github-mirrors/elasticsearch
1
0
Fork 0
mirror of https://github.com/elastic/elasticsearch.git synced 2025-04-24 23:27:25 -04:00
Code Issues Projects Releases Packages Wiki Activity

Node selector per client rather than per request (#31471)

Browse source 
We have made node selectors configurable per request, but all 
of other language clients don't allow for that.
A good reason not to do so, is that having a different node selector 
per request breaks round-robin. This commit makes NodeSelector 
configurable only at client initialization. It also improves the docs 
on this matter, important given that a single node selector can still 
affect round-robin.
This commit is contained in:
Luca Cavanna 2018-06-22 17:15:29 +02:00 • committed by GitHub
parent 59e7c6411a
commit 16e4e7a7cf
No known key found for this signature in database
GPG key ID: 4AEE18F83AFDEB23
20 changed files with 208 additions and 148 deletions
Download patch file Download diff file Expand all files Collapse all files

27
docs/java-rest/low-level/configuration.asciidoc
View file

@ -99,3 +99,30 @@ http://docs.oracle.com/javase/8/docs/technotes/guides/net/properties.html[`netwo
to your
http://docs.oracle.com/javase/8/docs/technotes/guides/security/PolicyFiles.html[Java
security policy].
=== Node selector
The client sends each request to one of the configured nodes in round-robin
fashion. Nodes can optionally be filtered through a node selector that needs
to be provided when initializing the client. This is useful when sniffing is
enabled, in case only dedicated master nodes should be hit by HTTP requests.
For each request the client will run the eventually configured node selector
to filter the node candidates, then select the next one in the list out of the
remaining ones.
["source","java",subs="attributes,callouts,macros"]
--------------------------------------------------
include-tagged::{doc-tests}/RestClientDocumentation.java[rest-client-init-allocation-aware-selector]
--------------------------------------------------
<1> Set an allocation aware node selector that allows to pick a node in the
local rack if any available, otherwise go to any other node in any rack. It
acts as a preference rather than a strict requirement, given that it goes to
another rack if none of the local nodes are available, rather than returning
no nodes in such case which would make the client forcibly revive a local node
whenever none of the nodes from the preferred rack is available.
WARNING: Node selectors that do not consistently select the same set of nodes
will make round-robin behaviour unpredictable and possibly unfair. The
preference example above is fine as it reasons about availability of nodes
which already affects the predictability of round-robin. Node selection should
not depend on other external factors or round-robin will not work properly.

22
docs/java-rest/low-level/usage.asciidoc
View file

@ -196,6 +196,16 @@ include-tagged::{doc-tests}/RestClientDocumentation.java[rest-client-init-failur
<1> Set a listener that gets notified every time a node fails, in case actions
need to be taken. Used internally when sniffing on failure is enabled.
["source","java",subs="attributes,callouts,macros"]
--------------------------------------------------
include-tagged::{doc-tests}/RestClientDocumentation.java[rest-client-init-node-selector]
--------------------------------------------------
<1> Set the node selector to be used to filter the nodes the client will send
requests to among the ones that are set to the client itself. This is useful
for instance to prevent sending requests to dedicated master nodes when
sniffing is enabled. By default the client sends requests to every configured
node.
["source","java",subs="attributes,callouts,macros"]
--------------------------------------------------
include-tagged::{doc-tests}/RestClientDocumentation.java[rest-client-init-request-config-callback]
@ -283,8 +293,7 @@ instance and share it between all requests:
include-tagged::{doc-tests}/RestClientDocumentation.java[rest-client-options-singleton]
--------------------------------------------------
<1> Add any headers needed by all requests.
<2> Set a `NodeSelector`.
<3> Customize the response consumer.
<2> Customize the response consumer.
`addHeader` is for headers that are required for authorization or to work with
a proxy in front of Elasticsearch. There is no need to set the `Content-Type`
@ -315,15 +324,6 @@ adds an extra header:
include-tagged::{doc-tests}/RestClientDocumentation.java[rest-client-options-customize-header]
--------------------------------------------------
Or you can send requests to nodes with a particular attribute:
["source","java",subs="attributes,callouts,macros"]
--------------------------------------------------
include-tagged::{doc-tests}/RestClientDocumentation.java[rest-client-options-customize-attribute]
--------------------------------------------------
<1> Replace the node selector with one that selects nodes on a particular rack.
==== Multiple parallel asynchronous actions
The client is quite happy to execute many actions in parallel. The following