Different index per user

[Erdinc Yilmazel <er...@yilmazel.com>]

I am building an online application where I want to provide search
functionality to users and each user is to search only within his own data.
Can you give me some ideas about the structure of the index that I should
have in this scenario? I know I could use a different index per user account
but that is simply not practical. I could also create a single index and use
a separate field to identify the documents that are related to a user
account and create my queries accordingly. However I am guessing that there
might be better ways to partition my index based on the user identifiers.

Thanks,

Erdinc

Re: Different index per user

[Erick Erickson <er...@gmail.com>]

If you're thinking about sharding etc, have you looked at SOLR which
manages a lot of the hard stuff for you?

Best
Erick

On Mon, Apr 19, 2010 at 1:34 PM, Erdinc Yilmazel
<er...@gmail.com>wrote:

> Thanks Erick,
> I don't have any access control constraints. The index won't be exposed to
> the users. I am just concerned about scalability issues. I'll probably use
> adding a user identifier field approach first and I'll try sharding based
> on
> that value when my index grows.
>
> Erdinc
>
> On Mon, Apr 19, 2010 at 1:22 AM, Erick Erickson <erickerickson@gmail.com
> >wrote:
>
> > No really better ways that spring to mind. At root you have to somehow
> > know what document is added by each user, and adding a field to
> > the document with this info is quite reasonable, then just tack
> > on an +user:<current user> to each query.
> >
> > You can get creative with Lucene Filters. That is, create a Filter for
> > a user when s/he logs on pass that to queries, see some of the
> > Searcher.search methods that take a Filter. But there's no
> > real reason to go there unless you have performance problems.
> >
> > This is the fairly simple case, you're not indicating that you're after
> > some generic Access Control List functionality so I'd stick with
> > the simple approach until proven inadequate...
> >
> > HTH
> > Erick
> >
> > On Sun, Apr 18, 2010 at 5:37 PM, Erdinc Yilmazel <erdinc@yilmazel.com
> > >wrote:
> >
> > > I am building an online application where I want to provide search
> > > functionality to users and each user is to search only within his own
> > data.
> > > Can you give me some ideas about the structure of the index that I
> should
> > > have in this scenario? I know I could use a different index per user
> > > account
> > > but that is simply not practical. I could also create a single index
> and
> > > use
> > > a separate field to identify the documents that are related to a user
> > > account and create my queries accordingly. However I am guessing that
> > there
> > > might be better ways to partition my index based on the user
> identifiers.
> > >
> > > Thanks,
> > >
> > > Erdinc
> > >
> >
>

Re: Different index per user

[Erdinc Yilmazel <er...@gmail.com>]

Thanks Erick,
I don't have any access control constraints. The index won't be exposed to
the users. I am just concerned about scalability issues. I'll probably use
adding a user identifier field approach first and I'll try sharding based on
that value when my index grows.

Erdinc

On Mon, Apr 19, 2010 at 1:22 AM, Erick Erickson <er...@gmail.com>wrote:

> No really better ways that spring to mind. At root you have to somehow
> know what document is added by each user, and adding a field to
> the document with this info is quite reasonable, then just tack
> on an +user:<current user> to each query.
>
> You can get creative with Lucene Filters. That is, create a Filter for
> a user when s/he logs on pass that to queries, see some of the
> Searcher.search methods that take a Filter. But there's no
> real reason to go there unless you have performance problems.
>
> This is the fairly simple case, you're not indicating that you're after
> some generic Access Control List functionality so I'd stick with
> the simple approach until proven inadequate...
>
> HTH
> Erick
>
> On Sun, Apr 18, 2010 at 5:37 PM, Erdinc Yilmazel <erdinc@yilmazel.com
> >wrote:
>
> > I am building an online application where I want to provide search
> > functionality to users and each user is to search only within his own
> data.
> > Can you give me some ideas about the structure of the index that I should
> > have in this scenario? I know I could use a different index per user
> > account
> > but that is simply not practical. I could also create a single index and
> > use
> > a separate field to identify the documents that are related to a user
> > account and create my queries accordingly. However I am guessing that
> there
> > might be better ways to partition my index based on the user identifiers.
> >
> > Thanks,
> >
> > Erdinc
> >
>

Re: Different index per user

[Erick Erickson <er...@gmail.com>]

No really better ways that spring to mind. At root you have to somehow
know what document is added by each user, and adding a field to
the document with this info is quite reasonable, then just tack
on an +user:<current user> to each query.

You can get creative with Lucene Filters. That is, create a Filter for
a user when s/he logs on pass that to queries, see some of the
Searcher.search methods that take a Filter. But there's no
real reason to go there unless you have performance problems.

This is the fairly simple case, you're not indicating that you're after
some generic Access Control List functionality so I'd stick with
the simple approach until proven inadequate...

HTH
Erick

On Sun, Apr 18, 2010 at 5:37 PM, Erdinc Yilmazel <er...@yilmazel.com>wrote:

> I am building an online application where I want to provide search
> functionality to users and each user is to search only within his own data.
> Can you give me some ideas about the structure of the index that I should
> have in this scenario? I know I could use a different index per user
> account
> but that is simply not practical. I could also create a single index and
> use
> a separate field to identify the documents that are related to a user
> account and create my queries accordingly. However I am guessing that there
> might be better ways to partition my index based on the user identifiers.
>
> Thanks,
>
> Erdinc
>