Save 37% off PRO during our Black Friday Sale! »

Elasticsearch Deepdive

Elasticsearch Deepdive

Elastic{ON} Tour London - June 22, 2017

Dive in with Clinton to learn how to use the latest features in Elasticsearch.

Clint Gormley | Elasticsearch Team Lead | Elastic

Dd9d954997353b37b4c2684f478192d3?s=128

Elastic Co

June 22, 2017
Tweet

Transcript

  1. Clinton Gormley Team Lead, Elasticsearch Elasticsearch past, present, future

  2. 2 Elasticsearch 5.0 26 October 2016

  3. 3 Better at Numbers Safe Simple Things
 Should Be Simple

    Elasticsearch 5.0
  4. 4 Great for Metrics • Faster to index • Faster

    to search • Smaller on disk • Less heap • IPv6
  5. 5 Keep Calm and Index On • Bootstrap checks •

    Fully sandboxed scripting (Painless) • Strict settings • Soft limits • All-new circuit breakers
  6. 6 ‘Time-series’ not ‘time consuming’ • Ingest node • Rollover

    API • Shrink API
  7. 7 Elasticsearch 5.x Feature rich

  8. 8 Elasticsearch 5.x Still ^ • Keyword normalization • Unified

    highlighter • Field collapse • Multi-word synonyms+proximity • Cancellable searches • Parallel scroll & reindex
  9. 9 Elasticsearch 5.x Still ^ • Numeric & date range

    fields • Automatic optimizations for range searches • Massive aggregations with partitioning • Faster geo-distance sorting • Faster geo-ip lookups and for logs and for numbers and for geo and ... ^
  10. 10 Where to next?

  11. 11 What are the pain points?

  12. © Marie-Lan Nguyen Wikimedia Commons / CC-BY 2.5 12 What

    are the pain points? • Ever increasing scale • Major version upgrades • Slow recovery • Sparse data and disk usage
  13. © Marie-Lan Nguyen Wikimedia Commons / CC-BY 2.5 13 What

    are the pain points? • Ever increasing scale • Major version upgrades • Slow recovery • Sparse data and disk usage
  14. 14 Ever increasing scale • More clusters, not bigger clusters

    • Easier to manage • Easier to upgrade • Reduce potential outages • Need to query across clusters
  15. 15 Tribe Node Yesterday’s solution

  16. 16 Cluster Sales Master Nodes Data Node Data Node Data

    Node Cluster R&D Master Nodes Data Node Data Node Data Node How the Tribe Node Works
  17. 17 Cluster Sales Master Nodes Data Node Data Node Data

    Node Tribe Node Cluster R&D Master Nodes Data Node Data Node Data Node tribe: t1: cluster.name: sales t2: cluster.name: r_and_d How the Tribe Node Works
  18. 18 Cluster Sales Master Nodes Data Node Data Node Data

    Node Tribe Node t1 Node Client Cluster R&D Master Nodes Data Node Data Node Data Node How the Tribe Node Works
  19. 19 Cluster Sales Master Nodes Data Node Data Node Data

    Node Tribe Node t1 Node Client Cluster R&D Master Nodes Data Node Data Node Data Node t2 Node Client How the Tribe Node Works
  20. Cluster Sales Master Nodes Data Node Data Node Data Node

    Tribe Node t1 Node Client Cluster R&D Master Nodes Data Node Data Node Data Node t2 Node Client 20 Cluster State Cluster State How the Tribe Node Works
  21. Cluster Sales Master Nodes Data Node Data Node Data Node

    Tribe Node t1 Node Client Cluster R&D Master Nodes Data Node Data Node Data Node t2 Node Client 21 Cluster State Cluster State How the Tribe Node Works
  22. Cluster Sales Master Nodes Data Node Data Node Data Node

    Tribe Node t1 Node Client Cluster R&D Master Nodes Data Node Data Node Data Node t2 Node Client 22 Merged Cluster State How the Tribe Node Works
  23. 23 Kibana Cluster Sales Master Nodes Data Node Data Node

    Data Node Tribe Node t1 Node Client Cluster R&D Master Nodes Data Node Data Node Data Node t2 Node Client Merged Cluster State How the Tribe Node Works
  24. Cluster Sales Master Nodes Data Node Data Node Data Node

