Foreword
The company ELK system needs reconstruction, then I am going to try to upgrade Es6.x to Es7.x, turned over many documents, stepped on a lot of pits, the following are some of the new features interpretation (theory), mainly from the ES6.x- "7.x.
The next course will be introduced ES7 + Java coding practices (expectations).
text
1. Cluster connection change: TransportClient discarded
So, java codes es7, use only restclient. Then, the individual integrated a little, for java programming, we recommend using High-level-rest-client operate in a manner ES cluster
2. ES data storage structure changes: In addition to Type
es6, the official mentions the es7 will delete type, and es6 time has been specified for each index can have only one type. _Doc as the default type in es7, the official said, completely removed type in the 8.x versions.
api send request mode changes, such as to obtain a document ID for an index: GET index/_doc/idwhere index id is a specific value and
3. High-level REST client change
Header has been removed accept the API method parameters; Cluster Health API default cluster level;
4. ES package default package jdk: 7.x version of the program that abrupt edge packet size 300MB +
Contrast 6.x found a large bag 200MB +, it is the size of JDK
5. The default configuration changes: The default node name as a host name, the default number of fragments to 1, 5 is no longer.
6. relevant of speed optimization: Weak-AND algorithm
What is weak-and algorithm?
The core principle: Take TOP N result set, estimate the number of hits recorded.
In short, in general we calculate the correlation of the text when the inverted index will be carried out by way of inquiry, by the inverted index than the full amount has to traverse save a lot of time, but sometimes still very slow.
The reason is that many times we really just want to top n results, some of the results were also significantly poorer complex correlation calculation,
and weak-and correlation algorithm to estimate the upper limit of the document by calculating the upper limit of the contribution of each word, thereby establishing a threshold value results in inverted Save branches, to thereby obtain the effect of speed.
7. Query interval (Intervals queries):
Some search use cases (for example, legal and patent search) introduced the need to look up a word or phrase from one another at a distance of record.
Interval of 7.0 Elasticsearch query introduces a new way of constructing such queries, compared with the previous method (span query span queries), the use and definition easier.
Compared with the span query, the query interval adaptability to the edge of the case stronger.
8. The introduction of a new sub-cluster coordination
Remove minimum_master_nodes parameters, let Elasticsearch choose node can form arbitration.
A typical primary node election is now only a short time to complete.
Scalable cluster safer, easier, and may result in loss of data the system configuration options less.
Node more clearly document their status, help diagnose why they can not join a cluster or Why not elect a master node.
Nanosecond timestamps support, improve data accuracy
Bold Style
No memory overflow
New Circuit Breaker more robust monitoring levels before the JVM heap memory usage, Elasticsearch ratio.
Set defaults indices.breaker.fielddata.limit JVM heap size from 60% to 40% reduction.
reference
Elastic Stack 7.0.0 release blockbuster
Breaking-Changes-7.0
elasticsearch 7.0 officially released, disk him!
elasticsearch7.0 cluster configuration learning
elasticsearch- versions Characteristics