二、nova-api
步驟3：nova-api接收請求
nova-api接收請求，也不是隨便怎么來都接收的，而是需要設定rate limits，默認的實現是在ratelimit的middleware里面實現的。
然而有時候，我們希望實現distributed rate-limiting，從而Turnstile是一個不錯的選擇。
https://github.com/klmitch/turnstile
http://pypi.python.org/pypi/turnstile
步驟4：對Token的驗證
步驟5：查看Policy
這兩步已經在keystone的時候研究過
步驟6：檢查quota
nova, neutron, Cinder各有各的quota，並且可以從命令行進行管理
# nova -h | grep quota
quota-class-show List the quotas for a quota class.
quota-class-update Update the quotas for a quota class.
quota-defaults List the default quotas for a tenant.
quota-delete Delete quota for a tenant/user so their quota will
quota-show List the quotas for a tenant/user.
quota-update Update the quotas for a tenant/user.
# nova quota-show
+-----------------------------+-------+
| Quota | Limit |
+-----------------------------+-------+
| instances | 10 |
| cores | 20 |
| ram | 51200 |
| floating_ips | 10 |
| fixed_ips | -1 |
| metadata_items | 128 |
| injected_files | 5 |
| injected_file_content_bytes | 10240 |
| injected_file_path_bytes | 255 |
| key_pairs | 100 |
| security_groups | 10 |
| security_group_rules | 20 |
+-----------------------------+-------+
# cinder -h | grep quota
quota-class-show List the quotas for a quota class.
quota-class-update Update the quotas for a quota class.
quota-defaults List the default quotas for a tenant.
quota-show List the quotas for a tenant.
quota-update Update the quotas for a tenant.
quota-usage List the quota usage for a tenant.
# cinder quota-show 1779b3bc725b44b98726fb0cbdc617b1
+-----------+-------+
| Property | Value |
+-----------+-------+
| gigabytes | 1000 |
| snapshots | 10 |
| volumes | 10 |
+-----------+-------+
# neutron -h | grep quota
quota-delete Delete defined quotas of a given tenant.
quota-list List quotas of all tenants who have non-default quota values.
quota-show Show quotas of a given tenant
quota-update Define tenant's quotas not to use defaults.
# neutron quota-show 1779b3bc725b44b98726fb0cbdc617b1
+---------------------+-------+
| Field | Value |
+---------------------+-------+
| floatingip | 50 |
| network | 10 |
| port | 50 |
| router | 10 |
| security_group | 10 |
| security_group_rule | 100 |
| subnet | 10 |
+---------------------+-------+
推薦下面的文章
openstack nova 基礎知識——Quota（配額管理）
http://www.sebastien-han.fr/blog/2012/09/19/openstack-play-with-quota/
步驟7：在數據庫中創建Instance實例
有關nova的database schema參考下面的文章
http://www.prestonlee.com/2012/05/03/openstack-nova-essex-mysql-database-schema-diagram-and-sql/
MySQL是Openstack中最重要的組件之一，所以在生產環境中High Availability是必須的。
MySQL的HA有下面幾種方式：
http://dev.mysql.com/doc/mysql-ha-scalability/en/index.html
| Requirement | MySQL Replication | MySQL with DRBD with Corosync and Pacemaker | MySQL Cluster |
| Availability | |||
| Platform Support | All Supported by MySQL Server | Linux | All Supported by MySQL Cluster |
| Automated IP Failover | No | Yes | Depends on Connector and Configuration |
| Automated Database Failover | No | Yes | Yes |
| Automatic Data Resynchronization | No | Yes | Yes |
| Typical Failover Time | User / Script Dependent | Configuration Dependent, 60 seconds and Above | 1 Second and Less |
| Synchronous Replication | No, Asynchronous and Semisynchronous | Yes | Yes |
| Shared Storage | No, Distributed | No, Distributed | No, Distributed |
| Geographic redundancy support | Yes | Yes, via MySQL Replication | Yes, via MySQL Replication |
| Update Schema On-Line | No | No | Yes |
| Scalability | |||
| Number of Nodes | One Master, Multiple Slaves | One Active (primary), one Passive (secondary) Node | 255 |
| Built-in Load Balancing | Reads, via MySQL Replication | Reads, via MySQL Replication | Yes, Reads and Writes |
| Supports Read-Intensive Workloads | Yes | Yes | Yes |
| Supports Write-Intensive Workloads | Yes, via Application-Level Sharding | Yes, via Application-Level Sharding to Multiple Active/Passive Pairs | Yes, via Auto-Sharding |
| Scale On-Line (add nodes, repartition, etc.) | No | No | Yes |
要想系統的學習Mysql replication，推薦下面的這本書
《MySQL High Availability Tools for Building Robust Data Centers》
還有一種方式是Mysql + galera，可以搭建Active + Active的Mysql應用
參考下面的兩篇文章
http://www.sebastien-han.fr/blog/2012/04/08/mysql-galera-cluster-with-haproxy/
http://www.sebastien-han.fr/blog/2012/04/01/mysql-multi-master-replication-with-galera/
還有一種常見的HA的技術，就是pacemaker
最底層是通信層corosync/openais
負責cluster中node之間的通信
上一層是Resource Allocation Layer，包含下面的組件：
CRM Cluster Resouce Manager
是總管，對於resource做的任何操作都是通過它。每個機器上都有一個CRM。
CIB Cluster Information Base
CIB由CRM管理，是在內存中的XML數據庫，保存了cluster的配置和狀態。我們查詢出來的configuration都是保存在CIB里面的。nodes, resources, constraints, relationship.
