Aman Sinha created DRILL-4846:
-
Summary: Eliminate extra operations during metadata cache pruning
Key: DRILL-4846
URL: https://issues.apache.org/jira/browse/DRILL-4846
Project: Apache Drill
Matt Keranen created DRILL-4845:
---
Summary: Malformed CSV throws IllegalArgumentException
Key: DRILL-4845
URL: https://issues.apache.org/jira/browse/DRILL-4845
Project: Apache Drill
Issue Type:
P.S. I changed NullableValueVectors.java to NOT expose an unsafe public version
of copyFrom, and instead have the public copyFrom call copyFromSafe. The query
with a where clause is now working for a simple case. The unexpected cast
behavior (to FLOAT8) is still an open issue.
-Original
Thanks Jinfeng, you nailed it. I added some incorrect code in
GetSetVectorHelper.java for VARDECIMAL
Regarding the Cast to FLOAT8, I also don't think it's correct. It should cast
to Int or BigInt. I found lots of pretty cryptic (to me) logic related to
casting rules, precedence, and other
Have you looked at GetSetVectorHelper [1]?
Are you trying to model decimal as something similar to VARCHAR, since
you use VARDECMIAL? If that's the case, you need check the code
corresponding to VARCHAR, which has both "data" and "offset".
Also, from your generated code, you are cast (100) and
Jinfeng,
I worked around the problem with the protected copyFrom method by renaming it,
basically, without figuring out why the wrong call was being made. The copier
generated code now compiles at runtime.
Now, the filtering code is being generated incorrectly. A sample is shown
below
Rahul Challapalli created DRILL-4844:
Summary: Partition pruning not working on a view which is created
using a '*' query
Key: DRILL-4844
URL: https://issues.apache.org/jira/browse/DRILL-4844
That sinks the RC0.
Once this performance issue is fixed, I'll send out RC1 for vote.
Thanks to everyone who voted on RC0!
Jinfeng
On Fri, Aug 12, 2016 at 1:10 PM, Aman Sinha wrote:
> I am changing my vote to a -1 based on some performance results that
> Dechang has
I am changing my vote to a -1 based on some performance results that
Dechang has obtained for metadata caching.
These are new tests with about 50K files spread over 5K directories that
are showing regression when used with certain types of views containing
SELECT *.
I have asked DC to file a
Just some confirmation required on a minor issue - I get the following for
version (see blank fields). Is this expected?
> select * from sys.version;
*+--++-+--+--+-+*
*| **version ** | **commit_id ** | **commit_message ** |
+1
Built from source zip file with unit test all passed on Linux
Randomly verified DRILL-4825
On Fri, Aug 12, 2016 at 3:52 AM, Arina Yelchiyeva <
arina.yelchiy...@gmail.com> wrote:
> +1 (non-binding)
> On Windows built from source, ran in embedded mode.
> Installed on cluster.
> Checked some
Jinfeng,
Actually, it turned out that the error message was because some of the hooks in
the code to setup the Casting mechanism were missing for VARDECIMAL. Once I
added these (in a number of places), I'm now encountering a weird problem as
below signature. It's using the copyFrom method,
+1 (non-binding)
On Windows built from source, ran in embedded mode.
Installed on cluster.
Checked some new features and run some basic queries.
On Fri, Aug 12, 2016 at 2:48 AM, Padma Penumarthy
wrote:
> +1
>
> Installed on the cluster and ran some basic queries
>
+1
Installed on the cluster and ran some basic queries
Tried few queries in embedded mode on the mac
Thanks,
Padma
> On Aug 10, 2016, at 10:58 PM, Aman Sinha wrote:
>
> +1 (binding)
> Downloaded source as well as binary tarballs.
> Checked git.properties and README.
>
On Drill's Team page[1] it says “It currently includes dozens of contributors
employed by many organizations”. It doesn’t make any claim about number of
*committers*, only *contributors*. I think you mis-read the page.
I count 99 distinct email addresses and 88 distinct names in the commit log,
Micheal,
I am having trouble locating the source of this assertion.
The closest to what you say is on Drill's team web page[1] which states,
and I quote "It currently includes dozens of *contributors*...".
The word is "contributors" and not "committers" and as per Github's
stat[2], the project
16 matches
Mail list logo