>
<listitem>
<para>
Are in a set of group hosts (if requested)
(<literal>ServerGroupAffinityFilter</literal>).</para>
</listitem>
</itemizedlist>
<para>The scheduler caches its list of available hosts; use the
<option>scheduler_driver_task_period</option> option to
specify how often the list is updated.</para>
<note>
<para>Do not configure <option>service_down_time</option> to
be much smaller than
<option>scheduler_driver_task_period</option>;
otherwise, hosts appear to be dead while the host list is
being cached.</para>
</note>
<para>For information about the volume scheduler, see the Block
Storage section of <link
xlink:href="http://docs.openstack.org/admin-guide-cloud/blockstorage-manage-volumes.html">
<citetitle>OpenStack Cloud Administrator
Guide</citetitle></link>.</para>
<para>The scheduler chooses a new host when an instance is
migrated.</para>
<para>When evacuating instances from a host, the scheduler service
honors the target host defined by the administrator on the evacuate
command. If a target is not defined by the administrator, the
scheduler determines the target host. For information about
instance evacuation, see <link
xlink:href="http://docs.openstack.org/admin-guide-cloud/compute-node-down.html#evacuate-instances"
>Evacuate instances</link> section of the
<citetitle>OpenStack Cloud Administrator
Guide</citetitle>.</para>
<section xml:id="filter-scheduler">
<title>Filter scheduler</title>
<para>The filter scheduler
(<literal>nova.scheduler.filter_scheduler.FilterScheduler</literal>)
is the default scheduler for scheduling virtual machine
instances. It supports filtering and weighting to make
informed decisions on where a new instance should be
created.</para>
</section>
<section xml:id="scheduler-filters">
<?dbhtml stop-chunking?>
<title>Filters</title>
<para>When the filter scheduler receives a request for a
resource, it first applies filters to determine which
hosts are eligible for consideration when dispatching a
resource. Filters are binary: either a host is accepted by
the filter, or it is rejected. Hosts that are accepted by
the filter are then processed by a different algorithm to
decide which hosts to use for that request, described in
the <link linkend="weights">Weights</link> section.</para>
<figure xml:id="filter-figure">
<title>Filtering</title>
<mediaobject>
<imageobject>
<imagedata
fileref="../../common/figures/filteringWorkflow1.png"
scale="80"/>
</imageobject>
</mediaobject>
</figure>
<para>The <option>scheduler_available_filters</option>
configuration option in <filename>nova.conf</filename>
provides the Compute service with the list of the filters
that are used by the scheduler. The default setting
specifies all of the filter that are included with the
Compute service:</para>
<programlisting language="ini">scheduler_available_filters = nova.scheduler.filters.all_filters</programlisting>
<para>This configuration option can be specified multiple
times. For example, if you implemented your own custom
filter in Python called
<literal>myfilter.MyFilter</literal> and you wanted to
use both the built-in filters and your custom filter, your
<filename>nova.conf</filename> file would
contain:</para>
<programlisting language="ini">scheduler_available_filters = nova.scheduler.filters.all_filters
scheduler_available_filters = myfilter.MyFilter</programlisting>
<para>The <literal>scheduler_default_filters</literal>
configuration option in <filename>nova.conf</filename>
defines the list of filters that are applied by the
<systemitem class="service"
>nova-scheduler</systemitem> service. The default
filters are:</para>
<programlisting language="ini">scheduler_default_filters = RetryFilter, AvailabilityZoneFilter, RamFilter, ComputeFilter, ComputeCapabilitiesFilter, ImagePropertiesFilter, ServerGroupAntiAffinityFilter, ServerGroupAffinityFilter</programlisting>
<para>The following sections describe the available
filters.</para>
<section xml:id="aggregate-corefilter">
<title>AggregateCoreFilter</title>
<para>
Filters host by CPU core numbers with a per-aggregate
<literal>cpu_allocation_ratio</literal> value. If the
per-aggregate value is not found, the value falls back
to the global setting. If the host is in more than one
aggregate and more than one value is found, the minimum
value will be used. For information about how to use
this filter, see <xref linkend="host-aggregates"/>. See
also <xref linkend="corefilter"/>.
</para>
</section>
<section xml:id="aggregate-diskfilter">
<title>AggregateDiskFilter</title>
<para>
Filters host by disk allocation with a per-aggregate
<literal>disk_allocation_ratio</literal> value. If the
per-aggregate value is not found, the value falls back to
the global setting. If the host is in more than one
aggregate and more than one value is found, the minimum
value will be used. For information about how to use this
filter, see <xref linkend="host-aggregates"/>. See also
<xref linkend="diskfilter"/>.
