Elasticsearch [email protected] - CTO Federico Panini CTO @ fazland.com email : [email protected] LikedIn : https://uk.linkedin.com/in/federicopanini slides : http://www.slideshare.net/FedericoPanini
Aug 16, 2015
Elasticsearch
[email protected] - CTO
Federico Panini
CTO @ fazland.comemail : [email protected]
LikedIn : https://uk.linkedin.com/in/federicopaninislides : http://www.slideshare.net/FedericoPanini
What is Elasticsearch
[email protected] - CTO
full-text search engine
“A search engine is an automated system which, upon request, uses a set of data and return an index of its content classifying them based on math/stats algorithm used to set the relevance, based in a search key.”
[email protected] - CTO
“It’s a distributed, scalable, and highly available Real-time search and analytics software.”
What’s Elasticsearch ?full-text search engine
[email protected] - CTO
Real-time data Realtime data analysis Distributed system High Availability Full-text searches Document oriented DB Schemaless DB
RESTFul Api Persistence per-operation Open Source Based on Apache Lucene Optimistic version control
What’s Elasticsearch ?features
Apache Lucene #1
[email protected] - CTO
It’s the heart of Elasticsearch
Lucene is the search engine of Elasticsearch
Apache Lucene #1
[email protected] - CTO
It’s in Java
It’s an Apache Software Foundation, so Open Source!
What has more than Lucene
[email protected] - CTO
full-text searches
horizontal scaling high availability Easy to use near real time
Architecture
[email protected] - CTO
requirements - CPU
Elasticsearch doesn’t need a lot of CPU.
The advice is to use the last CPU model available.
In general is a good practice to use machines with 2 to 8 cores.
Architecture
[email protected] - CTO
requirements - Disco
The I/O disk need is really important for all clusters.
Please use SSD disks.
Architecture
[email protected] - CTO
requirements - HD - bonus slide …
One very important thing to know is you have to pay attention where data is stored and mostly how. The word you have to remember is scheduler. The scheduler on *nix system is responsible to decide when data should be “written” to disc and on which priority. Usually common unix OS setup cfq as scheduler, which for instance is a scheduler for rotating disks and optimised for them. The advice is to use SSD disks and to setup the SO to use “noop” or “deadline” which are scheduler optimised for SSD’s.
If you use the right scheduler you can reach improvements of 500x !!!
[email protected] - CTO
Operating Systems
Elasticsearch is written in Java, so it’s a multiplatform solution. Use the last JDK available.
Architecture
[email protected] - CTO
requirements - RAM
Elasticsearch is eager of RAM!!!
https://www.elastic.co/guide/en/elasticsearch/guide/current/heap-sizing.html
Architecture
[email protected] - CTO
memory !?!?
Use solutions with 64GB is fine not more give to the Java heap size not more than 32GB of RAM use more than one machine for elasticsearch in order
setup correctly the cluster.
Architecture
[email protected] - CTO
Installation
curl -L -O http://download.elasticsearch.org/PATH/TO/VERSION.zip unzip elasticsearch-$VERSION.zip cd elasticsearch-$VERSION
There are availbes packages for many distribution as Debian or RPM, and Puppet or Chef modules
Architecture
Java based
[email protected] - CTO
elasticsearch
Elasticsearch has been developed in JAVA
Robust Scalable Multiplatform
Talking to Elasticsearch
[email protected] - CTO
clients Java #1
There are 2 clients available in JAVA:
Node client : the client join the cluster as non-data node, this mean that the client knows perfectly where data are and on which node of the cluster.
[email protected] - CTO
clients Java #2
Transport client : is a lightweight client and is the tool used to comunicate with the cluster remotely.
Talking to Elasticsearch
There are 2 clients available in JAVA:
[email protected] - CTO
clients Java #2
Both Java clients talk to the cluster on port 9300, which is the same port use by the cluster itself.
Talking to Elasticsearch
There are 2 clients available in JAVA:
[email protected] - CTO
client API RESTful
All programming languages other than Java can talk to the Elasticsearch cluster through its API Rest available on port 9200.
There are many official clients available in different programming languages.: Groovy, JavaScript, .NET, PHP, Perl, Python, e Ruby
Talking to Elasticsearch
Elastic
[email protected] - CTO
Document oriented
NoSqlElasticsearch is a document oriented database. This mean Elasticsearch is a schema-less database.
After inserting documents inside Elasticsearch, the documents will be immediately indexed.
Elastic
[email protected] - CTO
Document oriented
JSONElasticseach uses JSON as interchange language between the server and the API clients.
Elastic
[email protected] - CTO
glossary
cluster nodes indexes shards replica segments in-memory buffers translog
Elastic
[email protected] - CTO
cluster
The cluster is a set which belong one or more nodes, which shares the same property cluster.name. The cluster is used to balance the load of the server itself. A node could be deleted or inserted to the cluster, the cluster itself will re-organise itself.
