Black friday logs - Scaling Elasticsearch
5 likes1,429 views
The document presents a comprehensive discussion on managing Black Friday logs using the Elastic Stack 6.0, emphasizing various platform architectures and Elasticsearch cluster sizing. It highlights strategies for optimizing data handling, including shard sizing, the use of indices and aliases, and the implementation of automation via rollover and shrink APIs. Key considerations are provided for ensuring optimal performance in big data environments, including bulk sizes, load distribution, and enhancing write throughput.
![66
Test it ;)
DATE=`date +%Y.%m.%d`
LOG=logs/logs.txt
exec_test () {
curl -s -XDELETE "http://USER:PASS@HOST:9200/logstash-$DATE"
sleep 10
export SIZE=$1
time cat $LOG | ./bin/logstash -f logstash.conf
}
for SIZE in 100 500 1000 3000 5000 10000; do
for i in {1..20}; do
exec_test $SIZE
done; done;
input { stdin{} }
filter {}
output {
elasticsearch {
hosts => ["10.12.145.189"]
flush_size => "${SIZE}"
} }
In Beats set "bulk_max_size"
in the output.elasticsearch](https://image.slidesharecdn.com/2017-11capitoledulibre-managingblackfridaylog-180212105215/85/Black-friday-logs-Scaling-Elasticsearch-63-320.jpg)
![70
Clients
• Most clients implement round robin
‒ you specify a seed list
‒ the client sniffs the cluster
‒ the client implement different selectors
• Logstash allows an array (no sniffing)
• Beats allows an array (no sniffing)
• Kibana only connects to one single node
output {
elasticsearch {
hosts => ["node1","node2","node3"]
} }](https://image.slidesharecdn.com/2017-11capitoledulibre-managingblackfridaylog-180212105215/85/Black-friday-logs-Scaling-Elasticsearch-68-320.jpg)
![73
Test it ;)
#docs
time(s)
100 500 1000
NO Round Robin 191.7 161.9 163.5
Round Robin 189.7 159.7 159.0
• 2 node cluster (m3.large)
‒ 2 vCPU, 7.5GB Memory, 1x32GB SSD
• 1 index server (m3.large)
‒ logstash (round robin configured)
‒ hosts => ["10.12.145.189", "10.121.140.167"]
‒ kibana](https://image.slidesharecdn.com/2017-11capitoledulibre-managingblackfridaylog-180212105215/85/Black-friday-logs-Scaling-Elasticsearch-71-320.jpg)
Black friday logs - Scaling Elasticsearch
- 1. Sylvain Wallez @bluxte Software Engineer, Elastic Cloud Managing your Black Friday Logs
- 3. Elastic Stack 6.0 in orbit!
- 4. 4 Agenda Data Platform Architectures Elasticsearch Cluster Sizing Optimal Bulk Size Distribute the Load 1 2 3 4 5 Optimizing Disk IO 6 Final Remarks
- 6. life:universe user:soulmate _Search? outside the box city:restaurant car:model fridge:leftovers work:dreamjob
- 8. Logging
- 9. Metrics
- 12. APM
- 13. APM
- 14. 17 Hosted Elasticsearch & Kibana Includes X-Pack features Starts at $45/mo Available in Amazon Web Service Google Cloud Platform
- 15. 18 Provision and manage multiple Elastic Stack environments and provide search-aaS, logging-aaS, BI-aaS, data-aaS to your entire organization
- 16. 26 The Elastic Journey of Data Beats Log Files Metrics Wire Data your{beat} Data Store Web APIs Social Sensors Elasticsearch Master Nodes (3) Ingest Nodes (X) Data Nodes Hot (X) Data Notes Warm (X) Logstash Nodes (X) Kafka Redis Messaging Queue Kibana Instances (X) NotificationQueues Storage Metrics X-Pack X-PackX-Pack
