Jump to: navigation, search

KeystoneCentralizedQuotaManagement

Revision as of 14:19, 4 July 2013 by Drusskikh (talk | contribs) (REST API)

Introduction

In Keystone v3 (Grizzly release), Domains encapsulates users and projects into logical entities that can represent accounts, organizations, etc. Currently there is no capability or mechanism to manage or enforce quotas at domain level. Assigning or updating quota values or limits to a domain will allow the cloud administrator to evaluate domain lists and consumption. In order to achieve these capabilities it will be required to implement quota management for Keystone domains. The goal of this blueprint is to support quotas at the OpenStack Domain level. The design of the feature models, as far as possible, the style of project quotas.


Openstack Quotas

Today OpenStack services make usage of quotas to limit the project resources. For example, the “Instances” quota represents the number of instances that can be created in a project. The table below summarizes the existing project quotas.

quotas type default values description
nova.instances reservable 10 number of instances allowed per project
nova.cores reservable 20 number of instance cores allowed per project
nova.ram reservable 50*1024 megabytes of instance ram allowed per project
nova.floating_ips reservable 10 number of floating ips allowed per project
nova.fixed_ips reservable -1 number of fixed ips allowed per project
nova.metadata_items absolute 128 number of metadata items allowed per instance
nova.injected_files absolute 5 number of injected files allowed
nova.injected_files_content_bytes absolute 10*1024 number of bytes allowed per injected file
nova.injected_file_path_bytes absolute 255 number of bytes allowed per injected file path
nova.security_groups reservable 10 number of security groups per project
nova.security_groups_rules countable 20 number of security rules per security group
nova.key_pairs countable 100 number of key pairs per user
cinder.volumes reservable 10 number of volumes allowed per project
cinder.snapshots reservable 10 number of volume snapshots allowed per project
cinder.gigabytes reservable 1000 number of volume gigabytes (snapshots are also included) per project
quantum.network countable 10 Number of networks allowed per tenant
quantum.subnet countable 10 Number of subnets allowed per tenant
quantum.port countable 50 number of ports allowed per tenant

User Stories

Domain Quotas might impact partitioned OpenStack deployments (regions, cells, etc). Here we consider only the impact on regions. These can be

  1. Per–Region domain quotas
  2. Across region domain quotas

The 1st approach works similar to the current implementation of quotas per project; in a multi-region scenario, the domain quotas are enforced by each service in a non-centralized fashion. The 2nd approach, a domain quota is shared dynamically among regions, e.g. if a service from a given region needs more quota than the others, it requests more quotas. This blueprint addresses the per–region domain quotas. The user stories are listed below:

  • As a cloud administrator, I want to create a domain with default domain quotas
  • As a cloud/domain administrator, I want to see the domain quotas for a domain in a region
  • As a cloud/domain administrator, I want to see the domain quotas usage for a domain in a region
  • As a cloud administrator, I want to update the quotas for a domain in a region
  • As a cloud administrator, I want to delete the quotas for a domain in a region

Since quotas deals with sensible aspects of resource consumption, we identified the need to log the interactions of users when they manage domain quotas.

Design

Our proposal have 2 main blocks: one in keystone we are calling the Domain Quota Proxy (DQP), the other we are calling Domain Quota Driver (DQD). The DQP is responsible to give to user one point of domain quota management, so it acts as a proxy. The DQP is a set of extensions that can be improved to serve as a single point of management for other quotas. The DQD is a piece of code located in the quota module of Nova, Cinder and Quantum projects, and it's designed in the same fashion as the other drivers present in such module.


Domainquota.png

Quota modules need to be refactored to add DQD. Also we should extend the services REST API to provide domain quotas usage to Domain Quota Proxy. The DQP is designed to be implemented as a discrete extension and not included in the default pipeline of Keystone. The Domain Quota Driver design is similar to the current quota driver from quota.py module, given the possibility to the user to option to use it or not; they will be responsible to enforce all quotas listed in the table above in the context of domains.

REST API

Get resource list.

GET v3/os-quotas/resources
Content-Type application/json
Accept application/json

Normal Response Code: 200
Error Response Codes: Unauthorized (401)
Response:

{
    "resources": [
        "nova.instances",
        "nova.cores",
        "nova.ram",
        "cinder.volumes"
    ]
}


Create resource.

POST v3/os-quotas/resources
Content-Type application/json
Accept application/json

Normal Response Code: 200
Error Response Codes: Unauthorized (401)
Request:

{
    "resource": {
        "name": "nova.instances",
        "default_value": 10
    }
}


Get resource.

GET v3/os-quotas/resources/[resource-id]
Content-Type application/json
Accept application/json

Normal Response Code: 200
Error Response Codes: Unauthorized (401)
Response:

{
    "resource": {
        "name": "nova.instances",
        "default_value": 10
    }
}


Update resource.

PUT v3/os-quotas/resources/[resource-id]
Content-Type application/json
Accept application/json

Normal Response Code: 200
Error Response Codes: Unauthorized (401)
Request:

{
    "resource": {
        "name": "nova.instances",
        "default_value": 10
    }
}

Response:

{
    "resource": {
        "name": "nova.instances",
        "default_value": 10
    }
}


Delete resource.

DELETE v3/os-quotas/resources/[resource-id]
Content-Type application/json
Accept application/json

Normal Response Code: 200
Error Response Codes: Unauthorized (401)


Get quota list.

