lucene-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Per Steffensen (JIRA)" <j...@apache.org>
Subject [jira] [Updated] (SOLR-3383) Async responses in SolrJ
Date Wed, 09 May 2012 11:25:49 GMT

     [ https://issues.apache.org/jira/browse/SOLR-3383?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]

Per Steffensen updated SOLR-3383:
---------------------------------

    Attachment: SOLR-3383.patch

Early patch attached to show details about suggested implementation and to get feedback.

The patch should fit on top of your revision 1327417. The branch for doing the code for this
issue, was created from the branch where we did the SOLR-3173, SOLR-3178 and SOLR-3382 changes,
so it contains parts of an old fix for those. Some of those parts have been reverted or changed
in the latest patch attached to SOLR-3178, so they are basically outdated. Hope you are able
to distill what is related to SOLR-3383 and what is not.

Basically the idea is that all public methods for doing requests using SolrJ SolrServer's
or SolrRequest-subclasses, has been changed from returning responses (in the general form
of NamedList<Object> or in the specialized form of SolrResponse-subclasses) into returning
Futures of responses.

Exceptions originating from the server, that you used to get when you called one of those
public methods, you will now not get before you call get on the returned Future.

The benefits are that you (if your chain of SolrServers contains a real async implementation)
as a client, can go do other stuff from the time you send the request (by calling one of those
public methods) and until the time you inspect the response (fetched by calling get on the
returned Future). Besides that we will be able to get responses back from ConcurrentSolrServer.

As is right now you need to have ConcurrentSolrServer in your chain of SolrServers in order
to get real async behaviour, because the rest of the SolrServers does not really support it.
If you dont use ConcurrentSolrServer your client thread will basically not get its Future
back from a call to one of those public methods mentioned at the top, before the response
has been received from the server and calling get on the Future will immediately give the
client the reponse anyway. So not much acheived. But that can all easily be changed by modifying
the HttpSolrServer to be able to run in a async mode where it has its own threadpool in which
it does requests. That can easily be coded, and will to a big extend eliminate the need for
a ConcurrentSolrServer to begin with. We will make this feature of the HttpSolrServer, and
attach it if you like the idea represented by this issue SOLR-3383.

The patch is a little bit dependent on the patch attached to SOLR-3178, in the way that it
deals with e.g. PartialErrors in places where it needs to, but we can deal with that later.

This is just to present the idea in code and give you something to look at and comment on.

Regards, Per Steffensen
                
> Async responses in SolrJ
> ------------------------
>
>                 Key: SOLR-3383
>                 URL: https://issues.apache.org/jira/browse/SOLR-3383
>             Project: Solr
>          Issue Type: Improvement
>          Components: clients - java
>    Affects Versions: 3.5
>            Reporter: Per Steffensen
>            Assignee: Per Steffensen
>              Labels: async, asynchronous, concurrency, query, solrj, update
>             Fix For: 4.0
>
>         Attachments: SOLR-3383.patch
>
>
> Today it is like this
> - SolrServer.request returns NamedList<Object>
> - SolrRequest.process returns SolrResponse
> - Public methods on SolrServer like addX, optimize, commit, queryX etc. returns subclasses
of SolrResponse (e.g. "add" returns UpdateResponse)
> - etc
> This is all synchronous - that is, the calling thread of those methods will wait for
the response before being able to continue. I believe the industry today agrees that "operations"
like client-server network-requireing operations should be done asynchronously seens from
the client API. Therefore basically we should change those methods
> - SolrServer.request returns Future<NamedList<Object>>
> - SolrRequest.process returns Future<SolrResponse>
> - SolrServer.xxx returns Future<subclass of SolrResponse>
> and make the appropriate changes in the implementations below.
> My main argument for this right now, is that ConcurrentUpdateSolrServer really is not
able to hand over responses to the calling client. Guess that it is also the reason why it
is only a "Update"-SolrServer and not a complete SolrServer (being able to do queries etc.)
- updates does not require sending responses (except primitive errors) back to the client,
queries etc does. Now that we do "finegrained error propagation" (SOLR-3382) in order to send
"unique key constraint"- and "versioning"-errors (SOLR-3173 and SOLR-3178) back to the client
in responses to update-request, suddenly it is not true anymore that updates does not require
sending responses back to the client.
> Making the changes suggested above (returning Futures) would
> - Allow ConcurrentUpdateSolrServer to be used for updates potentially resulting in "unique
key constraint"- and "versioning"-errors
> - Allow ConcurrentUpdateSolrServer to become ConcurrentSolrServer - also being able to
do queries etc
> - Do cool stuff like SOLR-3384
> - Make SolrJ more modern with respect to asynchronous principles

--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators: https://issues.apache.org/jira/secure/ContactAdministrators!default.jspa
For more information on JIRA, see: http://www.atlassian.com/software/jira

        

---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@lucene.apache.org
For additional commands, e-mail: dev-help@lucene.apache.org


Mime
View raw message