overview

kubernetes集群中scheduler对pod各个阶段的调度耗时也展示了scheduler的性能，目前在社区中没有找到相关的监控项，于是本文将使用filebeat+logstash+influxdb+grafana技术栈将Metres展示监控起来。

kube-scheduler.INFO相关Trace日志

I0117 18:08:43.224106   87811 trace.go:76] Trace[1863564834]: "Scheduling fat-app/xxxxxxxx-xxxxxxxx-3404-0" (started: 2019-01-17 18:08:43.106820332 +0800 CST m=+116911.416416886) (total time: 117.213043ms):
Trace[1863564834]: [117.187µs] [117.187µs] Computing predicates
Trace[1863564834]: [2.093583ms] [1.976396ms] Prioritizing
Trace[1863564834]: [117.172797ms] [115.079214ms] Selecting host
Trace[1863564834]: [117.213043ms] [40.246µs] END

filebeat配置

filebeat.prospectors:
- type: log
  paths:
    - /var/log/kubernetes/kube-scheduler.*.INFO.*
  include_lines: ['trace','Trace']
# pattern: 正则表达式
# negate: true 或 false(默认是false)，false表示匹配pattern的行合并到上一行；true表示不匹配pattern的行合并到上一行
#  match:  after 或 before，合并到上一行的末尾或开头
  multiline:
    pattern: '^Trace\[[0-9]+\]'
    negate: false
    match: after
  fields:
    tag: scheduler  #添加tag
  fields_under_root: true
  scan_frequency: 10s
  ignore_older: 6h
  close_inactive: 5m
  close_removed: true
  clean_removed: true
  tail_files: false

fields:
  zone: $zone_name  #添加自己的环境信息
fields_under_root: true

output.logstash:
  enabled: true
  hosts:
    - $logstash_host:5044  #logstash的地址

logging.level: info
logging.metrics.enabled: false
http.enabled: true
http.host: localhost
http.port: 5066

setup.dashboards.enabled: false
setup.template.enabled: false

path.home: /usr/share/filebeat
path.data: /var/lib/filebeat
path.logs: /var/log/filebeat
filebeat.registry_file: ${path.data}/registry
max_procs: 2

note:

• debug时可通过 filebeat -e -c filebeat.yml -d "publish"来确认采集的信息是否正确
• 若有如下报错，则需要将filebeat stop掉，将/var/lib/filebeat/registry 删除，重启filebeat重新注册即可

ERROR registrar/registrar.go:346 Writing of registry returned error: rename /var/lib/filebeat/registry.new /var/lib/filebeat/registry: no such file or directory. Continuing...

Logstash的安装及配置

install logstash

• 下载logstash安装

$ wget https://artifacts.elastic.co/downloads/logstash/logstash-6.5.4.rpm
$ rpm -ivh

• 到logstash的path目录下，链接/etc/logstash

$ cd /usr/share/logstash/config
$ ln -s /etc/logstash/* .

• 更改logstash相关目录权限

$ chown -R /etc/logstash
$ chown -R /usr/share/logstash
$ chown -R /var/lib/logstash

• 安装所需插件，详细见下文

$ cd /usr/share/logstash/
$ bin/logstash-plugin install logstash-output-influxdb

• 检查配置文件是否正常

$ bin/logstash -f /etc/logstash/logstash.conf --config.test_and_exit

• 启动

$ bin/logstash -f /etc/logstash/logstash.conf --config.reload.automatic

install-plugin

• logstash-filter-grok
• logstash-output-influxdb
  插件需要单独安装，安装方式如果能不能连接外网则可采用以下方式
  https://gems.ruby-china.com/

（1）尽可能用比较新的 RubyGems 版本，建议 2.6.x 以上

$ yum install -y gem
$ gem update --system # 这里请翻墙一下
$ gem -v
  2.6.3

（2） 更换到国内的源，确保只有 gems.ruby-china.com

$ gem sources --add https://gems.ruby-china.com/ --remove https://rubygems.org/
$ gem sources -l
  https://gems.ruby-china.com

（3）安装单个插件

$ gem install logstash-output-influxdb

（4）安装所有Gemfile中的插件

# 在Gemfile中添加想要安装的plugin
  gem 'logstash-output-influxdb', '~> 5.0', '>= 5.0.5'
$ bin/logstash-plug ininstall --no-verify

self-defined grok-pattern

KUBESCHEDULER .*Scheduling %{NOTSPACE:namespace:tag}/%{NOTSPACE:podname:tag}\".*total time: %{NUMBER:total_time}%{NOTSPACE:total_mesurement}\).*Trace\[\S+\].*\[\S+\].*\[%{NUMBER:computing_time}%{NOTSPACE:computing_mesurement}\]\s+Computing %{WORD:ComputingPredicates:tag}.*Trace\[\S+\].*\[\S+\].*\[%{NUMBER:prioritizing_time}%{NOTSPACE:prioritizing_mesurement}\]\s+%{WORD:Prioritizing:tag}.*Trace\[\S+\].*\[\S+\].*\[%{NUMBER:SelectingHost_time}%{NOTSPACE:SelectingHost_mesurement}\]\s+%{WORD:SelectingHost:tag}.**Trace\[\S+\].*\[\S+\].*\[%{NUMBER:end_time}%{NOTSPACE:end_mesurement}\]\s+%{WORD:END:tag}

