mirror of
https://github.com/honeymoose/OpenSearch.git
synced 2025-03-09 14:34:43 +00:00
This change optimizes the merge of terms aggregations by removing the priority queue that was used to collect all the buckets during a non-final reduction. We don't need to keep the result sorted since the merge of buckets in a subsequent reduce can modify the order. I wrote a small micro-benchmark to test the change and the speed ups are significative for small merge buffer sizes: ```` ########## Master: Benchmark (bufferSize) (cardinality) (numShards) (topNSize) Mode Cnt Score Error Units TermsReduceBenchmark.reduceTopHits 5 10000 1000 1000 avgt 10 2459,690 ± 198,682 ms/op TermsReduceBenchmark.reduceTopHits 16 10000 1000 1000 avgt 10 1030,620 ± 91,544 ms/op TermsReduceBenchmark.reduceTopHits 32 10000 1000 1000 avgt 10 558,608 ± 44,915 ms/op TermsReduceBenchmark.reduceTopHits 128 10000 1000 1000 avgt 10 287,333 ± 8,342 ms/op TermsReduceBenchmark.reduceTopHits 512 10000 1000 1000 avgt 10 257,325 ± 54,515 ms/op ########## Patch: Benchmark (bufferSize) (cardinality) (numShards) (topNSize) Mode Cnt Score Error Units TermsReduceBenchmark.reduceTopHits 5 10000 1000 1000 avgt 10 805,611 ± 14,630 ms/op TermsReduceBenchmark.reduceTopHits 16 10000 1000 1000 avgt 10 378,851 ± 17,929 ms/op TermsReduceBenchmark.reduceTopHits 32 10000 1000 1000 avgt 10 261,094 ± 10,176 ms/op TermsReduceBenchmark.reduceTopHits 128 10000 1000 1000 avgt 10 241,051 ± 19,558 ms/op TermsReduceBenchmark.reduceTopHits 512 10000 1000 1000 avgt 10 231,643 ± 6,170 ms/op ```` The code for the benchmark can be found [here](). It seems to be up to 3x faster for terms aggregations that return 10,000 unique terms (1000 terms per shard). For a cardinality of 100,000 terms, this patch is up to 5x faster: ```` ########## Patch: Benchmark (bufferSize) (cardinality) (numShards) (topNSize) Mode Cnt Score Error Units TermsReduceBenchmark.reduceTopHits 5 100000 1000 1000 avgt 10 12791,083 ± 397,128 ms/op TermsReduceBenchmark.reduceTopHits 16 100000 1000 1000 avgt 10 3974,939 ± 324,617 ms/op TermsReduceBenchmark.reduceTopHits 32 100000 1000 1000 avgt 10 2186,285 ± 267,124 ms/op TermsReduceBenchmark.reduceTopHits 128 100000 1000 1000 avgt 10 914,657 ± 160,784 ms/op TermsReduceBenchmark.reduceTopHits 512 100000 1000 1000 avgt 10 604,198 ± 145,457 ms/op ########## Master: Benchmark (bufferSize) (cardinality) (numShards) (topNSize) Mode Cnt Score Error Units TermsReduceBenchmark.reduceTopHits 5 100000 1000 1000 avgt 10 60696,107 ± 929,944 ms/op TermsReduceBenchmark.reduceTopHits 16 100000 1000 1000 avgt 10 16292,894 ± 783,398 ms/op TermsReduceBenchmark.reduceTopHits 32 100000 1000 1000 avgt 10 7705,444 ± 77,588 ms/op TermsReduceBenchmark.reduceTopHits 128 100000 1000 1000 avgt 10 2156,685 ± 88,795 ms/op TermsReduceBenchmark.reduceTopHits 512 100000 1000 1000 avgt 10 760,273 ± 53,738 ms/op ```` The merge of buckets can also be optimized. Currently we use an hash map to merge buckets coming from different shards so this can be costly if the number of unique terms is high. Instead, we could always sort the shard terms result by key and perform a merge sort to reduce the results. This would save memory and make the merge more linear in terms of complexity in the coordinating node at the expense of an additional sort in the shards. I plan to test this possible optimization in a follow up. Relates #51857
= Elasticsearch == A Distributed RESTful Search Engine === https://www.elastic.co/products/elasticsearch[https://www.elastic.co/products/elasticsearch] Elasticsearch is a distributed RESTful search engine built for the cloud. Features include: * Distributed and Highly Available Search Engine. ** Each index is fully sharded with a configurable number of shards. ** Each shard can have one or more replicas. ** Read / Search operations performed on any of the replica shards. * Multi Tenant. ** Support for more than one index. ** Index level configuration (number of shards, index storage, ...). * Various set of APIs ** HTTP RESTful API ** All APIs perform automatic node operation rerouting. * Document oriented ** No need for upfront schema definition. ** Schema can be defined for customization of the indexing process. * Reliable, Asynchronous Write Behind for long term persistency. * (Near) Real Time Search. * Built on top of Apache Lucene ** Each shard is a fully functional Lucene index ** All the power of Lucene easily exposed through simple configuration / plugins. * Per operation consistency ** Single document level operations are atomic, consistent, isolated and durable. == Getting Started First of all, DON'T PANIC. It will take 5 minutes to get the gist of what Elasticsearch is all about. === Installation * https://www.elastic.co/downloads/elasticsearch[Download] and unpack the Elasticsearch official distribution. * Run `bin/elasticsearch` on Linux or macOS. Run `bin\elasticsearch.bat` on Windows. * Run `curl -X GET http://localhost:9200/`. * Start more servers ... === Indexing Let's try and index some twitter like information. First, let's index some tweets (the `twitter` index will be created automatically): ---- curl -XPUT 'http://localhost:9200/twitter/_doc/1?pretty' -H 'Content-Type: application/json' -d ' { "user": "kimchy", "post_date": "2009-11-15T13:12:00", "message": "Trying out Elasticsearch, so far so good?" }' curl -XPUT 'http://localhost:9200/twitter/_doc/2?pretty' -H 'Content-Type: application/json' -d ' { "user": "kimchy", "post_date": "2009-11-15T14:12:12", "message": "Another tweet, will it be indexed?" }' curl -XPUT 'http://localhost:9200/twitter/_doc/3?pretty' -H 'Content-Type: application/json' -d ' { "user": "elastic", "post_date": "2010-01-15T01:46:38", "message": "Building the site, should be kewl" }' ---- Now, let's see if the information was added by GETting it: ---- curl -XGET 'http://localhost:9200/twitter/_doc/1?pretty=true' curl -XGET 'http://localhost:9200/twitter/_doc/2?pretty=true' curl -XGET 'http://localhost:9200/twitter/_doc/3?pretty=true' ---- === Searching Mmm search..., shouldn't it be elastic? Let's find all the tweets that `kimchy` posted: ---- curl -XGET 'http://localhost:9200/twitter/_search?q=user:kimchy&pretty=true' ---- We can also use the JSON query language Elasticsearch provides instead of a query string: ---- curl -XGET 'http://localhost:9200/twitter/_search?pretty=true' -H 'Content-Type: application/json' -d ' { "query" : { "match" : { "user": "kimchy" } } }' ---- Just for kicks, let's get all the documents stored (we should see the tweet from `elastic` as well): ---- curl -XGET 'http://localhost:9200/twitter/_search?pretty=true' -H 'Content-Type: application/json' -d ' { "query" : { "match_all" : {} } }' ---- We can also do range search (the `post_date` was automatically identified as date) ---- curl -XGET 'http://localhost:9200/twitter/_search?pretty=true' -H 'Content-Type: application/json' -d ' { "query" : { "range" : { "post_date" : { "from" : "2009-11-15T13:00:00", "to" : "2009-11-15T14:00:00" } } } }' ---- There are many more options to perform search, after all, it's a search product no? All the familiar Lucene queries are available through the JSON query language, or through the query parser. === Multi Tenant - Indices Man, that twitter index might get big (in this case, index size == valuation). Let's see if we can structure our twitter system a bit differently in order to support such large amounts of data. Elasticsearch supports multiple indices. In the previous example we used an index called `twitter` that stored tweets for every user. Another way to define our simple twitter system is to have a different index per user (note, though that each index has an overhead). Here is the indexing curl's in this case: ---- curl -XPUT 'http://localhost:9200/kimchy/_doc/1?pretty' -H 'Content-Type: application/json' -d ' { "user": "kimchy", "post_date": "2009-11-15T13:12:00", "message": "Trying out Elasticsearch, so far so good?" }' curl -XPUT 'http://localhost:9200/kimchy/_doc/2?pretty' -H 'Content-Type: application/json' -d ' { "user": "kimchy", "post_date": "2009-11-15T14:12:12", "message": "Another tweet, will it be indexed?" }' ---- The above will index information into the `kimchy` index. Each user will get their own special index. Complete control on the index level is allowed. As an example, in the above case, we might want to change from the default 1 shards with 1 replica per index, to 2 shards with 1 replica per index (because this user tweets a lot). Here is how this can be done (the configuration can be in yaml as well): ---- curl -XPUT http://localhost:9200/another_user?pretty -H 'Content-Type: application/json' -d ' { "settings" : { "index.number_of_shards" : 2, "index.number_of_replicas" : 1 } }' ---- Search (and similar operations) are multi index aware. This means that we can easily search on more than one index (twitter user), for example: ---- curl -XGET 'http://localhost:9200/kimchy,another_user/_search?pretty=true' -H 'Content-Type: application/json' -d ' { "query" : { "match_all" : {} } }' ---- Or on all the indices: ---- curl -XGET 'http://localhost:9200/_search?pretty=true' -H 'Content-Type: application/json' -d ' { "query" : { "match_all" : {} } }' ---- And the cool part about that? You can easily search on multiple twitter users (indices), with different boost levels per user (index), making social search so much simpler (results from my friends rank higher than results from friends of my friends). === Distributed, Highly Available Let's face it, things will fail.... Elasticsearch is a highly available and distributed search engine. Each index is broken down into shards, and each shard can have one or more replicas. By default, an index is created with 1 shard and 1 replica per shard (1/1). There are many topologies that can be used, including 1/10 (improve search performance), or 20/1 (improve indexing performance, with search executed in a map reduce fashion across shards). In order to play with the distributed nature of Elasticsearch, simply bring more nodes up and shut down nodes. The system will continue to serve requests (make sure you use the correct http port) with the latest data indexed. === Where to go from here? We have just covered a very small portion of what Elasticsearch is all about. For more information, please refer to the https://www.elastic.co/products/elasticsearch[elastic.co] website. General questions can be asked on the https://discuss.elastic.co[Elastic Forum] or https://ela.st/slack[on Slack]. The Elasticsearch GitHub repository is reserved for bug reports and feature requests only. === Building from Source Elasticsearch uses https://gradle.org[Gradle] for its build system. In order to create a distribution, simply run the `./gradlew assemble` command in the cloned directory. The distribution for each project will be created under the `build/distributions` directory in that project. See the xref:TESTING.asciidoc[TESTING] for more information about running the Elasticsearch test suite. === Upgrading from older Elasticsearch versions In order to ensure a smooth upgrade process from earlier versions of Elasticsearch, please see our https://www.elastic.co/guide/en/elasticsearch/reference/current/setup-upgrade.html[upgrade documentation] for more details on the upgrade process.
Description
Languages
Java
99.5%
Groovy
0.4%