DC Designated Coordinator
每個node都有CRM，會有一個被選為DC，是整個Cluster的大腦，這個DC控制的CIB是master CIB，其他的CIB都是副本。
PE Policy Engine
當DC需要進行一些全局配置的時候，首先由PE根據當前的狀態和配置，計算出將來的狀態，並生成一系列的action，使得cluster從初始狀態變為結果狀態。PE僅僅在DC上運行。
LRM Local Resource Manager
本地的resource管理，調用resource agent完成操作，啟停resource，將結果返回給CRM
再上一層是Resource Layer
包含多個resource agent。resource agent往往是一些shell script，用來啟動，停止，監控resource的狀態。
要弄懂Pacemaker，推薦讀《SUSE high availability guide》
https://www.suse.com/documentation/sle_ha/singlehtml/book_sleha/book_sleha.html
本人做了一些筆記和實驗，請參考
步驟8：創建filter_properties，用於nova scheduler
步驟9：發送RPC給nova-conductor
有關nova-conductor的文章
http://cloudystuffhappens.blogspot.com/2013/04/understanding-nova-conductor-in.html
在Openstack中，RPC的發送是通過RabbitMQ
RabbitMQ可以通過Pacemaker進行HA，當然也可以搭建自己的RabbitMQ Cluster
學習RabbitMQ當然首推《RabbitMQ in Action》
本人也做了一些筆記
RabbitMQ in Action (1): Understanding messaging
RabbitMQ in Action (2): Running and administering Rabbit
RabbitMQ in Action(5): Clustering and dealing with failure
還沒完全讀完，敬請諒解
當然Openstack中對於RabbitMQ的使用，一篇很好的文章是
本人也對RPC的調用過程進行了代碼分析
步驟10：nova-condutor創建request_spec，用於scheduler
步驟11：nova-conductor發送RPC給nova-scheduler
三、nova-scheduler
選擇一個物理機來創建虛擬機，我們稱為schedule的過程
nova scheduler的一個經典的圖如下
就是先Filter再Weighting，其實scheduler的過程在很早就參與了。
步驟13：對Host進行Filtering
Filtering主要通過兩個變量進行，request_spec和filter_properties，而這些變量在前面的步驟，都已經准備好了。
而不同的Filter只是利用這些信息，然后再根據從HostManager統計上來的HostState信息，選出匹配的Host。
request_spec中的第一個信息就是image的properties信息，尤其是當你想支持多種Hypervisor的時候，Xen的image, KVM的image, Hyper-V的image各不相同，如何保證image跑在正確的Hypervisor上？在image里面這種hypervisor_type property就很必要。
請閱讀下面的文章
http://www.cloudbase.it/filtering-glance-images-for-hyper-v/
image properties還會有min_ram, min_disk，只有內存和硬盤夠大才可以。
Flavor里面可以設置extra_specs，這是一系列key-value值，在數據結構中，以instance_type變量實現，可以在里面輸入這個Flavor除了資源需求的其他參數，從而在Filter的時候，可以使用。
host aggregates將所有的Host分成多個Group，當然不同的Group可以根據不同的屬性Metadata划分，一種是高性能和低性能。
在Openstack文檔中，這個例子很好的展示了host aggregates和Flavor extra_specs的配合使用
http://docs.openstack.org/trunk/config-reference/content/section_compute-scheduler.html
Example: Specify compute hosts with SSDs
This example configures the Compute service to enable users to request nodes that have solid-state drives (SSDs). You create a fast-io host aggregate in the nova availability zone and you add the ssd=true key-value pair to the aggregate. Then, you add the node1, and node2 compute nodes to it.
