Professional Documents
Culture Documents
Elasticsearch Best
Practice Architecture
1
whoami
Eric Westberg
Solutions Architect
Stockholm, Sweden
Joined Elastic last year
2
Webinar Housekeeping & Logistics
3
Application Site Enterprise
Search Search Search Logging
Future Solutions
Business Security
Metrics APM
Analytics Analytics
5
Elasticsearch Node Types
Nodes can play one or more roles, for workload isolation and scaling
• Master Nodes
– Control the cluster, requires a minimum of 3, one is active at any given time
Elasticsearch
• Data Nodes
– Hold indexed data and perform data related operations
Master (3) – Differentiated Hot and Warm Data nodes can be used
Ingest (X)
• Ingest Nodes
– Use ingest pipelines to transform and enrich before indexing
Coordinating (X)
• Coordinating Nodes
Data – Hot (X)
– Route requests, handle search reduce phase, distribute bulk indexing
– All nodes function as coordinating nodes
Data – Warm (X)
• Alerting Nodes
– Run alerting jobs
Alerting (X)
• Machine Learning Nodes
Machine Learning (2+) – Run machine learning jobs
6 All product names, logos, and brands are property of their respective owners and are used only for identification purposes. This is not an endorsement.
Split Brain
Network disconnection
7
Avoiding Split Brain
Network partition
8
Inside a Large Elasticsearch Logging Cluster
Reduce infrastructure costs, isolate workloads, and manage data lifecycle
Master Elasticsearch
Query
Alerting
Index
Response
Ingest Coordinating
Machine
Hot Data Warm Data
Learning
Curator
9 All product names, logos, and brands are property of their respective owners and are used only for identification purposes. This is not an endorsement.
Logging Architectures
10
Quick Start
Beats, Elasticsearch and Kibana
Beats
Elasticsearch
FILEBEAT WINLOGBEAT
Kibana
HEARTBEAT METRICBEAT
Uniform Nodes (3+)
PACKETBEAT AUDITBEAT
File Spool
Queue
11 All product names, logos, and brands are property of their respective owners and are used only for identification purposes. This is not an endorsement.
Advanced Processing and Resiliency
Adding Logstash processing, differentiated Elasticsearch node types
Beats
Logstash Elasticsearch
FILEBEAT WINLOGBEAT
Master Nodes (3)
Kibana
HEARTBEAT METRICBEAT Workers (1+)
Data Nodes (2+)
Persistent
PACKETBEAT AUDITBEAT Queues ML Nodes (2+)
File Spool
Queue
12
Flexible ingestion and input sources
Beats
Elasticsearch
13 All product names, logos, and brands are property of their respective owners and are used only for identification purposes. This is not an endorsement.
Application Metric Collection with Elastic APM
Datastore JMX
APM Curated
App
apm-agents
apm-server
NodeJS Python
14
Deployment Best
Practices
15
Centralized Monitoring Cluster
Maintain isolated monitoring cluster for monitoring workload isolation
Elasticsearch Elasticsearch
Monitoring
Data
Beats
16 All product names, logos, and brands are property of their respective owners and are used only for identification purposes. This is not an endorsement.
Cloud Monitoring Cluster
Opt-in Elastic Cloud cluster for monitoring on-premise stack
Kibana Monitoring
Data
Elastic Cloud
Logstash Monitoring
Cluster
Elasticsearch
Beats
On-Prem Cluster
17 All product names, logos, and brands are property of their respective owners and are used only for identification purposes. This is not an endorsement.
Isolated Audit Logging Cluster
Maintain isolated audit logging cluster for increased security and compliance
Elasticsearch
Cluster 1
Elasticsearch Elasticsearch
Audit Logs Audit Log
Elasticsearch
Cluster Kibana
Cluster 2
Audit and
Compliance
Elasticsearch
Cluster 3
18 All product names, logos, and brands are property of their respective owners and are used only for identification purposes. This is not an endorsement.
Multi Data Center
19
Multiple Data Centers, Duplicate Data
Kafka
Redis
Logstash Elasticsearch Kibana
Messaging
Queue
Replication
ACTIVE
Load
Logstash Elasticsearch Kibana Balancer
Kafka
ACTIVE
Redis
Messaging
Queue
20 All product names, logos, and brands are property of their respective owners and are used only for identification purposes. This is not an endorsement.
Multi Data Centers with a Queue at Each DC
DC1
Beats
Kafka
Logstash Messaging
Queue
Datastore Web APIs
Logstash Elasticsearch
DC2
Beats
Kafka
Logstash Messaging
Queue
Datastore Web APIs
21
Multi Data Center, Distinct Data and Cross-Cluster Search
Elasticsearch
Cross-Cluster
Search Cluster Kibana
22 All product names, logos, and brands are property of their respective owners and are used only for identification purposes. This is not an endorsement.
Scaling Kibana
23
High Availability
Pair two coordinating nodes with two independent Kibana nodes
Host A
Elasticsearch Kibana
Coordinating Node
Host B
Load
Elasticsearch Kibana Balancer
Coordinating Node
24
Separating Dashboards by Groups
Isolate user content by group in different Kibana instances
Group A
Kibana
Elasticsearch
Kibana Group B
Kibana
Group C
25
Questions?
Thank You
● Web : www.elastic.co
● Products : https://www.elastic.co/products
● Forums : https://discuss.elastic.co/
● Community : https://www.elastic.co/community/meetups
● Twitter : @elastic
● Contact us : www.elastic.co/contact