drill-user mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Ted Dunning <ted.dunn...@gmail.com>
Subject Re: New Drill R "dplyr" interface
Date Mon, 26 Dec 2016 20:34:29 GMT
On Mon, Dec 26, 2016 at 7:22 AM, Bob Rudis <bob@rud.is> wrote:

> I guess I read more into "ANSI SQL" than I should have.
>

Or not.  Could be a problem with Drill, too.


> While it does
> handle some complex SQL generated by dplyr (which you can see at
> https://github.com/hrbrmstr/sergeant
>

Btw... this is really pretty danged awesome.



> if you scroll down) there are
> instances when if, say you try to filter and left join two different
> tables it generates valid SQL that Drill dislikes. You can see the
> dplyr that fails here: http://rpubs.com/hrbrmstr/drillex
>
> dplyr generates the following SQL:
>
>     SELECT *
>     FROM ( SELECT *
>            FROM ( SELECT *
>                   FROM  dfs.tmp.`/in/c.parquet`
>                   WHERE ( cyl  = 6.0 ) ) xpintqtlzw
>            LEFT JOIN dfs.tmp.`/in/b.json`
>            USING ( car ) ) fyxhnlrrca
>     LIMIT 10
>
> The error message (if one doesn't want to go to the rpubs URL) is:
>
>     SYSTEM ERROR: DrillRuntimeException: Join only supports implicit
> casts between
>     1. Numeric data
>     2. Varchar, Varbinary data
>     3. Date, Timestamp data Left type: INT, Right type: VARCHAR.
>     Add explicit casts to avoid this error
>

I haven't been able to dig into this, but this looks pretty straightforward
to replicate.

Mime
  • Unnamed multipart/alternative (inline, None, 0 bytes)
View raw message