GET v3/os-quotas/quotas/[subject-type]/[subject-id]
Content-Type application/json
Accept application/json

Normal Response Code: 200
Error Response Codes: Unauthorized (401)
Response:

{
    "quotas": [
        {
            "id": "000-id-000",
            "resource-name": "nova.ram",
            "limit": 1024
        },
        {
            "id": "111-id-111",
            "resource-name": "nova.vcpu",
            "limit": 16
        },
    ]
}


Create quota.

POST v3/os-quotas/quotas/[subject-type]/[subject-id]
Content-Type application/json
Accept application/json

Normal Response Code: 200
Error Response Codes: Unauthorized (401)
Request:

{
    "quota": {
        "resource-name": "nova.ram",
        "limit": 1024
    }
}
>

Response:

{
    "quota": {
        "resource-name": "nova.ram",
        "limit": 1024
    }
}

Implementation

For Domain Quota Proxy, information will be stored in new tables in keystone which are as shown below 

DB Tables Domain Quota Proxy.png

Tables are made generic enough so as to accommodate quotas for any resource of any service. 'resources' table stores the information required for the resources. 'quotas' table is capable enough to store quota for any entity (and not just domain). Entity for which a quota is set is called as child in generic terms. Entity responsible for managing quota of a child is called parent. A child and a parent, need not be described just by a single field like user-id alone. There can be multiple fields (like user-id, role, project-id, etc.) and its values which in combination describes a child or a parent. This can be seen as best represented in general by a dictionary. For simple DB implementation, it is suggested to store this dictionary as separate key-value data in separate tables. These tables are 'child_field_data' and 'parent_field_data' tables.

Details of fields of various DB tables is mentioned below,

Resources Table

Column Description
id primary key
name name of the resource in the format <Service-Name>.<Resource Name>. For eg., nova.instances
parameters list of additional parameters required to actually track the usage and enforce quotas. For eg. Service-Group-Name for the nova quota of secuity-rules-per-group
child_identity a set of list of fields which should be present in order to set quota for a child (or entity). In case of match with multiple lists, list with maximum number of fields is considered. For eg. for domain quota, domain and region should be there. For project quota, project, region and domain should be there. For user quota, user, role, project should be there.
type ENUM('Absolute','Countable';'Reservable')
purge_quota_after time in seconds after which the closed records in the quota table should be purged. Useful for managing book keeping.
created_at time at which the record is created
created_by dictionary having minimum keys as user and role. This indicates who created the record.
closed_at time at which the record is created
closed_by dictionary having minimum keys as user and role.. This indicates who closed the record.

Quotas Table

Column Description
id primary key
resource_id foreign key to resources table
ceiling absolute quota limit
available indicates available quota. Calculated as 'ceiling' – (resources used by corresponding child )
created_at time at which the record is created
created_by dictionary having minimum keys as user and role. This indicates who created the record.
closed_at time at which the record is created
closed_by dictionary having minimum keys as user and role.. This indicates who closed the record.

Child_Field_Data Table

Column Description
id primary key
quota_id foreign key to quotas table. This helps in getting the quota for a child mentioned in this table.
key name of an attribute of child, like user-id, role-id, etc.
value value of the attribute of child.

Parent_Field_Data Table

Column Description
id primary key
quota_id foreign key to quotas table. This helps in getting the child information managed by the parent mentioned in this table (or vice-versa).
key name of an attribute of parent, like user-id, role-id, etc.
value value of the attribute of parent.


For history tracking, updates done in table <table-name> (orange heading) will be stored in the corresponding history table h_<table-name> (blue heading). For eg., 'h_quotas' table will be used for keeping track of updates done in the table 'quotas'.

For Domain Quota Driver, the information will be stored in new tables in services which are as shown below 
DB Tables Domain Quota Driver.png
Columns of 'domain_quotas' table are similar to columns of 'quotas' table in keystone mentioned above. It will be used to store only domain quotas, as there are other tables already present to store other quotas. This can be used in future to store all types of quotas in a single table. 'resource_id' is a foreign key to already existing resource table in corresponding service.

Open Stack Quota References

This is a list of URLs of work on quotas within OpenStack.

Topic URI Notes
Per-user quotas support https://blueprints.launchpad.net/nova/+spec/per-user-quotas Blocked to be re-introduced in Grizzly
Stores Quotas centrally in Keystone http://wiki.openstack.org/KeystoneStoreQuotaData
Quantum Quotas http://docs.openstack.org/api/openstack-network/2.0/content/List_Quotas.html
Swift Quotas https://blueprints.launchpad.net/swift/+spec/storage-quotas Implementation status unknown
Quota Project: An effective way to manage the usage of your Swift-based storage cloud http://www.zmanda.com/blogs/?cat=22
Update Limits and Quotas to Key On volume_type https://blueprints.launchpad.net/cinder/+spec/quotas-limits-by-voltype Not started, grizzly-3 target
Demystifying OpenStack Folsom Quotas http://ops.anthonygoddard.com/OpenStack/demystifying-openstack-folsom-quotas/ Explains Cinder/Nova quota separation
per-user quotas support https://blueprints.launchpad.net/nova/+spec/per-user-quotas Blocked by bug https://bugs.launchpad.net/nova/+bug/1034384. It will be re-introduced into Grizzly