- 18. 28 Terminology Cluster my_cluster Server 1 Node A d1 d2 d3 d4 d5 d6 d7 d8d9 d10 d11 d12 Index twitter d6d3 d2 d5 d1 d4 Index logs
- 19. 29 Partition Cluster my_cluster Server 1 Node A d1 d2 d3 d4 d5 d6 d7 d8d9 d10 d11 d12 Index twitter d6d3 d2 d5 d1 d4 Index logs Shards 0 1 4 2 3 0 1
- 20. 30 Distribution Cluster my_cluster Server 1 Node A Server 2 Node Btwitter shard P4 d1 d2 d6 d5 d10 d12 twitter shard P2 twitter shard P1 logs shard P0 d2 d5 d4 logs shard P1 d3 d4 d9 d7 d8 d11 twitter shard P3 twitter shard P0 d6d3 d1
- 21. 31 Replication Cluster my_cluster Server 1 Node A Server 2 Node Btwitter shard P4 d1 d2 d6 d5 d10 d12 twitter shard P2 twitter shard P1 logs shard P0 d2 d5 d4 logs shard P1 d3 d4 d9 d7 d8 d11 twitter shard P3 twitter shard P0 twitter shard R4 d1 d2 d6 d12 twitter shard R2 d5 d10 twitter shard R1 d6d3 d1 d6d3 d1 logs shard R0 d2 d5 d4 logs shard R1 d3 d4 d9 d7 d8 d11 twitter shard R3 twitter shard R0 • Primaries • Replicas
- 22. 32 Scaling Data
- 23. 33 Scaling Data
- 24. 34 Scaling Data
- 26. 36 Scaling • In Elasticsearch, shards are the working unit • More data → More shards Big Data ... ... But how many shards?
- 27. 37 How much data? • ~1000 events per second • 60s * 60m * 24h * 1000 events ~87M events per day • 1kB per event ~82GB per day • 3 months ~7TB
- 28. 38 Shard Size • It depends on many different factors ‒ document size, mapping, use case, kinds of queries being executed, desired response time, peak indexing rate, budget, ... • After the shard sizing*, each shard should handle 45GB • Up to 10 shards per machine * https://www.elastic.co/elasticon/conf/2016/sf/quantitative-cluster-sizing
- 29. 39 How many shards? • Data size: ~7TB • Shard Size: ~45GB* • Total Shards: ~160 • Shards per machine: 10* • Total Servers: 16 Cluster my_cluster 3 months of logs ... * https://www.elastic.co/elasticon/conf/2016/sf/quantitative-cluster-sizing
- 30. 40 But... • How many indices? • What do you do if the daily data grows? • What do you do if you want to delete old data?
- 31. 41 Time-Based Data • Logs, social media streams, time-based events • Timestamp + Data • Do not change • Typically search for recent events • Older documents become less important • Hard to predict the data size
- 32. 42 Time-Based Data • Time-based Indices is the best option ‒ create a new index each day, week, month, year, ... ‒ search the indices you need in the same request
- 36. 46 Templates • Every new created index starting with 'logs-' will have ‒ 2 shards ‒ 1 replica (for each primary shard) ‒ 60 seconds refresh interval PUT _template/logs { "template": "logs-*", "settings": { "number_of_shards": 2, "number_of_replicas": 1, "refresh_interval": "60s" } } More on that later
- 37. 47 Alias Cluster my_cluster d6d3 d2 d5 d1 d4 logs-2017-10-06 users Application logs-write logs-read
- 42. 51 Rollover config example Create a new index every (7 days or 10k docs) PUT /logs-000001 { "aliases": { "logs_write": {} } } # Add > 10000 documents to logs-000001 POST /logs_write/_rollover { "conditions": { "max_age": "7d", "max_docs": 10000 } } # logs_write now points to logs-000002