</para>
</section>
<section xml:id="aggregate-imagepropertiesisolationfilter">
<title>AggregateImagePropertiesIsolation</title>
<para>Matches properties defined in an image's metadata
against those of aggregates to determine host
matches:</para>
<itemizedlist>
<listitem>
<para>If a host belongs to an aggregate and the
aggregate defines one or more metadata that
matches an image's properties, that host is a
candidate to boot the image's instance.</para>
</listitem>
<listitem>
<para>If a host does not belong to any aggregate,
it can boot instances from all images.</para>
</listitem>
</itemizedlist>
<para>For example, the following aggregate
<systemitem>myWinAgg</systemitem> has the Windows
operating system as metadata (named 'windows'):</para>
<screen><prompt>$</prompt> <userinput>nova aggregate-details MyWinAgg</userinput>
<computeroutput>+----+----------+-------------------+------------+---------------+
| Id | Name | Availability Zone | Hosts | Metadata |
+----+----------+-------------------+------------+---------------+
| 1 | MyWinAgg | None | 'sf-devel' | 'os=windows' |
+----+----------+-------------------+------------+---------------+</computeroutput></screen>
<para>In this example, because the following Win-2012
image has the <property>windows</property> property,
it boots on the <systemitem>sf-devel</systemitem> host
(all other filters being equal):</para>
<screen><prompt>$</prompt> <userinput>glance image-show Win-2012</userinput>
<computeroutput>+------------------+--------------------------------------+
| Property | Value |
+------------------+--------------------------------------+
| Property 'os' | windows |
| checksum | f8a2eeee2dc65b3d9b6e63678955bd83 |
| container_format | ami |
| created_at | 2013-11-14T13:24:25 |
| ...</computeroutput></screen>
<para>You can configure the
<systemitem>AggregateImagePropertiesIsolation</systemitem>
filter by using the following options in the
<filename>nova.conf</filename> file:</para>
<programlisting language="ini"># Considers only keys matching the given
namespace (string). Multiple values can be given, as a comma-separated list.
aggregate_image_properties_isolation_namespace = &lt;None>
# Separator used between the namespace and keys (string).
aggregate_image_properties_isolation_separator = .</programlisting>
</section>
<section xml:id="aggregate-instanceextraspecsfilter">
<title>AggregateInstanceExtraSpecsFilter</title>
<para>Matches properties defined in extra specs for an
instance type against admin-defined properties on a host
aggregate. Works with specifications that are scoped with
<literal>aggregate_instance_extra_specs</literal>. Multiple
values can be given, as a comma-separated list. For
backward compatibility, also works with non-scoped
specifications; this action is highly discouraged because
it conflicts with <link
linkend="computecapabilitiesfilter">
ComputeCapabilitiesFilter</link> filter when you enable
both filters. For information about how to use this
filter, see the <link linkend="host-aggregates">host
aggregates</link> section.</para>
</section>
<section xml:id="aggregate-ioopsfilter">
<title>AggregateIoOpsFilter</title>
<para>
Filters host by disk allocation with a per-aggregate
<literal>max_io_ops_per_host</literal> value. If the
per-aggregate value is not found, the value falls back to
the global setting. If the host is in more than one
aggregate and more than one value is found, the minimum
value will be used. For information about how to use this
filter, see <xref linkend="host-aggregates"/>. See
also <xref linkend="ioopsfilter"/>.
</para>
</section>
<section xml:id="aggregate-multi-tenancy-isolation">
<title>AggregateMultiTenancyIsolation</title>
<para>Isolates tenants to specific <link
linkend="host-aggregates">host aggregates</link>.
If a host is in an aggregate that has the
<literal>filter_tenant_id</literal> metadata key,
the host creates instances from only that tenant or
list of tenants. A host can be in different
aggregates. If a host does not belong to an aggregate
with the metadata key, the host can create instances
from all tenants.</para>
</section>
<section xml:id="aggregate-numinstances-filter">
<title>AggregateNumInstancesFilter</title>
<para>
Filters host by number of instances with a per-aggregate
<literal>max_instances_per_host</literal> value. If the
per-aggregate value is not found, the value falls back to
the global setting. If the host is in more than one
aggregate and thus more than one value is found, the
minimum value will be used. For information about how to
use this filter, see <xref
linkend="host-aggregates"/>. See also <xref
linkend="numinstancesfilter"/>.
</para>
</section>
<section xml:id="aggregate-ram-filter">
<title>AggregateRamFilter</title>
<para>
Filters host by RAM allocation of instances with a per-aggregate
<literal>ram_allocation_ratio</literal> value. If the
per-aggregate value is not found, the value falls back to
the global setting. If the host is in more than one
aggregate and thus more than one value is found, the
minimum value will be used. For information about how to
use this filter, see <xref
linkend="host-aggregates"/>. See also <xref
linkend="ramfilter"/>.
