You are viewing a plain text version of this content. The canonical link for it is here.
Posted to dev@drill.apache.org by Jason Altekruse <al...@gmail.com> on 2016/01/05 18:56:27 UTC
Hangout starting in 5 minutes
Come join the Drill community in our weekly hangout meeting to find out
what is going on with Drill right now.
https://plus.google.com/hangouts/_/dremio.com/drillhangout
Some items I would like to discuss this week:
- 1.5 release, issues left to fix, when would we like to target for a vote
- Drill parquet date bug: https://issues.apache.org/jira/browse/DRILL-4203
Feel free to respond with other items you would like to have discussed, or
just jump on the call.
Re: Hangout starting in 5 minutes
Posted by Jason Altekruse <al...@gmail.com>.
Notes: Drill hangout - 1/5/2016
Vicky, Andries, Hakim, Aman, Julien, Jason, Charles
Drill 1.5 release thread, number of outstanding issues to solve
Parquet dates
- metadata migration may be needed for old files
- check migration tool to make sure it doesn't update already known
versions to a newer one
- can use some combination of a whitelist of known good writers as well
as checking the year
on some of the values and that the flag to auto-correct bad dates is
set
Allocator bugs
- CTAS with partitioning was failing, sort running out of memory
- Flatten was having issues
- Functional suite
- intermittent
- Unit test, jacques did not file a bug
- sent an e-mail to parth and Hakim
- advanced suite, out of memory failures
- See if Dremio infrastructure is running advanced tests
- checked with Jacques, Dremio is not currently running the
advanced suite
Amit's branch - testing by Vicky
- blocked on the sort bug
- his fix is on top of 1.5 which fails with out of memory in sort
before it reaches the merge join operator
Aman hash skew - Drill-4237
- sting of length 32 or more chars
- bad skew due to use of signed rather than unsigned long in the C
implementation
- revert to old hash functions
- performance of the hash is a little slower, need to measure how
much
- already tried using Guava UnsignedLong
- does not have required operations for the xxhash algorithm
- bit shifts
Hakim
- warming from parquet library
- parquet "corrupt statistics" message is showing with Drill 1.5
- issue with partitioned files
- partitioned by on date column seems to be the issue
On Tue, Jan 5, 2016 at 11:56 AM, Jason Altekruse <al...@gmail.com>
wrote:
> Come join the Drill community in our weekly hangout meeting to find out
> what is going on with Drill right now.
>
> https://plus.google.com/hangouts/_/dremio.com/drillhangout
>
> Some items I would like to discuss this week:
> - 1.5 release, issues left to fix, when would we like to target for a vote
> - Drill parquet date bug: https://issues.apache.org/jira/browse/DRILL-4203
>
> Feel free to respond with other items you would like to have discussed, or
> just jump on the call.
>
>
>
Re: Hangout starting in 5 minutes
Posted by Jason Altekruse <al...@gmail.com>.
Notes: Drill hangout - 1/5/2016
Vicky, Andries, Hakim, Aman, Julien, Jason, Charles
Drill 1.5 release thread, number of outstanding issues to solve
Parquet dates
- metadata migration may be needed for old files
- check migration tool to make sure it doesn't update already known
versions to a newer one
- can use some combination of a whitelist of known good writers as well
as checking the year
on some of the values and that the flag to auto-correct bad dates is
set
Allocator bugs
- CTAS with partitioning was failing, sort running out of memory
- Flatten was having issues
- Functional suite
- intermittent
- Unit test, jacques did not file a bug
- sent an e-mail to parth and Hakim
- advanced suite, out of memory failures
- See if Dremio infrastructure is running advanced tests
- checked with Jacques, Dremio is not currently running the
advanced suite
Amit's branch - testing by Vicky
- blocked on the sort bug
- his fix is on top of 1.5 which fails with out of memory in sort
before it reaches the merge join operator
Aman hash skew - Drill-4237
- sting of length 32 or more chars
- bad skew due to use of signed rather than unsigned long in the C
implementation
- revert to old hash functions
- performance of the hash is a little slower, need to measure how
much
- already tried using Guava UnsignedLong
- does not have required operations for the xxhash algorithm
- bit shifts
Hakim
- warming from parquet library
- parquet "corrupt statistics" message is showing with Drill 1.5
- issue with partitioned files
- partitioned by on date column seems to be the issue
On Tue, Jan 5, 2016 at 11:56 AM, Jason Altekruse <al...@gmail.com>
wrote:
> Come join the Drill community in our weekly hangout meeting to find out
> what is going on with Drill right now.
>
> https://plus.google.com/hangouts/_/dremio.com/drillhangout
>
> Some items I would like to discuss this week:
> - 1.5 release, issues left to fix, when would we like to target for a vote
> - Drill parquet date bug: https://issues.apache.org/jira/browse/DRILL-4203
>
> Feel free to respond with other items you would like to have discussed, or
> just jump on the call.
>
>
>