    Tribe Node t1 Node Client Cluster R&D Master Nodes Data Node Data Node Data Node t2 Node Client 24 Problems With How the Tribe Node Works Merged Cluster State Kibana
  25. 25 Cluster Sales Master Nodes Data Node Data Node Data

    Node Tribe Node t1 Node Client Cluster R&D Master Nodes Data Node Data Node Data Node t2 Node Client Merged Cluster State Kibana Static Configuration tribe: t1: cluster.name: sales t2: cluster.name: r_and_d Problems With How the Tribe Node Works
  26. 26 Cluster Sales Master Nodes Data Node Data Node Data

    Node Tribe Node t1 Node Client Cluster R&D Master Nodes Data Node Data Node Data Node t2 Node Client Kibana Merged Cluster State Connections to All Nodes Problems With How the Tribe Node Works
  27. 27 Cluster Sales Master Nodes Data Node Data Node Data

    Node Tribe Node t1 Node Client Cluster R&D Master Nodes Data Node Data Node Data Node t2 Node Client Kibana Merged Cluster State Frequent cluster state updates Problems With How the Tribe Node Works
  28. 28 Cluster Sales Master Nodes Data Node Data Node Data

    Node Tribe Node t1 Node Client Cluster R&D Master Nodes Data Node Data Node Data Node t2 Node Client Kibana Merged Cluster State Index names must be unique Problems With How the Tribe Node Works
  29. 29 Cluster Sales Master Nodes Data Node Data Node Data

    Node t1 Node Client Cluster R&D Master Nodes Data Node Data Node Data Node t2 Node Client Merged Cluster State Tribe Node Kibana No master node No index creation Problems With How the Tribe Node Works
  30. 30 Cluster Sales Master Nodes Data Node Data Node Data

    Node Tribe Node t1 Node Client Cluster R&D Master Nodes Data Node Data Node Data Node t2 Node Client Merged Cluster State Kibana Reduce results from many shards Problems With How the Tribe Node Works
  31. 31 Tribe is going away

  32. 32 Welcome to Cross-Cluster Search

  33. 33 Cross-Cluster Search • Minimum viable solution to supersede tribe

    • Reduces the problem domain to query execution • Cluster information is reduced to a namespace
  34. 34 How Cross-Cluster search works Cluster Sales Master Nodes Data

    Node Data Node Data Node Cluster R&D Master Nodes Data Node Data Node Data Node
  35. 35 How Cross-Cluster search works Cluster Sales Master Nodes Data

    Node Data Node Data Node Cluster R&D Master Nodes Data Node Data Node Data Node Any node can perform cross-cluster search
  36. 36 How Cross-Cluster search works Cluster Sales Master Nodes Data

    Node Data Node Data Node Optional dedicated cross-cluster search cluster Master/Data Node Cluster R&D Master Nodes Data Node Data Node Data Node Master/Data Node
  37. 37 How Cross-Cluster search works Cluster Sales Master Nodes Data

    Node Data Node Data Node Master/Data Node Cluster R&D Master Nodes Data Node Data Node Data Node Master/Data Node PUT _cluster/settings { "transient": { "search.remote": { "sales.seeds": "10.0.0.1:9300", “r_and_d.seeds”: "10.1.0.1:9300" } } } Dynamic settings Optional dedicated cross-cluster search cluster
  38. 38 How Cross-Cluster search works Cluster Sales Master Nodes Data

    Node Data Node Data Node Master/Data Node Cluster R&D Master Nodes Data Node Data Node Data Node Master/Data Node No cluster state updates Optional dedicated cross-cluster search cluster
  39. 39 How Cross-Cluster search works Cluster Sales Master Nodes Data

    Node Data Node Data Node Master/Data Node Cluster R&D Master Nodes Data Node Data Node Data Node Master/Data Node Kibana Optional dedicated cross-cluster search cluster
  40. 40 How Cross-Cluster search works Cluster Sales Master Nodes Data

    Node Data Node Data Node Master/Data Node Cluster R&D Master Nodes Data Node Data Node Data Node Master/Data Node Kibana Can create indices Optional dedicated cross-cluster search cluster
  41. 41 How Cross-Cluster search works Cluster Sales Master Nodes Data

