SlideShare a Scribd company logo

Debugging and Testing ES Systems

•
•
Chris Birchall
Chris Birchall

This document discusses debugging and testing Elasticsearch systems. It provides tips for debugging issues like typos in mappings, setting up a local environment for testing, useful commands like analyze and explain, tuning queries, and testing strategies using Java and Ruby. The document emphasizes the importance of testing representative queries to ensure expected results and the ability to tune queries without breaking other queries. It also recommends using Elasticsearch plugins like Head for visualizing clusters and indices.

1 of 21
Download to read offline
Debugging and Testing ES Systems Chris Birchall 2013/8/29 Elasticsearch 勉強会 第１回 #elasticsearchjp
Elasticsearch and me ● At Infoscience, helped build a log management product based on ES + Hadoop ● At M3, ES evangelist (??) ○ Maintain ES cluster ○ Help dev teams integrate ES into their apps Twitter: @cbirchall Github: https://github.com/cb372
Search at M3 ● Using ES for all new services ○ Search, recommendation (MoreLikeThis) ● Slowly migrating other services from Solr ● A few legacy services use Lucene directly ● Running all indices on one ES cluster ● Kuromoji for Japanese content
Debugging Mostly debugging of queries ● “Why doesn’t doc X match query Y?” ● “Why does this search return no results?” Operational issues are very rare ● ES’s clustering magic is surprisingly stable! ● No performance issues so far
Debugging - Step 1 Check for typos! ES will silently ignore many typos in settings/mapping definitions
Typo - Example $ curl -X PUT localhost:9200/myapp -d '{ "settings": { "number_of_shards": 3 }, "mapping" : { "article" : { "_source": { "enabled": false }, "properties": { "title": { "type": "string", "store": "true" }, "body": { "type": "string", "store": "true" }, ... } }, ... }' Let’s create a new index...
Typo - Example (cont’d) {"ok":true,"acknowledged":true} Response from ES: OK, seems fine...
Typo - Example (cont’d) $ curl localhost:9200/myapp/_mappings?pretty Response from ES: { "myapp" : { } } Eh? Where are my lovingly-crafted mappings?! Now check the mappings...
Typo - Example (cont’d) $ curl -X PUT localhost:9200/myapp -d '{ "settings": { "number_of_shards": 3 }, "mappings" : { "article" : { "_source": { "enabled": false }, "properties": { "title": { "type": "string", "store": "true" }, "body": { "type": "string", "store": "true" }, ... } }, ... }' Oops!
Debugging - Step 2 Set up a local environment ● Makes it easy to wipe & rebuild index
Setting up a local env (OSX) # Install $ brew install elasticsearch # Kuromoji plugin (optional) $ /usr/local/opt/elasticsearch/bin/plugin -install elasticsearch/elasticsearch-analysis-kuromoji/1.5.0 # Start $ elasticsearch # Create index $ curl -X PUT localhost:9200/my_app -d '{ ... }' # Insert some documents $ curl -X PUT localhost:9200/my_app/my_type/1 -d '{ ... }' $ curl -X PUT localhost:9200/my_app/my_type/2 -d '{ ... }' # Done!
Useful commands - Analyze $ curl 'localhost:9200/myindex/_analyze?pretty' / -d '東京特許許可局許可局長' { "tokens" : [ { "token" : "東京", "start_offset" : 0, "end_offset" : 2, "type" : "word", "position" : 1 }, { "token" : "特許", "start_offset" : 2, "end_offset" : 4, "type" : "word", ... How is my document/query being tokenized?
Useful commands - Explain $ curl 'localhost:9200/kuro/docs/123/_explain?pretty' / -d '{ "query": { "term": { "body": "東京" } } }' { ... "matched" : true, "explanation" : { "value" : 0.375, "description" : "weight(body:東京 in 0) [PerFieldSimilarity], result of:", "details" : [ { "value" : 0.375, "description" : "fieldWeight in 0, product of:", "details" : [ { "value" : 1.0, "description" : "tf(freq=1.0), with freq of:", "details" : [ { "value" : 1.0, "description" : "termFreq=1.0" ... Why does this document (not) match this query? Specify document ID
Tuning queries Parameters to tweak ● default_operator (AND/OR) ● auto_generate_phrase_queries ● minumum_should_match ● Stop words/tags ● Kuromoji ○ Segmentation mode ○ Reading form filter ○ Disable Kuromoji! (for some fields)
Why disable Kuromoji? Problem: occasionally weird tokenization ● AND query will fail, because not all terms match ● OR query will match any document with 病院 → low precision Phrase Terms 特定医療法人財団　日本会　東日本病院 (document field) 特定、医療、法人、財団、 日本、会、東日本、病院 東日本 (query) 東日、東日本、本 東日本病院 (query) 東、東日本、日本、病院
Useful plugin - Head $ bin/plugin -install mobz/elasticsearch-head http://mobz.github.io/elasticsearch-head/
Testing Main goal: Ensure that queries return the results that we expect ● Test coverage of representative queries ○ Freedom to tune for a given query without breaking other queries Ideally, tests should: ● Run fast ● Run standalone (i.e. no need to have an ES server running)
Testing - Java elasticsearch-test is awesome ● DSL to set up/tear down ES ● Annotations + JUnit runner ● ES runs in-process ○ No need to start an external ES server ● Index is stored in-memory ○ Runs quickly https://github.com/tlrx/elasticsearch-test
https://github.com/cb372/elasticsearch-test-example Testing - Java Simple elasticsearch-test example
Testing - Ruby Simple Rails + Tire + RSpec example https://github.com/cb372/elasticsearch-rspec-example
We’re hiring! TODO We are hiring slide http://bit.ly/m3jobs

