Monitoring Reactive Architecture Like Never Before / 今までになかったリアクティブアーキテクチャの監視 by Sahil Sawhney
1. A Reactive Platform to monitor
Reactive Application
Sahil Sawhney
Lead DevOps Consultant
+91-9871211045
www.knoldus.com
今までになかったリアクティブアーキテクチャの監視
2. Agenda
What Is Reactive Monitoring
Reactive Monitoring Challenges
PremonR: The Solution
Demonstration
リアクティブな監視の課題から、PremonR というソリューションに
ついて、デモを交えて紹介します。
Monitoring Needs Of The Hour
3. Monitoring Needs Of The Hour
顧客にとって本当に重要なことを計測するとはどういうことだろう
どんなに注意深く、誠実であっても、問題は起こる。監視されてなければ、リリースさ
れたとは言えない。
“Traditional metrics of cpu and memory usage don’t matter to your customers. How ’bout measuring what really matters to your
customers?”
“No matter how careful or good you are, sh!t will happen.”
“If it isn’t monitored, it isn’t production!”
“DevOps simply adds the idea that small, cross-functional teams should own the entire delivery process from concept through user
feedback and production monitoring.”
“Application up, monitoring applied, alerting all set. Now relax until things go down”
4. What is Reactive Monitoring?
❑ Applications whose foundation is laid on Reactive Manifesto accounts for being Reactive Applications.
❑ But can any monitoring pipeline ensure that its worthy enough to monitor your reactive fleet?
リアクティブな監視とは?
監視パイプラインによってリアクティブなサービス群を監視することが十分に価値ある
ことは確かめられるだろうか?
6. Challenges
Moving and
decentralized
component
Lack of visibility
across the
enterprise
Lack of democracy in
monitoring, due to cost of
acquiring commercial tools
Insufficient,
unmanaged
alerting rules
Alerting of the
mishaps after they
have occured.
Lack of persistence of
custom visualizations
& alert rules
Dedicated environment (prod,
beta) based classification of viz
+ dashboard
Keeping up with
evolving
applications & tools
様々な難しさ。流動的な、分散コンポーネント。不十分で管理されて
いないアラートルール。問題の発生後に発報するアラートなど。
8. In comes PreMonR
With years of experience in Reactive stack; Knoldus compiles all its learning into a Premonition based
Reactive Monitoring and Alerting Platform.
The fabrication of such a tool was based on three driving forces:
Driving
Forces
Monitor the
reactive applications
The monitoring platform
must itself be Reactive
Containing the mishaps
before they turn into reality
リアクティブ・アプリケーションの監視 / 監視プラットフォーム自体
がリアクティブ / 問題が実際に問題となる前に抑制する
9. Features of PreMonR
Based on a
reactive
monitoring
pipeline.
Highly available
monitoring
platform.
One Subscription
all solution
Centralized
insights of your
distributed
platform
Prebuilt fleet of dashboards
& alerting rules as per the
project stack.
Customize as
per your
appetite
Specialized for distributed
environments like
Kubernetes and DCOS
Real time
monitoring and
premonition
based alerting
リアクティブなモニタリングパイプライン上に構築。
リアルタイムな監視と、予測ベースのアラート。高可用性。k8s などの分散環境に特化
。
11. PreMonR Architecture
1. Extractor extracts metrics from the
underlying infrastructure.
2. Collector collects logs of the
application as well as infrastructure.
3. Shipper exports the extracted metrics
to transformer in case there are
some transformations that must be
applied to collected data.
4. Transformer transforms the input
logs and metrics as per the use case.
5. Data Backend stores the metrics and
logs aggregated by Extractore and
Collectors
6. Premonition engine apply Machine
Learning algorithm to detect
anomalies and facilitate proactive
alerting.
7. Visualizer is the UI where all logs and
Dashboards could be visualized.
8. Alerter fires alerts in case of threshold
breaches.
Metrics
Extractor
Data
Backend
Alerter
Visualizer
Logs
Collector
PagerDuty
Email
Slack
System logs(syslog, journald)
Application logs(log4j, log4net)
Server Logs(Apache, Nginx)
Platform Logs(AWS, Baremetal)
Cluster Logs(K8S, Mesos)
System metrics
(CPU, memory, disk)
Infrastructure metrics
(AWS, Gcloud,
Baremetal)
Application Agents
(APM, error tracking)
TransformerShipper
PreMonition
Analytics
Engine
インフラからメトリクスを収集、アプリケーションからのログ収集、メトリクスの変換
、ストア、機械学習による予測や異常検知、ダッシュボード、アラート
12. ● Lagom Metrics
● Spring-Boot Metrics
● Akka Metrics
● Play Metrics
● Application Logs
● System Resource Metrics
● Cassandra metrics
● Dgraph metrics
● Elasticsearch Metrics
● Kafka Metrics
● Anything That Gives Metrics
Monitoring
Application
Monitoring
Infrastructure
Monitoring
What Can I Monitor ?
モニタリングできるもの
13. The PremonR Effect
Centralized monitoring and alerting tool for
cluster health, log analysis.
High availability, persistence of Dashboards
and alert rules.
Clean and convenient setup procedure
Automatically discover, configure and
customize all relevant metrics.
Easily Adaptable/customizable solution
based on BELK
Visualize the health and topology of their
distributed applications in real-time
Data science-driven anomaly detection quickly detects
the tentative problems.
Optimized thresholds ensuring real time
alerting.
適切なメトリクスを自動で発見、設定、カスタマイズ。分散アプリケーションの健全性
、トポロジーをリアルタイムに可視化
機械学習ドリブンな異常検知。BELK 上に構築されている。