ElasticSearch - index server used as a document database

Elasticsearch - index server used as a document database

!(with examples)

!Robert Lujo, 2014

about me

software

professionally 17 y.

freelancer

more info -> linkedin

Elasticsearchsearch server based on Apache Lucene

distributed, multitenant-capable

full-text search engine

RESTful web interface

schema-free JSON documents

NoSQL capabilities

https://en.wikipedia.org/wiki/Elasticsearch

Elasticsearchfirst release in February 2010

until now raised total funding > $100M

latest release 1.3 & 1.4 beta

+ Logstash+ Kibana => ELK stack

Apache 2 Open Source License

Very popularand used by

!

!

!

!

!

… Wikimedia, Mozilla, Stack Exchange, Quora, CERN …!

!

Professional services also available

What about docs?

FeaturesSys: !

real time data, distributed, multi-tenancy, real time analytics, high availability

Dev:!

restful api, document oriented, schema free, full text search, per-operation persistence, conflict management

http://www.elasticsearch.org/overview/elasticsearch/

Install, run …

prerequisite: JDK - Java (Lucene remember?)

wget https://download.elasticsearch.org/.../elasticsearch-1.3.4.zip

unzip elasticsearch-1.3.4.zip

elasticsearch-1.3.4/bin/elasticsearch

& use!# curl localhost:9200

{ "status" : 200, "name" : "The Night Man", "version" : { "number" : "1.3.4", "build_hash" : “…”, "build_timestamp" : “…”, "build_snapshot" : false, "lucene_version" : "4.9" }, "tagline" : "You Know, for Search" }

create index & put some data

# curl -XPUT localhost:9200/mainindex/company/1 -d '{ "name" : "CoolComp Ltd.", "employees" : 10, "founded" : "2014-10-05", "services" : ["software", "consulting"], "management": [ {"role" : "CEO", "name" : "Petar Petrovich"}, {"name" : "Ivan Ivić"} ], "updated" : "2014-10-05T22:31:55" }’ => {"_index":"mainindex","_type":"company","_id":"1","_version":4,"created":false}

fetch document by id (key/value database)

# curl -XGET localhost:9200/mainindex/company/1 !=> !{“_index":"mainindex", “_type":"company", “_id”:"1","_version" : 4, ”found”:true, "_source":{ "name" : "CoolComp Ltd.", "employees" : 10, … }}

search documents# curl -XGET 'http://localhost:9200/maindex/_search?q=management.name:petar' # no type! {“took”:128,"timed_out":false,"_shards":{"total":5,"successful":5,"failed":0}, “hits”:{ "total":1, "max_score":0.15342641, “hits” : [ {“_index":"mainindex","_type":"company", “_id":"1", “_score":0.15342641, "_source":{ "name" : "CoolComp Ltd.", … "updated" : "2014-10-05T22:31:55"

Database is …an organized (or structured) collection of data

!

Database management system (DBMS) is …!

software system provides interface between users and database(s)

4 common groups of interactions:

1. Data definition

2. Update - CrUD

3. Retrieval - cRud

4. Administration

Elasticsearch is a database?

1. Data definition

2. Update - CrUD

3. Retrieval - cRud

4. Administration

Data representation - document-oriented-databaseDocument-oriented-database - “NoSql branch”? Not really but …

Document is … blah blah blah … something like this: !

{ “_id” : 1, “_type” : “company”, "name" : "CoolComp Ltd.", "employees" : 10, "founded" : "2014-10-05", "services" : ["software", "consulting"], "management": [ {"role" : "CEO", "name" : "Petar Petrovich"}, {"name" : "Ivan Ivić"} ], "updated" : "2014-10-05T22:31:55" }

Data representation - relational databases

company: id name employees founded -- --------------- --------- ------------ 1 'CoolComp Ltd.' 10 '2014-10-05' !services: id value --- ---------------- 1 'software' 2 'consulting' !company_services: id id_comp id_serv --- ------- -------- 1 1 1 2 1 2 !person: id name -- ----------------- 1 'Petar Petrovich' 2 'Ivan Ivić' comp_management: id id_comp id_pers role --- ------- ------- ----- 1 1 1 CEO 2 1 2 MEM

Data definitionElasticsearch is “schemaless”

But it provides defining schema - mappings

Very important when setting up for search:

• data types - string, integer, float, date/tst, boolean, binary, array, object, nested, geo, attachment

• search analysers, boosting, etc.

