Skip to content

update#2

Merged
MirrorChu merged 24 commits intoMirrorChu:masterfrom
trinodb:master
Apr 20, 2021
Merged

update#2
MirrorChu merged 24 commits intoMirrorChu:masterfrom
trinodb:master

Conversation

@MirrorChu
Copy link
Copy Markdown
Owner

No description provided.

ebyhr and others added 24 commits April 18, 2021 22:03
In case of FULL join having a scalar source producing no outputs,
do not replace the join with the other source if it cannot be
determined to have at least one row.

If the other source turns out to be empty, we would get empty
result after such replacement. This is incorrect. The single input row
from the scalar source should be preserved on output.
This is an enhancement of the section about IntelliJ IDEA's inspections.
Proper handling of language injection is unfortunately not automatic,
and there is no objective truth to how to do it. But since we're working
with SQL statements a lot, I find it useful.
This is a legacy method that is only used with table layouts.
Use a row level delete instead of dropping partitions when
running a delete that exactly matches entire partitions.
This commit fixes bug #7298, in which SQL UPDATE failed
with an exception when applied to a transactional table
originally created using CREATE TABLE AS SELECT...
CREATE TABLE AS SELECT... results in table files that
don't have the ACID transaction, rowId and bucket columns
even though the table was declared as transactional.

DELETE of original files worked because OrcFileWriterFactory
adds a different rowId column adaptation for original files
than for normal transactional delta files.  However, UPDATE
did not have that special column adaptation.  This commit adds
that rowId adaptation that works for UPDATE on original files.

This commit adds a several tests, two that verify that UPDATE with
original files now work correctly for both partitioned and
unpartitioned tables, and another to show that DELETE already
worked correctly with original files.

This commit also adds a method to verify that the data file
paths for original files conform to the formats genenerated
by Hive and Trino, and uses that method in all the tests
of original files.
Replace List<Integer> and streams with IntList.
Simple benchmarks show roughly 2.5x-5x performance gain.
@MirrorChu MirrorChu merged commit 77ac3d1 into MirrorChu:master Apr 20, 2021
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.