More Related Content

Debugging and Testing ES Systems

  • 1. Debugging and Testing ES Systems Chris Birchall 2013/8/29 Elasticsearch 勉強会 第１回 #elasticsearchjp
  • 2. Elasticsearch and me ● At Infoscience, helped build a log management product based on ES + Hadoop ● At M3, ES evangelist (??) ○ Maintain ES cluster ○ Help dev teams integrate ES into their apps Twitter: @cbirchall Github: https://github.com/cb372
  • 3. Search at M3 ● Using ES for all new services ○ Search, recommendation (MoreLikeThis) ● Slowly migrating other services from Solr ● A few legacy services use Lucene directly ● Running all indices on one ES cluster ● Kuromoji for Japanese content
  • 4. Debugging Mostly debugging of queries ● “Why doesn’t doc X match query Y?” ● “Why does this search return no results?” Operational issues are very rare ● ES’s clustering magic is surprisingly stable! ● No performance issues so far
  • 5. Debugging - Step 1 Check for typos! ES will silently ignore many typos in settings/mapping definitions
  • 6. Typo - Example $ curl -X PUT localhost:9200/myapp -d '{ "settings": { "number_of_shards": 3 }, "mapping" : { "article" : { "_source": { "enabled": false }, "properties": { "title": { "type": "string", "store": "true" }, "body": { "type": "string", "store": "true" }, ... } }, ... }' Let’s create a new index...
  • 7. Typo - Example (cont’d) {"ok":true,"acknowledged":true} Response from ES: OK, seems fine...
  • 8. Typo - Example (cont’d) $ curl localhost:9200/myapp/_mappings?pretty Response from ES: { "myapp" : { } } Eh? Where are my lovingly-crafted mappings?! Now check the mappings...
  • 9. Typo - Example (cont’d) $ curl -X PUT localhost:9200/myapp -d '{ "settings": { "number_of_shards": 3 }, "mappings" : { "article" : { "_source": { "enabled": false }, "properties": { "title": { "type": "string", "store": "true" }, "body": { "type": "string", "store": "true" }, ... } }, ... }' Oops!
  • 10. Debugging - Step 2 Set up a local environment ● Makes it easy to wipe & rebuild index
  • 11. Setting up a local env (OSX) # Install $ brew install elasticsearch # Kuromoji plugin (optional) $ /usr/local/opt/elasticsearch/bin/plugin -install elasticsearch/elasticsearch-analysis-kuromoji/1.5.0 # Start $ elasticsearch # Create index $ curl -X PUT localhost:9200/my_app -d '{ ... }' # Insert some documents $ curl -X PUT localhost:9200/my_app/my_type/1 -d '{ ... }' $ curl -X PUT localhost:9200/my_app/my_type/2 -d '{ ... }' # Done!
  • 12. Useful commands - Analyze $ curl 'localhost:9200/myindex/_analyze?pretty' / -d '東京特許許可局許可局長' { "tokens" : [ { "token" : "東京", "start_offset" : 0, "end_offset" : 2, "type" : "word", "position" : 1 }, { "token" : "特許", "start_offset" : 2, "end_offset" : 4, "type" : "word", ... How is my document/query being tokenized?
  • 13. Useful commands - Explain $ curl 'localhost:9200/kuro/docs/123/_explain?pretty' / -d '{ "query": { "term": { "body": "東京" } } }' { ... "matched" : true, "explanation" : { "value" : 0.375, "description" : "weight(body:東京 in 0) [PerFieldSimilarity], result of:", "details" : [ { "value" : 0.375, "description" : "fieldWeight in 0, product of:", "details" : [ { "value" : 1.0, "description" : "tf(freq=1.0), with freq of:", "details" : [ { "value" : 1.0, "description" : "termFreq=1.0" ... Why does this document (not) match this query? Specify document ID
  • 14. Tuning queries Parameters to tweak ● default_operator (AND/OR) ● auto_generate_phrase_queries ● minumum_should_match ● Stop words/tags ● Kuromoji ○ Segmentation mode ○ Reading form filter ○ Disable Kuromoji! (for some fields)
  • 15. Why disable Kuromoji? Problem: occasionally weird tokenization ● AND query will fail, because not all terms match ● OR query will match any document with 病院 → low precision Phrase Terms 特定医療法人財団　日本会　東日本病院 (document field) 特定、医療、法人、財団、 日本、会、東日本、病院 東日本 (query) 東日、東日本、本 東日本病院 (query) 東、東日本、日本、病院
  • 16. Useful plugin - Head $ bin/plugin -install mobz/elasticsearch-head http://mobz.github.io/elasticsearch-head/
  • 17. Testing Main goal: Ensure that queries return the results that we expect ● Test coverage of representative queries ○ Freedom to tune for a given query without breaking other queries Ideally, tests should: ● Run fast ● Run standalone (i.e. no need to have an ES server running)
  • 18. Testing - Java elasticsearch-test is awesome ● DSL to set up/tear down ES ● Annotations + JUnit runner ● ES runs in-process ○ No need to start an external ES server ● Index is stored in-memory ○ Runs quickly https://github.com/tlrx/elasticsearch-test
  • 20. Testing - Ruby Simple Rails + Tire + RSpec example https://github.com/cb372/elasticsearch-rspec-example
  • 21. We’re hiring! TODO We are hiring slide http://bit.ly/m3jobs