    Node Data Node Data Node Master/Data Node Cluster R&D Master Nodes Data Node Data Node Data Node Master/Data Node Kibana Optional dedicated cross-cluster search cluster
  42. 42 How Cross-Cluster search works Cluster Sales Master Nodes Data

    Node Data Node Data Node Master/Data Node Cluster R&D Master Nodes Data Node Data Node Data Node Master/Data Node Kibana Few lightweight connections Optional dedicated cross-cluster search cluster
  43. 43 How Cross-Cluster search works Cluster Sales Master Nodes Data

    Node Data Node Data Node Master/Data Node Cluster R&D Master Nodes Data Node Data Node Data Node Master/Data Node Kibana Index namespacing GET sales:*,r_and_d:logs*/_search { "query": { … } } Optional dedicated cross-cluster search cluster
  44. 44 How Cross-Cluster search works Cluster Sales Master Nodes Data

    Node Data Node Data Node Master/Data Node Cluster R&D Master Nodes Data Node Data Node Data Node Master/Data Node Kibana With many shards Batched Reduce Phase Optional dedicated cross-cluster search cluster
  45. © Marie-Lan Nguyen Wikimedia Commons / CC-BY 2.5 45 What

    are the pain points? • Ever increasing scale • Major version upgrades • Slow recovery • Sparse data and disk usage
  46. 46 Major version upgrades • Upgrade Lucene • Add new

    features • Streamline existing features • Breaking changes • Remove backwards compatibility cruft • Keep codebase maintainable © Famartin Wikimedia Commons / CC-BY 2.5
  47. 47 Major version upgrade pain • Too many changes at

    once • Full cluster restart • Upgrade Java client at same time as Elasticsearch cluster • Data from major_version - 2 no longer readable
  48. 48 Too many changes at once • Most features backported

    to 5.x • Deprecation logging • Migration assistance API (X-Pack)
  49. 49 Full Cluster Restart © Paul Cross / CC-BY 2.5

  50. 50 Rolling upgrades • Upgrade from 5.latest to 6.latest without

    full cluster restart • 5.latest is the latest GA release of 5.x when 6.0.0 goes GA • All 6.x releases will allow upgrading from that 5.x release, unless there is a new 5.x release
  51. 51 Rolling upgrade caveats • If using security, must have

    TLS enabled • Reserve the right to require full cluster restart in the future, but only if absolutely necessary • All nodes must be upgraded to 5.latest before upgrading • Indices created in 2.x still need to be reindexed before upgrading to 6.x
  52. 52 Java client • All other languages use REST interface

    • Transport client tied to Elasticsearch major version • Second entry point into the cluster • Complicates distinguishing between clients and nodes
  53. 53 Java REST client • Released in 5.0 • JSON

    strings only • Resilient, but not user friendly
  54. 54 Java high level REST client • Works across major

    version upgrade • IDE friendly • Similar API to Transport Client - easy migration • Based on low-level REST client • Supports CRUD & Search • Currently targeted for release in 5.6 • Depends on elasticsearch-core
  55. 55 Data compatibility • Any index created in 5.x can

    be upgraded to 6.x • Any index created in 2.x must be reindexed in 5.x or imported with reindex-from-remote • How do you reindex a petabyte of data?
  56. 56 Cross Major Version Search v5.2.0 Kibana Master Nodes Data

    Node Data Node
  57. 57 Cross Major Version Search v5.2.0 Kibana v6.0.0 Master Nodes

    Data Node Data Node Master Nodes Data Node Data Node
  58. 58 Cross Major Version Search v5.2.0 Master Nodes Data Node

    Data Node v6.0.0 v5.latest Kibana Master Nodes Data Node Data Node
  59. 59 Cross Major Version Search v5.2.0 Master Nodes Data Node