</para>
</section>
<section xml:id="aggregate-typeaffinityfilter">
<title>AggregateTypeAffinityFilter</title>
<para>
This filter passes hosts if no <literal>instance_type</literal>
key is set or the <literal>instance_type</literal>
aggregate metadata value contains the name of the
<literal>instance_type</literal> requested.
The value of the <literal>instance_type</literal> metadata entry
is a string that may contain either a single
<literal>instance_type</literal> name or a comma-separated list
of <literal>instance_type</literal> names, such as
'm1.nano' or "m1.nano,m1.small."
For information about how to use this filter, see <xref
linkend="host-aggregates"/>. See also <xref
linkend="typeaffinityfilter"/>.
</para>
</section>
<section xml:id="allhostsfilter">
<title>AllHostsFilter</title>
<para>This is a no-op filter. It does not eliminate any of
the available hosts.</para>
</section>
<section xml:id="availabilityzonefilter">
<title>AvailabilityZoneFilter</title>
<para>Filters hosts by availability zone. You must enable
this filter for the scheduler to respect availability
zones in requests.</para>
</section>
<section xml:id="computecapabilitiesfilter">
<title>ComputeCapabilitiesFilter</title>
<para>Matches properties defined in extra specs for an
instance type against compute capabilities.</para>
<para>If an extra specs key contains a colon
(<literal>:</literal>), anything before the colon
is treated as a namespace and anything after the colon
is treated as the key to be matched. If a namespace is
present and is not <literal>capabilities</literal>,
the filter ignores the namespace. For backward
compatibility, also treats the extra specs key as the
key to be matched if no namespace is present; this
action is highly discouraged because it conflicts with
<link linkend="aggregate-instanceextraspecsfilter">
AggregateInstanceExtraSpecsFilter</link> filter
when you enable both filters.</para>
</section>
<section xml:id="computefilter">
<title>ComputeFilter</title>
<para>Passes all hosts that are operational and
enabled.</para>
<para>In general, you should always enable this filter.</para>
</section>
<section xml:id="corefilter">
<title>CoreFilter</title>
<para>Only schedules instances on hosts if sufficient CPU
cores are available. If this filter is not set, the
scheduler might over-provision a host based on cores.
For example, the virtual cores running on an instance
may exceed the physical cores.</para>
<para>You can configure this filter to enable a fixed
amount of vCPU overcommitment by using the
<option>cpu_allocation_ratio</option>
configuration option in
<filename>nova.conf</filename>. The default setting
is:</para>
<programlisting language="ini">cpu_allocation_ratio = 16.0</programlisting>
<para>With this setting, if 8 vCPUs are on a node, the
scheduler allows instances up to 128 vCPU to be run on
that node.</para>
<para>To disallow vCPU overcommitment set:</para>
<programlisting language="ini">cpu_allocation_ratio = 1.0</programlisting>
<note>
<para>The Compute API always returns the actual
number of CPU cores available on a compute node
regardless of the value of the
<option>cpu_allocation_ratio</option>
configuration key. As a result changes to the
<option>cpu_allocation_ratio</option> are not
reflected via the command line clients or the
dashboard. Changes to this configuration key are
only taken into account internally in the
scheduler.</para>
</note>
</section>
<section xml:id="numatopologyfilter">
<title>NUMATopologyFilter</title>
<para>Filters hosts based on the NUMA topology that was specified
for the instance through the use of flavor <literal>extra_specs</literal>in
combination with the image properties, as described in detail
in the related nova-spec document: <link
xlink:href="http://specs.openstack.org/openstack/nova-specs/specs/juno/implemented/virt-driver-numa-placement.html">
</link>
Filter will try to match
the exact NUMA cells of the instance to those of the host.
It will consider the standard over-subscription limits each cell,
and provide limits to the compute host accordingly.
</para>
<note>
<para>If instance has no topology defined, it will
be considered for any host. If instance has a topology
defined, it will be considered only for NUMA capable
hosts.
</para>
</note>
</section>
<section xml:id="differenthostfilter">
<title>DifferentHostFilter</title>
<para>Schedules the instance on a different host from a
set of instances. To take advantage of this filter,
the requester must pass a scheduler hint, using
<literal>different_host</literal> as the key and a
list of instance UUIDs as the value. This filter is
the opposite of the <literal>SameHostFilter</literal>.
Using the <command>nova</command> command-line tool,
use the <literal>--hint</literal> flag. For
example:</para>
<screen><prompt>$</prompt> <userinput>nova boot --image cedef40a-ed67-4d10-800e-17455edce175 --flavor 1 \
--hint different_host=a0cf03a5-d921-4877-bb5c-86d26cf818e1 \
--hint different_host=8c19174f-4220-44f0-824a-cd1eeef10287 server-1</userinput></screen>
<para>With the API, use the
<literal>os:scheduler_hints</literal> key. For
example:</para>
<programlisting language="json"><xi:include href="../../common/samples/server-scheduler-hints.json" parse="text"/></programlisting>
</section>
<section xml:id="diskfilter">
<title>DiskFilter</title>
<para>Only schedules instances on hosts if there is
sufficient disk space available for root and ephemeral
storage.</para>
<para>You can configure this filter to enable a fixed
amount of disk overcommitment by using the
<literal>disk_allocation_ratio</literal>
configuration option in the
<filename>nova.conf</filename> configuration file.