Elastic
[email protected] - CTO
cluster
Inside a cluster a node is elected as Master. The Master node is responsible to manage operations as creation or removal indexes, join or deletion of a node. Every node could be elected as Master.
Elastic
[email protected] - CTO
nodes
A node is a minimum element of Elasticsearch that ensures the proper working of the cluster.
Elastic
[email protected] - CTO
shards
If we want to start indexing data on Elasticsearch we need to create an index. Index is the term used only to identify a logical definition, which represent a pointer to one or more elements called SHARDS.
Elastic
[email protected] - CTO
shards
The shard is the low level element of Elasticsearch, and contains a subset of all the data inside and index.
The shard is in fact a single instance of Apache Lucene.
Elastic
[email protected] - CTO
Replica shards
Replica shards are mirrors of shards used to protect our data from hardware failures. As the shards they are used exactly as the shards.
Elastic
[email protected] - CTO
shards immutability
The number of shards for an index is defined at Index creation time and is IMMUTABLE.
Elastic
[email protected] - CTO
shards immutability
curl -X http://localhost:9200/blogs -d ‘{ "settings" : { "number_of_shards" : 3, "number_of_replicas" : 1 } }’
Elastic
[email protected] - CTO
shards immutability
curl http://localhost:9200/_cluster/health“{ "cluster_name": "elasticsearch", "status": "yellow", "timed_out": false, "number_of_nodes": 1, "number_of_data_nodes": 1, "active_primary_shards": 3, "active_shards": 3, "relocating_shards": 0, "initializing_shards": 0, "unassigned_shards": 3 }”
Elastic
[email protected] - CTO
shards immutability
Replica shards on a single node instance are useless, the meaning for cluster is nothing in this case. To make replica shard useful we need at least 2 nodes to have data redundancy.
Elastic
[email protected] - CTO
BONUS : manage conflicts #2 : Pessimistic Concurrency ControlUsed in standard RDBMS
This approach is based on the concept that conflict could happened frequently and so to avoid them the RDBMS lock the resource.
The process lock the access to the row before reading it, this way we the RDBMS is sure that only one process will access to this thread and can subsequently modify it and nobody else. At the end of its process (update/delete) the thread will release the LOCK.
Elastic
[email protected] - CTO
BONUS : manage conflicts #3 : Optimistic Concurrency Control
Elasticsearch uses OCC
This approach will consider conflicts as infrequent. The database won’t lock the resource when access to it.
The responsibility is given to the application : when data is amended between a read and write then the update fails. In this case you need to re-get the fresh new data and trying to update it.
Elastic
[email protected] - CTO
BONUS : manage conflicts#4 : Optimistic Concurrency Control
Elasticsearch is a distributed solution, concurrent and asynchronous. When a document is created / updated / deleted is absolutely necessary to replicate this information across the whole cluster.
Every command sent to the nodes is sent in parallel and could happen that some data will reach its destination (node) already expired.
Elastic
[email protected] - CTO
BONUS : manage conflicts#5 : Optimistic Concurrency Control
We need a way to understand that the entry we’re trying to update as been already updated by another process.
Elastic
[email protected] - CTO
BONUS : manage conflicts#6 : Optimistic Concurrency Control
VERSIONING
Elastic
[email protected] - CTO
BONUS : manage conflicts#7 : Optimistic Concurrency Control
In Elasticsearch every document has a field named:
_version
This system field is incremented every time an operation (update / delete) occurs over a document. In this way an update to _version:3 won’t be never applied to a document whose _version field value is at 4.
Elastic
[email protected] - CTO
BONUS : manage conflicts #8 : Optimistic Concurrency Control
This approach move all the responsibility from the database to the application! so WE are responsible to not create conflicts over a document or and index. If we want to be sure to not have loss of data we nee to implement writes with the use of versioning!
Elastic
[email protected] - CTO
BONUS : manage conflicts #9 : Optimistic Concurrency Control
http://www.jillesvangurp.com/2014/12/03/optimistic-locking-for-updates-in-elasticsearch/ h t tps : / / aphy r.com/pos ts /317 -ca l l -me-maybe-elasticsearch https://www.elastic.co/guide/en/elasticsearch/resiliency/current/index.html
Elastic
[email protected] - CTO
Simple searches - CREATE AN INDEX
curl -XPUT http://fazlab.fazland.com:9200/fazlab-d "{ "settings" :
{ "number_of_shards" : 3, "number_of_replicas" : 1
} }"
Elastic
[email protected] - CTO
Simple searches - INDEX A DOCUMENT
curl -XPUThttp://fazlab.fazland.com:9200/fazlab/categories/1?pretty -d '{
nome: "Federico"}'
Elastic
[email protected] - CTO
Simple searches - GET A DOCUMENT
curl http://fazlab.fazland.com:9200/fazlab/categories/1?pretty
Elastic
[email protected] - CTO
Simple searches - DELETE A DOCUMENT
curl -XDELETE http://fazlab.fazland.com:9200/fazlab/categories/2?pretty
Elastic
[email protected] - CTO
mapping and analysis
EXACT MATCH vs FULL TEXT
Exact match Full Text
where name = ‘Federico’
and user_id = 2
and date > “2014-09-15”
“Frank has been to
South beach”
Frank / FRANK / frank
Elastic
[email protected] - CTO
mapping and analysis
EXACT MATCH vs FULL TEXT
Exact match
Full Text
binary : the document contains these values ?