Data definition - compared to RDBMS

But we loose some things what RDBMS offers:

• data validation / integrity

• removing data redundancy - normalization

• “fine grained” structure definition

• standard and common usage (SQL)

RetrievalWe had this example before:!

!# curl -XGET 'http://localhost:9200/maindex/_search?q=management.name:petar' # no type!

!equivalent SQL query:!

!select * from company where exists( select 1 from comp_management cm inner join peron p on p.id=cm.id_pers where lower(p.name) like '%peter%');

Retrieval - ES-QDSL

based on my experience, I would rather use ES:• for searches: full text, fuzzy, multi field, multi

document types, multi indexes/databases• in programming - better to convert/deal with

JSON than with ORM/raw SQL results• single web page applications

Retrieval - SQL

on the other hand, I would rather use SQL and RDBMS:

• when composing complex query - easier to do with SQL

• for data exploring/researching!

SQL is much more expressive DSL

Joining & denormalization

object hierarchy … must be denormalized.

increases retrieval performance (since no query joining is necessary),

uses more space

makes keeping things consistent and up-to-date more difficult

They’re excellent for write-once-read-many-workloads

https://www.found.no/foundation/elasticsearch-as-nosql/

Joining optionsES has several ways to “join” objects/documents/types:

1. embedding objects

2. “nested” objects

3. parent / child relation between types

4. compose manual query

When fetching by id - very handy (1 & 2).

When quering - not so handy.

Updating - CrUD !Elasticsearch

I would rather use Elasticsearch:• when creating, updating and deleting single

nested document

Updating - CrUD!RDBMS

on the other hand, RDBS I found handy:• for flat entities/documents• for mass objects manipulation• transactions & integrity (ACID)

Administrationinstall, configure, maintenance, monitoring, scaling … quite satisfing!!

OS specific install - apt-get, yum, zypper, brew, …!

plugins installation ./bin/plugin -i Elasticsearch/marvel/latest

Administration - tools

Elasticsearch as Database

!

!

!

!

!

to avoid maintenance and development time overhead

Hybrid solutionElasticsearch + …

ES + … - hybrid solution

So why can you use ElasticSearch as a single point of truth (SPOT)?

Elasticsearch … used in addition to another database.

A database system for constraints, correctness and robustness, transactionally updatable,

master record which is then asynchronously pushed/pulled to Elasticsearch

Hybrid solution

Elasticsearch riversbesides classic indexing - rivers provide alternative way for inserting data into ES

service that fetches the data from an external source (one shot or periodically) and puts the data into the cluster

Besides listed on official site:

• RDBMS/JDBC

• MongoDB

• Redis

• Couchbase

• …

• Indexing & reindexing subdocuments is major job

• upsert mode• issues - not indexing, memory hungry, full

reindex when new field/subdoc• building AST when building a query - quite

demanding• satisfied with the final result!

Use-case - RDBMS & Elasticsearch

What about others?

Riak & Solr

September 16, 2014 - With 2.0, we have added distributed Solr to Riak Search. For every instance of Riak, there is an instance of Solr. While this drastically improves full-text search, it also improves Riak’s overall functionality. Riak Search now allows for Riak to act as a document store (instead of key/value) if needed.

Despite being a part of Riak, Riak Search is a separate Erlang application. It monitors for changes to data in Riak and propagates those changes to indexes managed by Solr.

Couchbase and Elasticsearch

integrates Couchbase Server and Elasticsearch,

by streaming data in real-time from Couchbase to Elasticsearch.

combined solution … with full-text search, indexing and querying and real-time analytics … content store or aggregation of data from different data sources.

Couchbase Server provides easy scalability, low-latency document access, indexing and querying of JSON documents and real-time analytics with incremental map reduce.

MongoDB and Elasticsearch

“addition of Elasticsearch represents only a first step in its mission to enable developers to choose the database that's right for their needs”

“big weakness of MongoDB is the free text search, which MongoDB tried to address in version 2.4 in some aspects.”

Not to forget good old school …

RDBMS with FTS

Elasticsearch use when …

you need very good, reliable, handy, web oriented search index engine

you have intensive read and document oriented application

“write” balance - depending on how much - ES as a NoSQL only or as a hybrid solution

Summaryno silver bullet, “the right tool for the job”

learn & get familiar with different solutions and choose optimal one

be objective & productive

General trend are heterogenous => lot of integration tasks lately

learn new things & have fun!

Thank you for your patience!

Questions?

!

[email protected]

@trebor74hr