KUBESCHEDULERSHORT .*Scheduling %{NOTSPACE:namespace:tag}/%{NOTSPACE:podname:tag}\".*total time: %{NUMBER:total_time}%{NOTSPACE:total_mesurement}\).*Trace\[\S+\].*\[\S+\].*\[%{NUMBER:computing_time}%{NOTSPACE:computing_mesurement}\]\s+Computing %{WORD:ComputingPredicates:tag}.**Trace\[\S+\].*\[\S+\].*\[%{NUMBER:end_time}%{NOTSPACE:end_mesurement}\]\s+%{WORD:END:tag}

logstash.conf配置文件

input {
    beats {
        port => "5044"
        client_inactivity_timeout => "60" #默认60s
    }   
}
filter {
    grok {
        # 需要安装grok插件，并且pattern能正确匹配才能有输出
        patterns_dir => ["/usr/share/logstash/config/pattern"] #自定义的grok pattern
        match => {
                   "message" => ["%{KUBESCHEDULER}", "%{KUBESCHEDULERSHORT}"]
                 }
        remove_field => ["message"]

    } 
    ruby {
    # logstash的算术运算操作，统一运算单位
        code => "
            total_unit = event.get('total_mesurement')
            computing_unit = event.get('computing_mesurement')
            prioritizing_unit = event.get('prioritizing_mesurement')
            selecting_unit = event.get('SelectingHost_mesurement')
            ending_unit = event.get('end_mesurement')
            if total_unit == 'ms'
                event.set('total_time',(event.get('total_time').to_f*1000))
            elsif total_unit == 's'
                event.set('total_time',(event.get('total_time').to_f*1000000))
            else
                event.set('total_time',(event.get('total_time').to_f))
            end
            if computing_unit == 'ms'
                event.set('computing_time',(event.get('computing_time').to_f*1000))
            elsif computing_unit == 's'
                event.set('computing_time',(event.get('computing_time').to_f*1000000))
            else
                event.set('computing_time',(event.get('computing_time').to_f))
            end
            if prioritizing_unit == 'ms'
                event.set('prioritizing_time',(event.get('prioritizing_time').to_f*1000))
            elsif prioritizing_unit == 's'
                event.set('prioritizing_time',(event.get('prioritizing_time').to_f*1000000))
            else
                event.set('prioritizing_time',(event.get('prioritizing_time').to_f))
            end
            if selecting_unit == 'ms'
                event.set('SelectingHost_time',(event.get('SelectingHost_time').to_f*1000))
            elsif selecting_unit == 's'
                event.set('SelectingHost_time',(event.get('SelectingHost_time').to_f*1000000))
            else
                event.set('SelectingHost_time',(event.get('SelectingHost_time').to_f))
            end
            if ending_unit == 'ms'
                event.set('end_time',(event.get('end_time').to_f*1000))
            elsif ending_unit == 's'
                event.set('end_time',(event.get('end_time').to_f*1000000))
            else
                event.set('end_time',(event.get('end_time').to_f))
            end
           "   
        remove_field => ["total_mesurement","computing_mesurement","prioritizing_mesurement","SelectingHost_mesurement","end_mesurement","prospector","offset","tags","beat","source"]
    }   
}

output {
  influxdb {
    db => "$dbname"
    host => "$influxdb-host"
    port => "8086"
    user => "$username"
    password => "$yourpasswd"
    measurement => "kubeSchedulerTimeCost"
    coerce_values => {
      "total_time" => "float"
      "computing_time" => "float"
      "prioritizing_time" => "float"
      "SelectingHost_time" => "float"
      "end_time" => "float"
    }
    data_points => {
      "namespace" => "%{namespace}"
      "podname" => "%{podname}"
      "ComputingPredicates" => "%{ComputingPredicates}"
      "Prioritizing" => "%{Prioritizing}"
      "SelectingHost" => "%{SelectingHost}"
      "End" => "%{END}"
      "zone" => "%{zone}"
      "total_time" => "%{total_time}"
      "computing_time" => "%{computing_time}"
      "prioritizing_time" => "%{prioritizing_time}"
      "SelectingHost_time" => "%{SelectingHost_time}"
      "end_time" => "%{end_time}"
      "host" => "%{host}"
     }   
    send_as_tags => ["host","zone","namespace","podname","ComputingPredicates","Prioritizing","SelectingHost","End"]
  }
#  elasticsearch {
#    hosts => "$es-host:9200"
#    manage_template => false
#    index => "%{[@metadata][beat]}-%{[@metadata][version]}-%{+YYYY.MM.dd}"
#    document_type => "%{[@metadata][type]}"
#  }
  stdout { codec => rubydebug }
}

note:

• 可通过执行bin/logstash -f ./config/conf.d/scheduler.conf -- config.reload.automatic
• 若启动logstash服务，但是input beats 5044 port没有打开，则可能时权限问题，更改一下queue的权限chown -R logstash:logstash /var/lib/logstash

influxdb

influxdb需要将相应的端口打开

grafana看板

可看到各个集群的pod在各个阶段的调度耗时及调度时间

scheduler.png