The default setting disables the possibility of the
overcommitment and allows launching a VM only if
there is a sufficient amount of disk space available
on a host:</para>
<programlisting language="ini">disk_allocation_ratio = 1.0</programlisting>
<para>DiskFilter always considers the value of the
<option>disk_available_least</option> property and not the
one of the <option>free_disk_gb</option> property
of a hypervisor's statistics:</para>
<screen><prompt>$</prompt> <userinput>nova hypervisor-stats</userinput>
<computeroutput>+----------------------+-------+
| Property | Value |
+----------------------+-------+
| count | 1 |
| current_workload | 0 |
| disk_available_least | 29 |
| free_disk_gb | 35 |
| free_ram_mb | 3441 |
| local_gb | 35 |
| local_gb_used | 0 |
| memory_mb | 3953 |
| memory_mb_used | 512 |
| running_vms | 0 |
| vcpus | 2 |
| vcpus_used | 0 |
+----------------------+-------+</computeroutput></screen>
<para>As it can be viewed from the command output above,
the amount of the available disk space
can be less than the amount of the free disk space.
It happens because the <option>disk_available_least</option>
property accounts for the virtual size rather
than the actual size of images. If you use an image format that
is sparse or copy on write so that each virtual instance
does not require a 1:1 allocation of a virtual disk
to a physical storage, it may be useful to allow the
overcommitment of disk space.</para>
<para>To enable scheduling instances while overcommitting disk
resources on the node, adjust the value of
the <literal>disk_allocation_ratio</literal>
configuration option to greater than <literal>1.0</literal>:</para>
<programlisting language="ini">disk_allocation_ratio > 1.0</programlisting>
<note>
<para>If the value is set to <literal>>1</literal>, we recommend
keeping track of the free disk space, as the value approaching
<literal>0</literal> may result in the incorrect functioning
of instances using it at the moment.</para>
</note>
</section>
<section xml:id="groupaffinityfilter">
<title>GroupAffinityFilter</title>
<note>
<para>This filter is deprecated in favor of <link
linkend="servergroupantiaffinityfilter"
>ServerGroupAffinityFilter</link>.</para>
</note>
<para>The GroupAffinityFilter ensures that an instance is
scheduled on to a host from a set of group hosts. To
take advantage of this filter, the requester must pass
a scheduler hint, using <literal>group</literal> as
the key and an arbitrary name as the value. Using the
<command>nova</command> command-line tool, use the
<literal>--hint</literal> flag. For
example:</para>
<screen><prompt>$</prompt> <userinput>nova boot --image <replaceable>IMAGE_ID</replaceable> --flavor 1 --hint group=foo server-1</userinput></screen>
<para>This filter should not be enabled at the same time
as <link linkend="groupantiaffinityfilter"
>GroupAntiAffinityFilter</link> or neither filter
will work properly.</para>
</section>
<section xml:id="groupantiaffinityfilter">
<title>GroupAntiAffinityFilter</title>
<note>
<para>This filter is deprecated in favor of <link
linkend="servergroupantiaffinityfilter"
>ServerGroupAntiAffinityFilter</link>.</para>
</note>
<para>The GroupAntiAffinityFilter ensures that each
instance in a group is on a different host. To take
advantage of this filter, the requester must pass a
scheduler hint, using <literal>group</literal> as the
key and an arbitrary name as the value. Using the
<command>nova</command> command-line tool, use the
<literal>--hint</literal> flag. For
example:</para>
<screen><prompt>$</prompt> <userinput>nova boot --image <replaceable>IMAGE_ID</replaceable> --flavor 1 --hint group=foo server-1</userinput></screen>
<para>This filter should not be enabled at the same time
as <link linkend="groupaffinityfilter"
>GroupAffinityFilter</link> or neither filter will
work properly.</para>
</section>
<section xml:id="imagepropertiesfilter">
<title>ImagePropertiesFilter</title>
<para>Filters hosts based on properties defined on the
instance's image. It passes hosts that can support the
specified image properties contained in the instance.
