You are viewing a plain text version of this content. The canonical link for it is here.
Posted to commits@cloudstack.apache.org by da...@apache.org on 2025/12/01 06:36:37 UTC
(cloudstack-documentation) branch 4.22 updated: Add Deployment planners documentation (#594)
This is an automated email from the ASF dual-hosted git repository.
dahn pushed a commit to branch 4.22
in repository https://gitbox.apache.org/repos/asf/cloudstack-documentation.git
The following commit(s) were added to refs/heads/4.22 by this push:
new 71b45432 Add Deployment planners documentation (#594)
71b45432 is described below
commit 71b45432cc1d3ac34e17f5d6a21cb8cea552e810
Author: Manoj Kumar <ma...@gmail.com>
AuthorDate: Mon Dec 1 12:06:31 2025 +0530
Add Deployment planners documentation (#594)
Co-authored-by: Jithin Raju <ra...@gmail.com>
---
.../_static/images/deployment-planner-diagram.png | Bin 0 -> 73711 bytes
source/adminguide/deployment_planners.rst | 99 +++++++++++++++++++++
source/adminguide/index.rst | 1 +
source/adminguide/vm_volume_allocators.rst | 37 ++++----
4 files changed, 119 insertions(+), 18 deletions(-)
diff --git a/source/_static/images/deployment-planner-diagram.png b/source/_static/images/deployment-planner-diagram.png
new file mode 100644
index 00000000..a041ff0b
Binary files /dev/null and b/source/_static/images/deployment-planner-diagram.png differ
diff --git a/source/adminguide/deployment_planners.rst b/source/adminguide/deployment_planners.rst
new file mode 100644
index 00000000..64cee50d
--- /dev/null
+++ b/source/adminguide/deployment_planners.rst
@@ -0,0 +1,99 @@
+.. Licensed to the Apache Software Foundation (ASF) under one
+ or more contributor license agreements. See the NOTICE file
+ distributed with this work for additional information#
+ regarding copyright ownership. The ASF licenses this file
+ to you under the Apache License, Version 2.0 (the
+ "License"); you may not use this file except in compliance
+ with the License. You may obtain a copy of the License at
+ http://www.apache.org/licenses/LICENSE-2.0
+ Unless required by applicable law or agreed to in writing,
+ software distributed under the License is distributed on an
+ "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY
+ KIND, either express or implied. See the License for the
+ specific language governing permissions and limitations
+ under the License.
+
+
+Deployment Planners
+======================
+
+
+Deployment planners determine *how and where instances* are placed across clusters within a zone.
+A planner builds and orders a *list of candidate clusters* based on a placement strategy such as available capacity, user dispersion, or pod concentration.
+This ordered list is then passed to the *host allocator*, which attempts to deploy the instance following the planner’s priority order.
+
+Administrators can configure the global setting ``vm.deployment.planner`` to define the default deployment planner for the environment.
+This can also be overridden per *Compute Offering*, allowing flexible control over how instances are distributed across the infrastructure.
+
+Available Planners
+------------------
+
+FirstFitPlanner
+~~~~~~~~~~~~~~~
+
+The ``FirstFitPlanner`` ranks all clusters in a zone by their *available (free) capacity*, placing clusters with the most available resources at the top of the list.
+This approach prioritizes capacity-driven placement, ensuring efficient utilization of resources across the zone.
+
+UserDispersingPlanner
+~~~~~~~~~~~~~~~~~~~~~
+
+The ``UserDispersingPlanner`` aims to *spread a user’s instances across multiple clusters*, reducing the impact of any single cluster failure on that user.
+
+#. The planner counts the number of instances in the *Running* or *Starting* state for the user’s account in each cluster.
+#. Clusters are sorted in **ascending order** based on this count, so clusters with fewer instances from the user are preferred.
+#. The global setting ``vm.user.dispersion.weight`` (default: ``1``) controls how strongly dispersion affects ordering:
+ * ``1``: Ranking is based entirely on dispersion.
+ * ``< 1``: Available capacity has more influence in placement decisions.
+
+Lowering the dispersion weight allows a balance between *even distribution* and *efficient capacity usage*.
+
+UserConcentratedPodPlanner
+~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+The ``UserConcentratedPodPlanner`` focuses on *pod-level affinity*, preferring pods where the user already has active instances.
+
+#. The planner identifies all pods in the zone that contain *Running* instances for the user’s account.
+#. Pods are sorted in **descending order** by the number of user instances — pods with more user instances come first.
+#. Clusters from these pods are then added to the top of the list in that order, so deployment is biased toward pods where the user is already active.
+#. Clusters within each pod are *not* further sorted by capacity or instance count.
+#. If no pods contain user instances, the cluster order remains unchanged.
+
+Summary of Planner Behavior
+---------------------------
+
+.. list-table::
+ :header-rows: 1
+
+ * - Planner
+ - Placement Focus
+ - Ordering Criteria
+ - Typical Use Case
+ * - FirstFitPlanner
+ - Capacity
+ - Descending by available resources
+ - Capacity-optimized or general-purpose placement
+ * - UserDispersingPlanner
+ - Dispersion
+ - Ascending by user instance count (optionally weighted with capacity)
+ - Distribute user instances evenly across clusters
+ * - UserConcentratedPodPlanner
+ - Pod Affinity
+ - Descending by user instance count per pod
+ - Keep user instances within the same pod for locality or data proximity
+
+Pod-Level vs Cluster-Level Allocation
+------------------------------------
+
+When ``apply.allocation.algorithm.to.pods = true``:
+The allocation algorithm (for example, *FirstFit*) is applied at *pod granularity* first.