- 43. 51 Rollover config example Create a new index every (7 days or 10k docs) PUT /logs-000001 { "aliases": { "logs_write": {} } } # Add > 10000 documents to logs-000001 POST /logs_write/_rollover { "conditions": { "max_age": "7d", "max_docs": 10000 } } # logs_write now points to logs-000002 # Use "date math" in the index name PUT /<logs-{now/d}-0001> { "aliases": { "logs_write": {} } } # log_write --> logs-2017-11-17-0001
- 44. 52 Do not Overshard • 3 different logs • 1 index per day each • 1GB each / day • 5 shards (default): so 200mb / shard vs 45gb • 6 months retention • ~900 shards for ~180GB • we needed ~4 shards! don't keep default values! Cluster my_cluster access-... d6d3 d2 d5 d1 d4 application-... d6d5 d9 d5 d1 d7 mysql-... d10d59 d3 d5 d0 d4
- 51. 55 Shrink config example PUT /my_source_index/_settings { "settings": { "index.routing.allocation.require._name": "shrink_node_name", "index.blocks.write": true } } POST my_source_index/_shrink/my_target_index { "settings": { "index.number_of_replicas": 1, "index.number_of_shards": 1, "index.codec": "best_compression" } }
- 52. 56 Scaling Big Data ... ...1M users
- 53. 56 Scaling Big Data ... ...1M users But what happens if we have 2M users?
- 54. 57 Scaling Big Data ... ...1M users ... ...1M users
- 55. 58 Scaling Big Data ... ...1M users ... ...1M users ... ...1M users
- 56. 59 Scaling Big Data ... ... ... ... ... ... U s e r s
- 57. 60 Shards are the working unit • Primaries ‒ More data → More shards ‒ write throughput (More writes → More primary shards) • Replicas ‒ high availability (1 replica is the default) ‒ read throughput (More reads → More replicas)
- 59. 62 What is Bulk? Elasticsearch Master Nodes (3) Ingest Nodes (X) Data Nodes Hot (X) Data Notes Warm (X) __________ __________ __________ __________ __________ __________ __________ __________ __________ __________ __________ __________ __________ __________ __________ __________ __________ __________ __________ __________ __________ __________ __________ __________ __________ __________ __________ __________ __________ __________ __________ __________ __________ __________ __________ __________ __________ __________ __________ __________ _____ 1000 log events Beats Logstash Application 1000 index requests with 1 document 1 bulk request with 1000 documents
- 60. 63 What is the optimal bulk size? Elasticsearch Master Nodes (3) Ingest Nodes (X) Data Nodes Hot (X) Data Notes Warm (X) __________ __________ __________ __________ __________ __________ __________ __________ __________ __________ __________ __________ __________ __________ __________ __________ __________ __________ __________ __________ __________ __________ __________ __________ __________ __________ __________ __________ __________ __________ __________ __________ __________ __________ __________ __________ __________ __________ __________ __________ _____ 1000 log events Beats Logstash Application 4 * 250? 1 * 1000? 2 * 500?