Properties include the architecture, hypervisor type,
hypervisor version (for Xen hypervisor type only),
and virtual machine mode.</para>
<para>For example, an instance
might require a host that runs an ARM-based processor,
and QEMU as the hypervisor. You can decorate an image
with these properties by using:</para>
<screen><prompt>$</prompt> <userinput>glance image-update img-uuid --property architecture=arm --property hypervisor_type=qemu</userinput></screen>
<para>The image properties that the filter checks for are:</para>
<itemizedlist>
<listitem>
<para><literal>architecture</literal>:
describes the machine
architecture required by the image. Examples
are <literal>i686</literal>, <literal>x86_64</literal>,
<literal>arm</literal>, and <literal>ppc64</literal>.</para>
</listitem>
<listitem>
<para><literal>hypervisor_type</literal>:
describes the hypervisor required by the image.
Examples are <literal>xen</literal>,
<literal>qemu</literal>, and <literal>xenapi</literal>.</para>
<note><para><literal>qemu</literal> is used for both
QEMU and KVM hypervisor types.</para></note>
</listitem>
<listitem>
<para><literal>hypervisor_version_requires</literal>:
describes the hypervisor version required by the image.
The property is supported for Xen hypervisor type only.
It can be used to enable support
for multiple hypervisor versions,
and to prevent instances with newer
Xen tools from being provisioned on an older version
of a hypervisor. If available, the property value
is compared to the hypervisor version of the compute host.</para>
<para>To filter the hosts by the hypervisor version,
add the <literal>hypervisor_version_requires</literal> property
on the image as metadata and pass an operator and a
required hypervisor version as its value:</para>
<screen><prompt>$</prompt> <userinput>glance image-update img-uuid --property hypervisor_type=xen --property hypervisor_version_requires=">=4.3"</userinput></screen>
</listitem>
<listitem>
<para><literal>vm_mode</literal>: describes the hypervisor application
binary interface (ABI) required by the image.
Examples are <literal>xen</literal> for Xen 3.0 paravirtual
ABI, <literal>hvm</literal> for native ABI,
<literal>uml</literal> for User Mode
Linux paravirtual ABI, <literal>exe</literal>
for container virt executable ABI.</para>
</listitem>
</itemizedlist>
</section>
<section xml:id="isolatedhostsfilter">
<title>IsolatedHostsFilter</title>
<para>Allows the admin to define a special (isolated) set
of images and a special (isolated) set of hosts, such
that the isolated images can only run on the isolated
hosts, and the isolated hosts can only run isolated
images. The flag
<literal>restrict_isolated_hosts_to_isolated_images</literal>
can be used to force isolated hosts to only run
isolated images.</para>
<para>The admin must specify the isolated set of images
and hosts in the <filename>nova.conf</filename> file
using the <literal>isolated_hosts</literal> and
<literal>isolated_images</literal> configuration
options. For example:</para>
<programlisting language="ini">isolated_hosts = server1, server2
isolated_images = 342b492c-128f-4a42-8d3a-c5088cf27d13, ebd267a6-ca86-4d6c-9a0e-bd132d6b7d09</programlisting>
</section>
<section xml:id="ioopsfilter">
<title>IoOpsFilter</title>
<para>
The IoOpsFilter filters hosts by concurrent I/O operations
on it. Hosts with too many concurrent I/O operations will
be filtered out. The <option>max_io_ops_per_host</option>
option specifies the maximum number of I/O intensive
instances allowed to run on a host. A host will be ignored
by the scheduler if more than
<option>max_io_ops_per_host</option> instances in build,
resize, snapshot, migrate, rescue or unshelve task states
are running on it.
</para>
</section>
<section xml:id="jsonfilter">
<title>JsonFilter</title>
<para>The JsonFilter allows a user to construct a custom
filter by passing a scheduler hint in JSON format. The
following operators are supported:<itemizedlist>
<listitem>
<para>=</para>
</listitem>
<listitem>
<para>&lt;</para>
</listitem>
<listitem>
<para>></para>
</listitem>
<listitem>
<para>in</para>
</listitem>
<listitem>
<para>&lt;=</para>
</listitem>
<listitem>
<para>>=</para>
</listitem>
<listitem>
<para>not</para>
</listitem>
<listitem>
<para>or</para>
</listitem>
<listitem>
<para>and</para>
</listitem>
</itemizedlist>The filter supports the following variables:<itemizedlist>
<listitem>
<para><code>$free_ram_mb</code></para>
</listitem>
<listitem>
<para><code>$free_disk_mb</code></para>
</listitem>
<listitem>
<para><code>$total_usable_ram_mb</code></para>
</listitem>
<listitem>
<para><code>$vcpus_total</code></para>
</listitem>
<listitem>
<para><code>$vcpus_used</code></para>
</listitem>
</itemizedlist>Using the <command>nova</command>
command-line tool, use the <literal>--hint</literal>
flag:</para>
<screen><prompt>$</prompt> <userinput>nova boot --image 827d564a-e636-4fc4-a376-d36f7ebe1747 \
--flavor 1 --hint query='[">=","$free_ram_mb",1024]' server1</userinput></screen>
<para>With the API, use the
<literal>os:scheduler_hints</literal> key:</para>
<programlisting language="json"><xi:include href="../../common/samples/server-scheduler-hints2.json" parse="text"/></programlisting>
</section>
<section xml:id="metricsfilter">
<title>MetricsFilter</title>
<para>Filters hosts based on meters
<literal>weight_setting</literal>. Only hosts with the
available meters are passed so that the metrics weigher
will not fail due to these hosts.