+The planner will evaluate or rank pods according to the allocation heuristics — for *FirstFit*, that means prioritizing pods with more available capacity according to the FirstFit capacity checks.
+After pods are ordered, the planner then considers clusters *inside each pod* — typically evaluating clusters within the selected pod in order (or applying cluster-level heuristics only within that pod).
+In other words, *pod-level ordering happens before cluster selection*.
+
+When ``apply.allocation.algorithm.to.pods = false`` (the default in many deployments):
+The allocation algorithm operates at the *cluster level* across the entire zone.
+
+|deployment-planner-diagram.png|
+
+.. |deployment-planner-diagram.png| image:: /_static/images/deployment-planner-diagram.png
+ :alt: Deployment Planner Diagram
diff --git a/source/adminguide/index.rst b/source/adminguide/index.rst
index 040ad1cd..1c12ffa9 100644
--- a/source/adminguide/index.rst
+++ b/source/adminguide/index.rst
@@ -138,6 +138,7 @@ Managing VM and Volume Allocation
.. toctree::
:maxdepth: 4
+ deployment_planners
host_and_storage_tags
arch_types
vm_volume_allocators
diff --git a/source/adminguide/vm_volume_allocators.rst b/source/adminguide/vm_volume_allocators.rst
index c15ebd87..dded93ec 100644
--- a/source/adminguide/vm_volume_allocators.rst
+++ b/source/adminguide/vm_volume_allocators.rst
@@ -37,14 +37,14 @@ VM allocator supports following algorithms to select a host in the cluster:
.. cssclass:: table-striped table-bordered table-hover
============================= ========================
-Algorithm Description
+Algorithm Description
============================= ========================
-random Selects a host in the cluster randomly.
-firstfit Selects the first available host in the cluster.
+random Selects a host in the cluster randomly.
+firstfit Selects the first available host in the cluster.
userdispersing Selects the host running least instances for the account, aims to spread out the instances belonging to a single user account.
-userconcentratedpod_random Selects the host randomly aiming to keep all instances belonging to single user account in same pod.
-userconcentratedpod_firstfit Selects the first suitable host from a pod running most instances for the user.
-firstfitleastconsumed Selects the first host after sorting eligible hosts by least allocated resources (such as CPU or RAM).
+userconcentratedpod_random Behaves same as random algorithm.
+userconcentratedpod_firstfit Behaves same as firstfit algorithm.
+firstfitleastconsumed Selects the first host after sorting eligible hosts by least allocated resources (such as CPU or RAM).
============================= ========================
Use global configuration parameter:
@@ -62,14 +62,14 @@ Volume allocator supports following algorithms to select a host in the cluster:
.. cssclass:: table-striped table-bordered table-hover
============================= ========================
-Algorithm Description
+Algorithm Description
============================= ========================
-random Selects a storage pool in the cluster randomly.
-firstfit Selects the first available storage pool in the cluster.
-userdispersing Selects the storage pool running least instances for the account, aims to spread out the instances belonging to a single user account.
-userconcentratedpod_random Selects the storage pool randomly aiming to keep all instances belonging to single user account in same pod.
-userconcentratedpod_firstfit Selects the first suitable pool from a pod running most instances for the user.
-firstfitleastconsumed Selects the first storage pool after sorting eligible pools by least allocated resources.
+random Selects a storage pool in the cluster randomly.
+firstfit Selects the first available storage pool in the cluster.
+userdispersing Selects the storage pool running least instances for the account, aims to spread out the instances belonging to a single user account.
+userconcentratedpod_random Behaves same as random algorithm.
+userconcentratedpod_firstfit Behaves same as firstfit algorithm.
+firstfitleastconsumed Selects the first storage pool after sorting eligible pools by least allocated resources.
============================= ========================
.. note::
@@ -98,11 +98,11 @@ Key: `host.capacityType.to.order.clusters`
.. cssclass:: table-striped table-bordered table-hover
========= ========================
-Value Behavior
+Value Behavior
========= ========================
-CPU Prioritizes resources with the most available CPU.
-RAM Prioritizes resources with the most available memory.
-COMBINED Uses a weighted formula to balance CPU and RAM in prioritization.
+CPU Prioritizes resources with the most available CPU.
+RAM Prioritizes resources with the most available memory.
+COMBINED Uses a weighted formula to balance CPU and RAM in prioritization.
========= ========================
**Additional Configuration for COMBINED**
@@ -132,8 +132,9 @@ Example Configuration
Above config prioritizes CPU at 70% weight and RAM at 30% when ranking pods, clusters, and hosts.
.. note::
- - `host.capacityType.to.order.clusters` is only respected for host ordering when:
+ - `host.capacityType.to.order.clusters` is only respected for cluster/host ordering when:
.. code:: bash
+ vm.deployment.planner: FirstFitPlanner, UserDispersingPlanner (when vm.user.dispersion.weight is < 1)
vm.allocation.algorithm: firstfitleastconsumed
- When using COMBINED, make sure to tune cpu.to.memory.capacity.weight to reflect your environment’s resource constraints and workload profiles.