OpenStack Ceilometer中的Pipeline机制

Pipeline作用

Pipeline翻译过来是管道的意思,它在ceilometer中的作用类似一个过滤器一样,或者说是转换器。它是一般是一个方法链,这个方法链前面一部分是transformer,transformer实现数据转换等功能,它可以有多个。在链尾是publisher,它负责将数据发送到AMQP中去。

Pipeline定义

在Agent的构造函数中,第一个创建的属性就是pipeline_manager

self.pipeline_manager = pipeline.setup_pipeline(
    transformer.TransformerExtensionManager(
        'ceilometer.transformer',
    ),
    publisher.PublisherExtensionManager(
        'ceilometer.publisher',
    ),
)

其中,transformer和publisher来自setup.cfg中

ceilometer.transformer =
    accumulator = ceilometer.transformer.accumulator:TransformerAccumulator

ceilometer.publisher =
    meter_publisher = ceilometer.publisher.meter:MeterPublisher
    meter = ceilometer.publisher.meter:MeterPublisher
    udp = ceilometer.publisher.udp:UDPPublisher

Pipeline设置

它调用了ceilometer.pipeline中的setup_pipline(),setup_pipeline()通过导入pipeline.yaml,获得pipeline的配置,默认配置如下

name: meter_pipeline
interval: 600
counters:
    - "*"
transformers:
publishers:
    - meter

最后它创建了一个PipelineManager给self.pipeline_manager

PipelineManager(pipeline_cfg,transformer_manager,publisher_manager)

PipelineManager做的事情如下:

self.pipelines = [Pipeline(pipedef, publisher_manager,transformer_manager) for pipedef in cfg]

它遍历cfg中对pipeline的定义(基本都是一个),然后生成一个Pipeline对象数组

def __init__(self, cfg, publisher_manager, transformer_manager):
    self.cfg = cfg
    self.name = cfg['name']
    self.interval = int(cfg['interval'])
    self.counters = cfg['counters']
    self.publishers = cfg['publishers']
    self.transformer_cfg = cfg['transformers'] or []
    self.publisher_manager = publisher_manager
    self._check_counters()
    self._check_publishers(cfg, publisher_manager)
    self.transformers = self._setup_transformers(cfg, transformer_manager)

Pipeline的构造函数如上,它的作用是处理transformer和publisher

Pipeline使用

pipeline的使用位置在agent.py中

def setup_polling_tasks(self):
    polling_tasks = {}
    for pipeline, pollster in itertools.product(
            self.pipeline_manager.pipelines,
            self.pollster_manager.extensions):
        for counter in pollster.obj.get_counter_names():
            if pipeline.support_counter(counter):
                polling_task = polling_tasks.get(pipeline.interval, None)
                if not polling_task:
                    polling_task = self.create_polling_task()
                    polling_tasks[pipeline.interval] = polling_task
                polling_task.add(pollster, [pipeline])
                break

    return polling_tasks

首先通过product生成pipeline和pollster的笛卡尔积,即将每一个pollster都和pipeline配对(一般只有一个pipeline)。

pipeline.support_counter(counter)用来检查这个counter是否同意进入pipeline

另外,每一个polling_task都在构造函数中

self.publish_context = pipeline.PublishContext(
    agent_manager.context,
    cfg.CONF.counter_source)

声明了一个pipeline.PublishContext()

在执行task.poll_and_publish前,会先执行

def add(self, pollster, pipelines):
    self.publish_context.add_pipelines(pipelines)
    self.pollsters.update([pollster])

即增加一个pipeline管理

最后是publish_context的使用位置

def poll_and_publish_instances(self, instances):
    with self.publish_context as publisher:
        for instance in instances:
            if getattr(instance, 'OS-EXT-STS:vm_state', None) != 'error':
                for pollster in self.pollsters:
                    publisher(list(pollster.obj.get_counters(
                        self.manager,
                        instance)))

这里用了with as作为pipeline的管理

__enter__()中,定义了一个函数

def p(counters):
    for p in self.pipelines:
        p.publish_counters(self.context,
                           counters,
                           self.source)

这个函数执行pipeline中的publish_counters,然后最终的执行代码来自

ext.obj.publish_counters(ctxt, counters, source)

即publisher的publish_counters,在这里是ceilometer.publisher.meter:publish_counters,它负责将数据发送到AMQP中去

总结

Pipeline机制一定程度上保证了数据的安全性,并且可以统一数据格式,了解它对于了解Ceilometer的数据流有一定帮助

Loading Disqus comments...
Table of Contents