    Data Node v6.0.0 Kibana Master Nodes Data Node Cross Cluster Client v5.latest
  60. © Marie-Lan Nguyen Wikimedia Commons / CC-BY 2.5 60 What

    are the pain points? • Ever increasing scale • Major version upgrades • Slow recovery • Sparse data and disk usage
  61. How is data stored? In memory buffer Transaction log Lucene

    segments
  62. How is data stored? In memory buffer Transaction log Lucene

    segments 1 1
  63. How is data stored? In memory buffer Transaction log Lucene

    segments 1 2 1 2
  64. How is data stored? In memory buffer Transaction log Lucene

    segments 1 2 3 1 2 3
  65. How is data stored? In memory buffer Transaction log Lucene

    segments 1 2 3 1 2 3 REFRESH
  66. How is data stored? In memory buffer Transaction log Lucene

    segments 1 2 3 1 2 3
  67. How is data stored? In memory buffer Transaction log Lucene

    segments 4 5 6 7 1 2 3 4 5 6 7 1 2 3
  68. How is data stored? In memory buffer Transaction log Lucene

    segments 4 5 6 7 1 2 3 4 5 6 7 1 2 3 REFRESH
  69. How is data stored? In memory buffer Transaction log Lucene

    segments 1 2 3 4 5 6 7 1 2 3 4 5 6 7
  70. How is data stored? In memory buffer Transaction log Lucene

    segments 8 9 1 2 3 4 5 6 7 8 9 1 2 3 4 5 6 7
  71. How is data stored? In memory buffer Transaction log Lucene

    segments 8 9 1 2 3 4 5 6 7 8 9 1 2 3 4 5 6 7 FLUSH
  72. How is data stored? In memory buffer Transaction log Lucene

    segments 1 2 3 4 5 6 7 8 9 1 2 3 4 5 6 7 8 9
  73. How is data stored? In memory buffer Transaction log Lucene

    segments 1 2 3 4 5 6 7 8 9 1 2 3 4 5 6 7 8 9
  74. How is data stored? In memory buffer Transaction log Lucene

    segments 1 2 3 4 5 6 7 8 9
  75. How is data stored? In memory buffer Transaction log Lucene

    segments 1 2 3 4 5 6 7 8 9 MERGE
  76. How is data stored? In memory buffer Transaction log Lucene

    segments 1 2 3 4 5 6 7 8 9 1 2 3 8 9
  77. How is data stored? In memory buffer Transaction log Lucene

    segments 4 5 6 7 1 2 3 8 9 1 2 3 8 9
  78. Data replication Client Primary shard Replica shard

  79. Data replication Client Primary shard Replica shard

  80. Data replication Client Primary shard Replica shard 1 2

  81. Data replication Client Primary shard Replica shard 1 2

  82. Data replication Client Primary shard Replica shard 1 2

  83. Data replication Client Primary shard Replica shard 1 2

  84. Data replication Client Primary shard Replica shard 1

  85. Data replication Lucene segments 4 5 6 7 1 2

    3 8 9 Primary Lucene segments 1 2 4 7 9 3 5 6 8 Replica
  86. Replica recovery Lucene segments 4 5 6 7 1 2

    3 8 9 Primary Lucene segments Replica 1 2 4 7 9 3 5 6 8
  87. Replica recovery Lucene segments 4 5 6 7 1 2

    3 8 9 Primary 4 5 6 7 1 2 3 8 9 Lucene segments Replica 1 2 4 7 9 3 5 6 8
  88. Replica recovery Lucene segments 4 5 6 7 1 2

    3 8 9 Primary 4 5 6 7 1 2 3 8 9 Lucene segments Replica
  89. Data at rest Lucene segments 4 5 6 7 1

    2 3 8 9 Primary Lucene segments Replica 1 2 4 7 9 3 5 6 8 SYNCED FLUSH
  90. Data at rest Lucene segments 4 5 6 7 1

    2 3 8 9 Primary Lucene segments Replica 1 2 4 7 9 3 5 6 8
  91. Data at rest Lucene segments 4 5 6 7 1

    2 3 8 9 Primary Lucene segments Replica 1 2 4 7 9 3 5 6 8
  92. Data at rest Lucene segments 4 5 6 7 1

    2 3 8 9 Primary Lucene segments Replica 1 2 4 7 9 3 5 6 8
  93. Active indexing Lucene segments 4 5 6 7 1 2

    3 8 9 Primary Lucene segments Replica 1 2 4 7 9 3 5 6 8
  94. Active indexing Lucene segments 4 5 6 7 1 2

    3 8 9 Primary Lucene segments Replica 1 2 4 7 9 3 5 6 8
  95. Active indexing Lucene segments 4 5 6 7 1 2

    3 8 9 Primary Lucene segments Replica 1 2 4 7 9 3 5 6 8 10 11
  96. Active indexing Lucene segments 1 2 3 4 5 6