$ nova aggregate-create fast-io nova
+----+---------+-------------------+-------+----------+
| Id | Name | Availability Zone | Hosts | Metadata |
+----+---------+-------------------+-------+----------+
| 1 | fast-io | nova | | |
+----+---------+-------------------+-------+----------+
$ nova aggregate-set-metadata 1 ssd=true
+----+---------+-------------------+-------+-------------------+
| Id | Name | Availability Zone | Hosts | Metadata |
+----+---------+-------------------+-------+-------------------+
| 1 | fast-io | nova | [] | {u'ssd': u'true'} |
+----+---------+-------------------+-------+-------------------+
$ nova aggregate-add-host 1 node1
+----+---------+-------------------+-----------+-------------------+
| Id | Name | Availability Zone | Hosts | Metadata |
+----+---------+-------------------+------------+-------------------+
| 1 | fast-io | nova | [u'node1'] | {u'ssd': u'true'} |
+----+---------+-------------------+------------+-------------------+
$ nova aggregate-add-host 1 node2
+----+---------+-------------------+---------------------+-------------------+
| Id | Name | Availability Zone | Hosts | Metadata |
+----+---------+-------------------+----------------------+-------------------+
| 1 | fast-io | nova | [u'node1', u'node2'] | {u'ssd': u'true'} |
+----+---------+-------------------+----------------------+-------------------+
Use the nova flavor-create command to create the ssd.large flavor called with an ID of 6, 8 GB of RAM, 80 GB root disk, and four vCPUs.
$ nova flavor-createssd.large68192804+----+-----------+-----------+------+-----------+------+-------+-------------+-----------+-------------+ | ID | Name | Memory_MB | Disk | Ephemeral | Swap | VCPUs | RXTX_Factor | Is_Public | extra_specs | +----+-----------+-----------+------+-----------+------+-------+-------------+-----------+-------------+ | 6 | ssd.large | 8192 | 80 | 0 | | 4 | 1 | True | {} | +----+-----------+-----------+------+-----------+------+-------+-------------+-----------+-------------+
Once the flavor is created, specify one or more key-value pairs that match the key-value pairs on the host aggregates. In this case, that is the ssd=true key-value pair. Setting a key-value pair on a flavor is done using the nova flavor-key command.
$ nova flavor-key ssd.large set ssd=true
Once it is set, you should see the extra_specs property of the ssd.large flavor populated with a key of ssd and a corresponding value of true.
$ nova flavor-show ssd.large
+----------------------------+-------------------+
| Property | Value |
+----------------------------+-------------------+
| OS-FLV-DISABLED:disabled | False |
| OS-FLV-EXT-DATA:ephemeral | 0 |
| disk | 80 |
| extra_specs | {u'ssd': u'true'} |
| id | 6 |
| name | ssd.large |
| os-flavor-access:is_public | True |
| ram | 8192 |
| rxtx_factor | 1.0 |
| swap | |
| vcpus | 4 |
+----------------------------+-------------------+
Now, when a user requests an instance with the ssd.large flavor, the scheduler only considers hosts with the ssd=true key-value pair. In this example, these are node1 and node2.
另一個作用是Xen和KVM的POOL分開，有利於XEN進行Live Migration
另一個作用是Windows和Linux的POOL分開，因為Windows是需要收費的，而Linux大多不需要，Windows的收費是按照物理機，而非虛擬機來收費的，所有需要盡量的將windows的虛擬機放到一個物理機上。
Filter_properties的里面scheduler_hints是一個json，里面可以設置任何值，用於Filter的時候使用。
例如JsonFilter
The JsonFilter allows a user to construct a custom filter by passing a scheduler hint in JSON format. The following operators are supported:
- =
- <
- >
- in
- <=
- >=
- not
- or
- and
The filter supports the following variables:
- $free_ram_mb
- $free_disk_mb
- $total_usable_ram_mb
- $vcpus_total
- $vcpus_used
Using the nova command-line tool, use the --hint flag:
$ nova boot --image 827d564a-e636-4fc4-a376-d36f7ebe1747 --flavor 1 --hint query='[">=","$free_ram_mb",1024]' server1
With the API, use the os:scheduler_hints key:
{
"server": {
"name": "server-1",
"imageRef": "cedef40a-ed67-4d10-800e-17455edce175",
"flavorRef": "1"
},
"os:scheduler_hints": {
"query": "[>=,$free_ram_mb,1024]"
}
}
我們可以指定某個物理機，用下面的命令--availability-zone <zone-name>:<host-name>
步驟14：對合適的Hosts進行weighting並且排序
選出了Hosts，接下來就是進行Weighting的操作
Weighting可以根據很多變量來，一般來說Memory和disk是最先需要滿足的，CPU和network io則需要次要考慮，一般來說，對於付錢較少的Flavor，能滿足memory和disk就可以了，對於付錢較多的Flavor，則需要保證其CPU和network io.