Elasticsearch Lessons Learned

Presentation held at the code.talks conference in Hamburg on October 9, 2014. Apart from the slides, the presentation featured various examples demonstrated using the Elasticsearch Marvel/Sense plugin. The slides contain a link to a gist showing those examples.

Transcript

1. codecentric AG
   Patrick Peschlow
   Elasticsearch Lessons Learned
   

   View Slide

2. codecentric AG
   Beginner’s lesson
   Quick introduction
   

   View Slide

3. codecentric AG
   Lessons learned
   Map carefully
   

   View Slide

4. codecentric AG
   Map carefully
   − Disable the _all field, you definitely don’t want it
   !
   − Keep the _source field enabled and don’t set any fields to _stored
   !
   − Disable dynamic mapping (except where really needed)
   !
   − Choose analyzers carefully (for many fields, „not_analyzed“ might be enough)
   !
   − Consider mapping fields more than once
   !
   − Existing mappings cannot be changed without deleting the type
   

   View Slide

5. codecentric AG
   Lessons learned
   Understand filters
   

   View Slide

6. codecentric AG
   Understand filters
   − Use filters instead of queries whenever you don’t need scoring
   − Elasticsearch is able to cache the results of filters
   !
   − Tricky caching behavior
   − Most simple filters are cached by default, but some not (e.g., geo)
   − Compound filters (bool/and/or/not) are not cached
   − You can still explicitly request caching by setting _cache
   − Bool filters query the cache for their (sub-)filters, but and/or/not filters don’t
   !
   − But: This topic seems to be a moving target
   !
   − Consider the scope of filters
   − Apply to query, facets/aggregations, or both?
   − Often „filtered query“ is what you need
   

   View Slide

7. codecentric AG
   Lessons learned
   Use analysis tooling
   

   View Slide

8. codecentric AG
   Use analysis tooling
   − Use the explain feature for search result scores
   !
   − Use the validate/explain feature to see what the query looks like after rewriting
   !
   − Make sure your analyzers work correctly
   − Use the analyze API or the „inquisitor“ plugin
   − And maybe the „extended-analyze“ plugin
   !
   − When in doubt, take a look at the terms in your index
   − Use the Lucene tool „Luke“
   − http://rosssimpson.com/blog/2014/05/06/using-luke-with-elasticsearch/
   − The „skywalker“ plugin is also useful to understand Elasticsearch indexing
   

   View Slide

9. codecentric AG
   Lessons learned
   Create regression tests
   

   View Slide

10. codecentric AG
    Create regression tests
    − Set up a comprehensive test suite
    − Test expectations about matches
    − Prevent regressions when changing or modifying analyzers
    !
    − The Elasticsearch Java client is embeddable
    − Consider using it for a unit test suite even if not using Java otherwise
    !
    − Try it by solving the „mapping challenge“!
    − https://github.com/peschlowp/elasticsearch-mapping-challenge
    

    View Slide

11. codecentric AG
    Lessons learned
    Know some Lucene basics
    

    View Slide

12. codecentric AG
    Know some Lucene basics
    

    View Slide

13. codecentric AG
    Know some Lucene basics
    Segment
    flush()
    

    View Slide

14. codecentric AG
    Know some Lucene basics
    Segment
    flush()
    

    View Slide

15. codecentric AG
    Know some Lucene basics
    Segment
    flush()
    Segment
    flush()
    

    View Slide

16. codecentric AG
    Know some Lucene basics
    Segment
    flush()
    Segment
    flush()
    commit()
    Synced to Disk
    

    View Slide

17. codecentric AG
    Know some Lucene basics
    Visible to newly opened readers
    Segment
    flush()
    Segment
    flush()
    commit()
    Synced to Disk
    If desired, visible via NRT
    

    View Slide

18. codecentric AG
    Know some Lucene basics
    Segment
    flush()
    Segment
    flush()
    commit()
    Synced to Disk
    Executed heuristically
    (or explicitly via NRT)
    Explicit call (transaction)
    

    View Slide

19. codecentric AG
    Know some Lucene basics
    

    View Slide

20. codecentric AG
    Know some Lucene basics
    Persisted
    

    View Slide

21. codecentric AG
    Know some Lucene basics
    Persisted
    refresh()
    

    View Slide

22. codecentric AG
    Know some Lucene basics
    Persisted
    + Reopen reader

    for NRT
    refresh()
    Segment
    flush()
    

    View Slide

23. codecentric AG
    Know some Lucene basics
    Persisted
    + Reopen reader

    for NRT
    refresh()
    Segment
    flush()
    

    View Slide

24. codecentric AG
    Know some Lucene basics
    flush()
    Persisted
    + Reopen reader

    for NRT
    refresh()
    Segment
    flush()
    

    View Slide

25. codecentric AG
    Know some Lucene basics
    Segment
    flush()
    Synced to Disk
    Persisted
    refresh()
    Segment
    flush() commit()
    + Reopen reader
    

    View Slide

26. codecentric AG
    Know some Lucene basics
    Segment
    flush()
    Persisted
    refresh()
    Segment
    flush()
    Executed heuristically
    Executed regularly
    commit()
    Synced to Disk
    + Reopen reader
    

    View Slide

27. codecentric AG
    Know some Lucene basics
    All documents persisted and searchable. Transaction log can be cleared.
    