- 61. 64 It depends... • on your application (language, libraries, ...) • document size (100b, 1kb, 100kb, 1mb, ...) • number of nodes • node size • number of shards • shards distribution
- 62. 65 Test it ;) Elasticsearch Master Nodes (3) Ingest Nodes (X) Data Nodes Hot (X) Data Notes Warm (X) __________ __________ __________ __________ __________ __________ __________ __________ __________ __________ __________ __________ __________ __________ __________ __________ __________ __________ __________ __________ __________ __________ __________ __________ __________ __________ __________ __________ __________ __________ __________ __________ __________ __________ __________ __________ __________ __________ __________ __________ _____ 1000000 log events Beats Logstash Application 4000 * 250-> 160s 1000 * 1000-> 155s 2000 * 500-> 164s
- 63. 66 Test it ;) DATE=`date +%Y.%m.%d` LOG=logs/logs.txt exec_test () { curl -s -XDELETE "http://USER:PASS@HOST:9200/logstash-$DATE" sleep 10 export SIZE=$1 time cat $LOG | ./bin/logstash -f logstash.conf } for SIZE in 100 500 1000 3000 5000 10000; do for i in {1..20}; do exec_test $SIZE done; done; input { stdin{} } filter {} output { elasticsearch { hosts => ["10.12.145.189"] flush_size => "${SIZE}" } } In Beats set "bulk_max_size" in the output.elasticsearch
- 64. 67 Test it ;) • 2 node cluster (m3.large) ‒ 2 vCPU, 7.5GB Memory, 1x32GB SSD • 1 index server (m3.large) ‒ logstash ‒ kibana # docs 100 500 1000 3000 5000 10000 time(s) 191.7 161.9 163.5 160.7 160.7 161.5
- 68. 70 Clients • Most clients implement round robin ‒ you specify a seed list ‒ the client sniffs the cluster ‒ the client implement different selectors • Logstash allows an array (no sniffing) • Beats allows an array (no sniffing) • Kibana only connects to one single node output { elasticsearch { hosts => ["node1","node2","node3"] } }
- 71. 73 Test it ;) #docs time(s) 100 500 1000 NO Round Robin 191.7 161.9 163.5 Round Robin 189.7 159.7 159.0 • 2 node cluster (m3.large) ‒ 2 vCPU, 7.5GB Memory, 1x32GB SSD • 1 index server (m3.large) ‒ logstash (round robin configured) ‒ hosts => ["10.12.145.189", "10.121.140.167"] ‒ kibana
- 73. 75 Durability index a doc time lucene flush buffer index a doc buffer index a doc buffer buffer segment
- 74. 76 Durability index a doc time lucene flush buffer segment trans_log buffer trans_log buffer trans_log elasticsearch flush doc op lucene commit segment segment
- 75. 77 refresh_interval • Dynamic per-index setting • Increase to get better write throughput to an index • New documents will take more time to be available for Search. PUT logstash-2017.05.16/_settings { "refresh_interval": "60s" } #docs time(s) 100 500 1000 1s refresh 189.7 159.7 159.0 60s refresh 185.8 152.1 152.6
- 76. 78 Translog fsync every 5s (1.7) index a doc buffer trans_log doc op index a doc buffer trans_log doc op Primary Replica redundancy doesn’t help if all nodes lose power
- 77. 79 Translog fsync on every request • For low volume indexing, fsync matters less • For high volume indexing, we can amortize the costs and fsync on every bulk • Concurrent requests can share an fsync bulk 1 bulk 2 single fsync
- 78. 80 Async Transaction Log • index.translog.durability ‒ request (default) ‒ async • index.translog.sync_interval (only if async is set) • Dynamic per-index settings • Be careful, you are relaxing the safety guarantees #docs time(s) 100 500 1000 Request fsync 185.8 152.1 152.6 5s sync 154.8 143.2 143.1
- 79. Wrapping up
- 80. 82 Wrapping up Beats Log Files Metrics Wire Data your{beat} Data Store Web APIs Social Sensors Elasticsearch Master Nodes (3) Ingest Nodes (X) Data Nodes Hot (X) Data Notes Warm (X) Logstash Nodes (X) Kafka Redis Messaging Queue Kibana Instances (X) NotificationQueues Storage Metrics X-Pack X-PackX-Pack
- 81. 83 Wrapping up • Primaries ‒ More data → More shards ‒ Do not overshard! • Replicas ‒ high availability (1 replica is the default) ‒ read throughput (More reads → More replicas) Big Data ... ... ... ... ... ... U s e r s
- 82. 84 • Bulk and Test • Distribute the Load • Refresh Interval • Async Trans Log (careful) #docs 100 500 1000 Default 191.7s 161.9s 163.5s RR+60s+Async5s 154.8s 143.2s 143.1s Wrapping up
- 83. 89 Sylvain Wallez @bluxte Software Engineer, Elastic Cloud Thanks! Questions?