    7 8 9 Primary Lucene segments Replica 1 2 4 7 9 3 5 6 8 10 11
  97. Active indexing Lucene segments 1 2 3 4 5 6

    7 8 9 Primary Lucene segments Replica 1 2 4 7 9 3 5 6 8 10 11 1 2 3 4 5 6 7 8 9 10 11
  98. Active indexing Lucene segments 1 2 3 4 5 6

    7 8 9 Primary Lucene segments Replica 10 11 1 2 3 4 5 6 7 8 9 10 11
  99. Sequence numbers Primary Replica

  100. Sequence numbers Transaction log 1 2 3 Primary Transaction log

    Replica 1 2 3
  101. Sequence numbers Transaction log 1 2 3 Primary Transaction log

    Replica 1 2 3
  102. 1 2 3 Sequence numbers Transaction log 1 2 3

    4 5 Primary Transaction log Replica
  103. Sequence numbers Transaction log 1 2 3 4 5 Primary

    Transaction log Replica 1 2 3
  104. Sequence numbers Transaction log 1 2 3 4 5 6

    7 8 9 Primary Transaction log Replica 1 2 3 4 5 7 8
  105. Trimming the transaction log Transaction log 1 2 3 4

    5 6 7 8 9 Primary Transaction log Replica 1 2 3 4 5 7 8
  106. Trimming the transaction log Transaction log Primary Transaction log Replica

    5 6 7 8 9 5 7 8
  107. 107 Slow recovery • 6.0: ‒ Fast replica recovery ‒

    Configurable transaction log retention period • Lays groundwork for: ‒ Replica syncing after primary failure ‒ Cross-data-centre recovery
  108. © Marie-Lan Nguyen Wikimedia Commons / CC-BY 2.5 108 What

    are the pain points? • Ever increasing scale • Major version upgrades • Slow recovery • Sparse data and disk usage
  109. 109 Sparse data and disk usage • Doc Values: Columnar

    store • Fast access to a field’s value for many documents • Used for aggregations, sorting, scripting, and some queries • Written to disk at index time • Cached in the file-system cache © Tony Weman / CC-BY 2.5
  110. 110 Doc values - Dense data Segment 2 Docs Field

    1 Field 2 1 Four D Segment 1 Docs Field 1 Field 2 1 One A 2 Two B 3 Three C
  111. 111 Doc values - Dense data Merged Segment 3 Docs

    Field 1 Field 2 1 One A 2 Two B 3 Three C 4 Four D Segment 1 Docs Field 1 Field 2 1 One A 2 Two B 3 Three C Segment 2 Docs Field 1 Field 2 1 Four D
  112. 112 Doc values - Sparse data Segment 1 Docs Field

    1 Field 2 1 One A 2 Two B 3 Three C Segment 2 Docs Field 3 Field 4 Field 5 1 Foo Null Null 2 Null Bar Null 3 Null Null Baz
  113. 113 Doc values - Sparse data Segment 1 Docs Field

    1 Field 2 1 One A 2 Two B 3 Three C Segment 2 Docs Field 3 Field 4 Field 5 1 Foo Null Null 2 Null Bar Null 3 Null Null Baz Merged Segment 3 Docs Field 1 Field 2 Field 3 Field 4 Field 5 1 One A Null Null Null 2 Two B Null Null Null 3 Three C Null Null Null 4 Null Null Foo Null Null 5 Null Null Null Bar Null 6 Null Null Null Null Baz
  114. 114 Doc values - Sparse data Segment 1 Docs Field

    1 Field 2 1 One A 2 Two B 3 Three C Segment 2 Docs Field 3 Field 4 Field 5 1 Foo 2 Bar 3 Baz Merged Segment 3 Docs Field 1 Field 2 Field 3 Field 4 Field 5 1 One A 2 Two B 3 Three C 4 Foo 5 Bar 6 Baz
  115. 115 Sparse doc value support • Coming in 6.0 •

    Big disk savings for sparse values - pay for what you use • Big file cache savings - 
 more data can be cached • Dense queries still more efficient 
 than sparse © Tony Weman / CC-BY 2.5
  116. Elasticsearch 6.0

  117. Coming soon to a cluster near you Elasticsearch 6.0