    View Slide

28. codecentric AG
    Lessons learned
    Understand updates
    

    View Slide

29. codecentric AG
    Understand updates
    − With Lucene we have Update = Delete + Add
    !
    − Elasticsearch offers two types of update: „Partial document“ and „Script“
    − Internally still delete + add
    − This means even small updates might take a while
    !
    − Partial document update trickiness
    − Updated fields are replaced, except for inner objects which are merged
    − To replace inner objects, consider wrapping them in a single-element array
    

    View Slide

30. codecentric AG
    Lessons learned
    Understand relations
    

    View Slide

31. codecentric AG
    Understand relations
    − Lucene documents are flat
    − Arrays of inner objects are flattened when indexing
    − Cannot build one-to-many relationships this way
    !
    − Elasticsearch offers two alternatives: nested objects and parent/child mapping
    !
    − Parent/child mapping is very flexible
    − Documents stay separated
    − May be queried together
    !
    − Nested objects
    − Better performance, but documents are combined into one
    

    View Slide

32. codecentric AG
    Lessons learned
    Embrace aliases
    

    View Slide

33. codecentric AG
    Embrace aliases
    − A logical name for one or more Elasticsearch index(es)
    − Decouples client view from physical storage
    !
    − Use cases
    − (Read-only) views on one or multiple indexes
    − Dynamically split indexes without affecting clients
    − Zero downtime reindexing
    !
    − Limitations
    − Writes are only permitted for aliases that point to a single index
    !
    − Recommendation: Use aliases right from the start
    

    View Slide

34. codecentric AG
    Lessons learned
    Don’t underestimate reindexing
    

    View Slide

35. codecentric AG
    Lessons learned
    Adjust default settings
    

    View Slide

36. codecentric AG
    Adjust default settings
    − Choose a unique cluster name
    !
    − Consider using unicast discovery
    !
    − Configure (at least) these settings according to your requirements
    − minimum_master_nodes
    − gateway.recover_after_nodes
    − gateway.recover_after_time
    − gateway.expected_nodes
    

    View Slide

37. codecentric AG
    Lessons learned
    Think about read preference
    

    View Slide

38. codecentric AG
    Think about read preference
    − Search has a „preference“ parameter
    − Which shard (primary or some replica) to contact
    !
    − By default „round robin“
    − May cause some nasty effects
    − Documents marked as deleted still affect scoring
    − If segment merging on shards differs, may return different search results
    !
    − Set „preference“ according to your needs:
    − Possible values: local, primary, only some shards or nodes, arbitrary string
    − One approach: Always direct a user to the same shard
    

    View Slide

39. codecentric AG
    Lessons learned
    Satisfy scaling requirements
    

    View Slide

40. codecentric AG
    Satisfy scaling requirements
    − Number of shards needs to be fixed at index creation time
    !
    − Choose number of shards depending on estimation and measurements
    − A little overallocation is OK, yet shards don’t come for free
    !
    − If it turns out you need to scale higher, index aliases may help
    − Create another, identically configured index and add new documents to it
    − Define an alias so that search considers both indexes
    − But: Requires additional effort for updates and deletes (which index to address?)
    − Alternative: Migrate to a new index with more shards
    !
    − Remember:
    − Search 1 index with 50 shards =~ Search 50 indexes with 1 shard each
    − In both cases, 50 Lucene indexes are searched
    

    View Slide

41. codecentric AG
    Lessons learned
    Give plugins a try
    

    View Slide

42. codecentric AG
    Resources
    − The official blog

    http://www.elasticsearch.org/blog/
    !
    − The official book

    http://www.elasticsearch.org/guide/en/elasticsearch/guide/current/index.html
    !
    − The official reference

    http://www.elasticsearch.org/guide/en/elasticsearch/reference/current/index.html
    !
    − The official Google Group

    https://groups.google.com/d/forum/elasticsearch
    !
    − The Sense examples shown in this talk

    https://gist.github.com/peschlowp/560411af9bac3be909c0
    

    View Slide

43. codecentric AG
    Questions?
    Dr. rer. nat. Patrick Peschlow

    codecentric AG

    Merscheider Straße 1

    42699 Solingen

    

    tel +49 (0) 212.23 36 28 54

    fax +49 (0) 212.23 36 28 79

    patrick.peschlow@codecentric.de

    

    www.codecentric.de
    

    View Slide