How much is relevant the document compared to the
term used inside the query ?
Elastic
[email protected] - CTO
mapping and analysis
Elasticsearch to help a full-text search analyse the text and uses this result to build an inverted index.
Inverted Index Analyzer
Elastic
[email protected] - CTO
Inverted Index
1. The quick brown fox jumped over the lazy dog
2. Quick brown foxes leap over lazy dogs in summer
Elastic
[email protected] - CTO
Inverted Index
If we want to search the word “quick” and “brown” we will pick
only the documents where these 2 words are.
1. The quick brown fox jumped over the lazy dog
2. Quick brown foxes leap over lazy dogs in summer
Elastic
[email protected] - CTO
Inverted Index
1. The quick brown fox jumped over the lazy dog
2. Quick brown foxes leap over lazy dogs in summer
Elastic
[email protected] - CTO
ANALYZERS
An analyzer has 3 functions:
Character filters
Tokenizer
Token Filters
Elastic
[email protected] - CTO
ANALYZERS - Character Filters
The first part of an analyser is to parse every string with character filer which will clean / reorganize the strings before tokenization.
During this phase special HTML chars will be removed or & will be converted in AND.
Elastic
[email protected] - CTO
ANALYZERS - Tokenizer
The second phase of an analyser is tokenisation which will divide a sentence in small terms.
Elastic
[email protected] - CTO
ANALYZERS - Token FiltersSuccessivamente alla fase di Tokenizzazione delle stringhe in singoli termini (terms), i filtri (selezionati) sono applicati in sequenza. After tokenisation filters will be applied in sequence. For example :
- put lower case the whole text - remove stop words - add synonyms
Elastic
[email protected] - CTO
Standard Analyzer
“Set the shape to semi-transparent by calling set_trans(5)”
The standard analyzer is the default analyzer of Elasticsearch. Divide text in single words and remove most of punctuation.
“set, the, shape, to, semi, transparent, by, calling, set_trans, 5”
Elastic
[email protected] - CTO
Simple Analyzer
“Set the shape to semi-transparent by calling set_trans(5)”
The simple analyser removes all characters which are not letters and put the whole text lowercase
“set, the, shape, to, semi, transparent, by, calling, set, trans”
Elastic
[email protected] - CTO
Whitespace Analyzer
“Set the shape to semi-transparent by calling set_trans(5)”
The whitespace analyser will create token by white space and put text in lowercase
“Set, the, shape, to, semi, transparent, by, calling, set_trans(5)”
Elastic
[email protected] - CTO
Language Analyzer
“Set the shape to semi-transparent by calling set_trans(5)”
This analyser uses a language specific feature to remove stop words or to do stemming.
“set, shape, semi, transpar, call, set_tran, 5”
Elastic
[email protected] - CTO
Language Analyzer
arabic, armenian, basque, brazilian, bulgarian, catalan, chinese, cjk, czech, danish, dutch, english, finnish, french, galician, german, greek, hindi, hungarian, indonesian, irish, italian, latvian, norwegian, persian, portuguese, romanian, russian, sorani, spanish, swedish, turkish, thai.
Elastic
[email protected] - CTO
Pre-built Analyzers
Standard Analyzer Simple Analyzer
Whitespace Analyzer Stop Analyzer
Keyword Analyzer Pattern Analyzer
Language Analyzers Snowball Analyzer Custom Analyzer
Elastic
[email protected] - CTO
Tokenizer
Standard Tokenizer Edge NGram Tokenizer
Keyword Tokenizer Letter Tokenizer
Lowercase Tokenizer NGram Tokenizer
Whitespace Tokenizer Pattern Tokenizer
UAX Email URL Tokenizer Path Hierarchy Tokenizer
Elastic
[email protected] - CTO
Token Filters
Standard Token Filter ASCII Folding Token Filter
Length Token Filter Lowercase Token Filter
NGram Token Filter Edge NGram Token Filter Porter Stem Token Filter
Shingle Token Filter Stop Token Filter
… more than 32 Filters
References• Elasticsearch : The Definitive Guide• https://en.wikipedia.org/wiki/Full_text_search• https://www.elastic.co/guide/en/elasticsearch/guide/current/
hardware.html• https://www.elastic.co/guide/en/elasticsearch/guide/current/
heap-sizing.html• https://mtalavera.wordpress.com/2015/02/16/monitoring-with-
collectd-and-kibana/• Fuzzy search : https://www.found.no/foundation/fuzzy-search/• Phonetic-plugin : https://github.com/elastic/elasticsearch-
analysis-phonetic
[email protected] - CTO