</para>
</section>
<section xml:id="numinstancesfilter">
<title>NumInstancesFilter</title>
<para>
Hosts that have more instances running than specified by
the <option>max_instances_per_host</option> option are
filtered out when this filter is in place.
</para>
</section>
<section xml:id="pcipassthroughfilter">
<title>PciPassthroughFilter</title>
<para>
The filter schedules instances on a host if the host has
devices that meet the device requests in the
<literal>extra_specs</literal> attribute for the flavor.
</para>
</section>
<section xml:id="ramfilter">
<title>RamFilter</title>
<para>Only schedules instances on hosts that have
sufficient RAM available. If this filter is not set,
the scheduler may over provision a host based on RAM
(for example, the RAM allocated by virtual machine
instances may exceed the physical RAM).</para>
<para>You can configure this filter to enable a fixed
amount of RAM overcommitment by using the
<literal>ram_allocation_ratio</literal>
configuration option in
<filename>nova.conf</filename>. The default setting
is:</para>
<programlisting language="ini">ram_allocation_ratio = 1.5</programlisting>
<para>This setting enables 1.5&nbsp;GB instances to run on
any compute node with 1&nbsp;GB of free RAM.</para>
</section>
<section xml:id="retryfilter">
<title>RetryFilter</title>
<para>Filters out hosts that have already been attempted
for scheduling purposes. If the scheduler selects a
host to respond to a service request, and the host
fails to respond to the request, this filter prevents
the scheduler from retrying that host for the service
request.</para>
<para>This filter is only useful if the
<literal>scheduler_max_attempts</literal>
configuration option is set to a value greater than
zero.</para>
<para>If there are multiple force hosts/nodes, this filter
helps to retry on the force hosts/nodes if a VM fails
to boot.</para>
</section>
<section xml:id="samehostfilter">
<title>SameHostFilter</title>
<para>Schedules the instance on the same host as another
instance in a set of instances. To take advantage of
this filter, the requester must pass a scheduler hint,
using <literal>same_host</literal> as the key and a
list of instance UUIDs as the value. This filter is
the opposite of the
<literal>DifferentHostFilter</literal>. Using the
<command>nova</command> command-line tool, use the
<literal>--hint</literal> flag:</para>
<screen><prompt>$</prompt> <userinput>nova boot --image cedef40a-ed67-4d10-800e-17455edce175 --flavor 1 \
--hint same_host=a0cf03a5-d921-4877-bb5c-86d26cf818e1 \
--hint same_host=8c19174f-4220-44f0-824a-cd1eeef10287 server-1</userinput></screen>
<para>With the API, use the
<literal>os:scheduler_hints</literal> key:</para>
<programlisting language="json"><xi:include href="../../common/samples/server-scheduler-hints3.json" parse="text"/></programlisting>
</section>
<section xml:id="servergroupaffinityfilter">
<title>ServerGroupAffinityFilter</title>
<para>The ServerGroupAffinityFilter ensures that an
instance is scheduled on to a host from a set of group
hosts. To take advantage of this filter, the requester
must create a server group with an
<literal>affinity</literal> policy, and pass a
scheduler hint, using <literal>group</literal> as the
key and the server group UUID as the value. Using the
<command>nova</command> command-line tool, use the
<literal>--hint</literal> flag. For
example:</para>
<screen><prompt>$</prompt> <userinput>nova server-group-create --policy affinity group-1</userinput>
<prompt>$</prompt> <userinput>nova boot --image <replaceable>IMAGE_ID</replaceable> --flavor 1 --hint group=<replaceable>SERVER_GROUP_UUID</replaceable> server-1</userinput></screen>
</section>
<section xml:id="servergroupantiaffinityfilter">
<title>ServerGroupAntiAffinityFilter</title>
<para>The ServerGroupAntiAffinityFilter ensures that each
instance in a group is on a different host. To take
advantage of this filter, the requester must create a
server group with an <literal>anti-affinity</literal>
policy, and pass a scheduler hint, using
<literal>group</literal> as the key and the server
group UUID as the value. Using the
<command>nova</command> command-line tool, use the
<literal>--hint</literal> flag. For
example:</para>
<screen><prompt>$</prompt> <userinput>nova server-group-create --policy anti-affinity group-1</userinput>
<prompt>$</prompt> <userinput>nova boot --image <replaceable>IMAGE_ID</replaceable> --flavor 1 --hint group=<replaceable>SERVER_GROUP_UUID</replaceable> server-1</userinput></screen>
</section>
<section xml:id="simplecidraffinityfilter">
