Hi Erick,

I don't think all Solr queries support indexed=false and docValues=true.
For example SOLR-11190 added such support to graph queries recently.

I added this to the ref guide under the graph query parser section ( I
notice a typo there which i"ll fix ):

+The supported fieldTypes are point fields with docValues enabled or string
fields with indexed=true or docValues=true.
+For string fields which are indexed=false and docValues=true please refer
to the javadocs for `DocValuesTermsQuery`
+for it's performance characteristics so indexed=true will perform better
for most use-cases.

Maybe your message sends out the message better?

On Tue, Aug 29, 2017 at 2:08 AM, Erick Erickson <[email protected]>
wrote:

> bq: What do you mean by 'be in the JVM'?
>
> I wasn't sure if a more efficient searching structure would be built
> in the JVM or not, building an inverted structure out of docValues
> there. But you're saying not, it's a linear scan of the uninverted
> structure out in the OS's memory.
>
> It would have been quite ironic if we started seeing a message like
> "inverting docValues field for searching" in the logs. Symmetrical to
> the background for docValues I'll admit... ;)
>
> Thanks for the confirmation.
>
> That leaves whether this is reasonable behavior or not. It feels like
> a documentation issue, something like
>
> 'While searching on fields having docValues="true", indexed="false" is
> possible, it is orders of magnitude slower than searching on fields
> with indexed="true". We _strongly_ recommend that any field that is
> used for searching be configured with indexed="true" '
>
> That's assuming that just dis-allowing searching on dv=true,
> indexed=false fields is not an option.
>
> WDYT?
>
> Erick
>
> On Mon, Aug 28, 2017 at 12:55 PM, Adrien Grand <[email protected]> wrote:
> > Indeed this will be a linear scan if it is not intersected with a
> selective
> > query, which is quite trappy.
> >
> > What do you mean by 'be in the JVM'?
> >
> > Le lun. 28 août 2017 à 21:49, Erick Erickson <[email protected]> a
> > écrit :
> >>
> >> You can search on fields with DV=true and indexed=false. But IIUC this
> >> is a "table scan". Does it really make sense to support this?
> >>
> >> NOTE: Haven't checked the code, but even if we build an efficient
> >> structure, it would be in the JVM, correct?
> >>
> >> ---------------------------------------------------------------------
> >> To unsubscribe, e-mail: [email protected]
> >> For additional commands, e-mail: [email protected]
> >>
> >
>
> ---------------------------------------------------------------------
> To unsubscribe, e-mail: [email protected]
> For additional commands, e-mail: [email protected]
>
>

Reply via email to