<title>SimpleCIDRAffinityFilter</title>
<para>Schedules the instance based on host IP subnet
range. To take advantage of this filter, the requester
must specify a range of valid IP address in CIDR
format, by passing two scheduler hints:</para>
<variablelist>
<varlistentry>
<term><literal>build_near_host_ip</literal></term>
<listitem>
<para>The first IP address in the subnet (for
example,
<literal>192.168.1.1</literal>)</para>
</listitem>
</varlistentry>
<varlistentry>
<term><literal>cidr</literal></term>
<listitem>
<para>The CIDR that corresponds to the subnet
(for example,
<literal>/24</literal>)</para>
</listitem>
</varlistentry>
</variablelist>
<para>Using the <command>nova</command> command-line tool,
use the <literal>--hint</literal> flag. For example,
to specify the IP subnet
<literal>192.168.1.1/24</literal></para>
<screen><prompt>$</prompt> <userinput>nova boot --image cedef40a-ed67-4d10-800e-17455edce175 --flavor 1 \
--hint build_near_host_ip=192.168.1.1 --hint cidr=/24 server-1</userinput></screen>
<para>With the API, use the
<literal>os:scheduler_hints</literal> key:</para>
<programlisting language="json"><xi:include href="../../common/samples/server-scheduler-hints4.json" parse="text"/></programlisting>
</section>
<section xml:id="trustedfilter">
<title>TrustedFilter</title>
<para>
Filters hosts based on their trust. Only passes hosts that
meet the trust requirements specified in the instance
properties.
</para>
</section>
<section xml:id="typeaffinityfilter">
<title>TypeAffinityFilter</title>
<para>
Dynamically limits hosts to one instance type. An instance
can only be launched on a host, if no instance with
different instances types are running on it, or if the
host has no running instances at all.
</para>
</section>
</section>
<section xml:id="weights">
<title>Weights</title>
<?dbhtml stop-chunking?>
<para>When resourcing instances, the filter scheduler filters
and weights each host in the list of acceptable hosts. Each
time the scheduler selects a host, it virtually consumes
resources on it, and subsequent selections are adjusted
accordingly. This process is useful when the customer asks
for the same large amount of instances, because weight is
computed for each requested instance.</para>
<para>All weights are normalized before being summed up; the
host with the largest weight is given the highest
priority.</para>
<figure xml:id="figure_weighting-hosts">
<title>Weighting hosts</title>
<mediaobject>
<imageobject>
<imagedata
fileref="../../common/figures/nova-weighting-hosts.png"
/>
</imageobject>
</mediaobject>
</figure>
<para>If cells are used, cells are weighted by the scheduler
in the same manner as hosts.</para>
<para>Hosts and cells are weighted based on the following
options in the <filename>/etc/nova/nova.conf</filename>
file:</para>
<table rules="all" xml:id="table_host-weighting-options">
<caption>Host weighting options</caption>
<col width="10%" title="Section"/>
<col width="25%" title="Option"/>
<col width="60%" title="Description"/>
<thead>
<tr>
<th>Section</th>
<th>Option</th>
<th>Description</th>
</tr>
</thead>
<tbody>
<tr valign="top">
<td>[DEFAULT]</td>
<td><literal>ram_weight_multiplier</literal></td>
<td>By default, the scheduler spreads instances
across all hosts evenly. Set the
<option>ram_weight_multiplier</option>
option to a negative number if you prefer
stacking instead of spreading. Use a
floating-point value.</td>
</tr>
<tr valign="top">
<td>[DEFAULT]</td>
<td><literal>scheduler_host_subset_size</literal></td>
<td>New instances are scheduled on a host that is
chosen randomly from a subset of the N best
hosts. This property defines the subset size
from which a host is chosen. A value of 1
chooses the first host returned by the
weighting functions. This value must be at
least 1. A value less than 1 is ignored, and 1
is used instead. Use an integer value.</td>
</tr>
<tr valign="top">
<td>[DEFAULT]</td>
<td><literal>scheduler_weight_classes</literal></td>
<td>Defaults to
<literal>nova.scheduler.weights.all_weighers</literal>,
which selects the
RamWeigher and MetricsWeigher. Hosts are then weighted and sorted
with the largest weight winning.</td>
</tr>
<tr valign="top">
<td>[DEFAULT]</td>
<td><literal>io_ops_weight_multiplier</literal></td>
<td>Multiplier used for weighing host I/O operations. A negative
value means a preference to choose light workload
compute hosts.</td>
</tr>
<tr valign="top">
<td>[metrics]</td>
<td><literal>weight_multiplier</literal></td>
<td>Multiplier for weighting meters. Use a
floating-point value.</td>
</tr>
<tr valign="top">
<td>[metrics]</td>
<td><literal>weight_setting</literal></td>
<td>Determines how meters are weighted. Use a
comma-separated list of metricName=ratio. For
example: "name1=1.0, name2=-1.0" results in:
<literal>name1.value * 1.0 + name2.value *
-1.0</literal>
</td>
</tr>
<tr valign="top">
<td>[metrics]</td>
<td><literal>required</literal></td>
<td><para>Specifies how to treat unavailable meters:<itemizedlist>
<listitem>
<para>True&mdash;Raises an
exception. To avoid the raised
exception, you should use the
scheduler filter
<literal>MetricFilter</literal> to
filter out hosts with unavailable
meters.</para>
</listitem>
<listitem>
<para>False&mdash;Treated as a
negative factor in the weighting
process (uses the
<option>weight_of_unavailable</option>
option).</para>
</listitem>
</itemizedlist></para></td>
</tr>
<tr valign="top">
<td>[metrics]</td>
<td><literal>weight_of_unavailable</literal></td>
<td>If <option>required</option> is set to False,
and any one of the meters set by
<option>weight_setting</option> is
unavailable, the
<option>weight_of_unavailable</option>
value is returned to the scheduler.</td>
</tr>
</tbody>
</table>
<para>For example:</para>
<programlisting language="ini">[DEFAULT]
scheduler_host_subset_size = 1
scheduler_weight_classes = nova.scheduler.weights.all_weighers
ram_weight_multiplier = 1.0
io_ops_weight_multiplier = 2.0
[metrics]
weight_multiplier = 1.0
weight_setting = name1=1.0, name2=-1.0
required = false
weight_of_unavailable = -10000.0</programlisting>
<table rules="all" xml:id="table_cell-weighting-options">
<caption>Cell weighting options</caption>
<col width="10%" title="Section"/>
<col width="25%" title="Option"/>
<col width="60%" title="Description"/>
<thead>
<tr>
<th>Section</th>
<th>Option</th>
<th>Description</th>
</tr>
</thead>
<tbody>
<tr valign="top">
<td>[cells]</td>
<td><literal>mute_weight_multiplier</literal></td>
<td>Multiplier to weight mute children (hosts which
have not sent capacity or capacity updates for
some time). Use a negative, floating-point
value.</td>
</tr>
<tr valign="top">
<td>[cells]</td>
<td><literal>offset_weight_multiplier</literal></td>
<td>Multiplier to weight cells, so you can specify
a preferred cell. Use a floating point
value.</td>
</tr>
<tr valign="top">
<td>[cells]</td>
<td><literal>ram_weight_multiplier</literal></td>
<td>By default, the scheduler spreads instances
across all cells evenly. Set the
<option>ram_weight_multiplier</option>
option to a negative number if you prefer
stacking instead of spreading. Use a
floating-point value.</td>
</tr>
<tr valign="top">
<td>[cells]</td>
<td><literal>scheduler_weight_classes</literal></td>
<td>Defaults to
<literal>nova.cells.weights.all_weighers</literal>,
which maps to all cell weighers included with
Compute. Cells are then weighted and sorted
with the largest weight winning.</td>
</tr>
</tbody>
</table>
<para>For example:</para>
<programlisting language="ini">[cells]
scheduler_weight_classes = nova.cells.weights.all_weighers
mute_weight_multiplier = -10.0
ram_weight_multiplier = 1.0
offset_weight_multiplier = 1.0</programlisting>
</section>
<section xml:id="chance-scheduler">
<title>Chance scheduler</title>
<?dbhtml stop-chunking?>
<para>As an administrator, you work with the filter scheduler.
However, the Compute service also uses the Chance
Scheduler,
<literal>nova.scheduler.chance.ChanceScheduler</literal>,
which randomly selects from lists of filtered
hosts.</para>
</section>
<section xml:id="utilization-aware-scheduling">
<title>Utilization aware scheduling</title>
<?dbhtml stop-chunking?>
<para>It is possible to schedule VMs using advanced scheduling
decisions. These decisions are made based on enhanced
usage statistics encompassing data like memory cache
utilization, memory bandwidth utilization, or network
bandwidth utilization. This is disabled by default.
The administrator can configure how the metrics are
weighted in the configuration file by using the <literal>
weight_setting</literal> configuration option in
the <filename>nova.conf</filename> configuration file.
For example to configure metric1 with ratio1 and metric2
with ratio2:</para>
<programlisting language="ini">
weight_setting = "metric1=ratio1, metric2=ratio2"
</programlisting>
</section>
<xi:include
href="../../common/section_cli_nova_host_aggregates.xml"/>
<section xml:id="compute-scheduler-config-ref">
<title>Configuration reference</title>
<para>To customize the Compute scheduler, use the
configuration option settings documented in <xref
linkend="config_table_nova_scheduler"/>.</para>
</section>
</section>