setop
/
sqlite-parquet-vtable
mirror of https://github.com/cldellow/sqlite-parquet-vtable.git
1
0
Fork 0
Code Releases Activity
121 Commits 2 Branches 0 Tags 570 KiB
Commit Graph

121 Commits

Author SHA1 Message Date
Colin Dellow 68ecfde4d9 travis: code coverage, maybe? 2018-07-05 18:47:33 -04:00
Colin Dellow e3c6bad9f5 Add status badge, run tests 
Fixes #16
Fixes #18
 2018-07-05 09:17:19 -04:00
Colin Dellow 83db07456e travis: revert sqlite hack 2018-07-05 09:13:37 -04:00
Colin Dellow efb22c5c5a travis: don't stomp CC in Makefile 2018-07-05 09:09:18 -04:00
Colin Dellow f87d823607 travis: maybe fix cmake gcc 2018-07-05 09:01:56 -04:00
Colin Dellow b4bf732eb1 travis: run cmake 
This ensures `parquet_version.h.in` gets procesed
 2018-07-05 08:55:46 -04:00
Colin Dellow aaa07fe629 travis: force sqlite to use gcc 2018-07-04 23:46:36 -04:00
Colin Dellow 16da95ca29 Revert "travis: ensure gcc" 
This reverts commit 9a0a19ea52.

...gcc is already on the box. Hmm.
 2018-07-04 23:45:01 -04:00
Colin Dellow 9a0a19ea52 travis: ensure gcc 2018-07-04 23:43:28 -04:00
Colin Dellow ea3cd39ae6 travis: --yes 2018-07-04 23:36:59 -04:00
Colin Dellow 9789a21865 travis: PPA for cmake 3.2 2018-07-04 23:35:30 -04:00
Colin Dellow b5de2799cd travis: run in sudo env 2018-07-04 23:25:21 -04:00
Colin Dellow 1f729e91b7 Add .travis.yml 
Progress towards #16
 2018-07-04 23:22:39 -04:00
Colin Dellow 100edb7015 Support prebuilt binaries 
...to make Travis CI builds reasonable.

Progress towards #16
 2018-07-04 23:17:57 -04:00
Colin Dellow 806d87c8e7 Derp, fix makefile 2018-07-04 22:36:43 -04:00
Colin Dellow 0aaff745f8 Stomp timestamp of arrow/pq directory 
Otherwise they'll be deemed to be newer than
the prebuilt libraries, and make will gallantly
kick off a build.
 2018-07-04 22:13:10 -04:00
Colin Dellow 9516955717 Separate targets for fetching/building arrow/pq 
This will allow us to use pre-built libs in Travis for
linking while still having the header files available for compilation.
 2018-07-04 22:01:13 -04:00
Colin Dellow b8df4b720b Provide `publish_libs` target 
...which publishes the newly-built supporting libs to my S3 bucket, for
future use by a Travis CI build.

Progress towards #16
 2018-07-04 21:45:59 -04:00
Colin Dellow 005d7a451f Add g++ as prerequisite 
Progress towards #16
 2018-07-04 21:30:51 -04:00
Colin Dellow 8084f14379 Add libicu for Ubuntu 14.04 
Progress towards #16
 2018-07-04 21:30:05 -04:00
Colin Dellow 7e961c4802 Skip tests for arrow 
Partial progress towards #16
 2018-07-04 20:50:47 -04:00
Colin Dellow ebb0eb7710 Add test case for #30 2018-07-04 19:59:55 -04:00
Colin Dellow 2e1ac92882 Revert "Add other random test case" 
This reverts commit 3bdc6f7078.
 2018-07-04 19:49:42 -04:00
Colin Dellow 3bdc6f7078 Add other random test case 2018-07-04 19:48:36 -04:00
Colin Dellow 33f8dbe4f4 Add test case for #26 2018-07-04 19:45:08 -04:00
Colin Dellow 5b26a78c1f Improve random query generation 
...throw in the occasional `NOT`
 2018-07-04 19:14:35 -04:00
Colin Dellow 373616ad1e Don't try to optimize IsNot 
Doesn't handle NULLs correctly, will open separate ticket
for it. Fixes the IS NOT case of #26
 2018-07-04 18:59:16 -04:00
Colin Dellow 0aa98ae1a5 Skip shared parquet/arrow libs, fix icu versions 2018-06-27 23:20:33 -04:00
Colin Dellow 2167d102b4 Add `make-linux-pgo` 
fixes #23, with perhaps some open questions about why PGO on
arrow/parquet-cpp regressed things.
 2018-06-27 22:23:22 -04:00
Colin Dellow 1a4f540e18 Stub PGO code in 
Incremental progress on #23 - should probably add a dedicated flag that
creates the instrumented binary, runs a test suite, then creates the
optimized binary.
 2018-06-26 23:50:11 -04:00
Colin Dellow 1d0d4c08b8 Build sqlite in parallel 2018-06-26 23:05:30 -04:00
Colin Dellow 76fb058dd7 Link Boost statically 
Fixes #15
 2018-06-26 22:44:50 -04:00
Colin Dellow 263a6af7ec Use Arrow's compression libraries 
Fixes #27
 2018-06-26 08:17:18 -04:00
Colin Dellow 129ff4e694
Merge pull request #25 from evansd/makefile-fixes 
Makefile fixes
 2018-06-25 13:54:29 -04:00
David Evans b7da04433b Include header locations we need 2018-06-25 18:20:24 +01:00
David Evans ab87b13b75 Reverse prereqs order to get build to work 2018-06-25 18:20:04 +01:00
Colin Dellow bc0be71546 Add brotli/snappy/gzip test files 
`test/test-supported` verifies they can be opened
 2018-06-25 08:32:36 -04:00
Colin Dellow 0bdcc9895e All-in-one build command 
`./make-linux` clones and builds:

- arrow
- brotli
- lz4
- parquet
- snappy
- zlib
- zstd
- this project

as a statically linked binary. Two Boost libs are still pulled in as
shared libs, should probably fix that, too, for ultimate portability.
 2018-06-24 21:11:07 -04:00
Colin Dellow ec6e970bbc Fix `order by rowid` to apply w/o clause 
Fixes #12, first screen of datasette is fast now
 2018-06-24 15:20:06 -04:00
Colin Dellow 5b59ba02fe Make ORDER BY ROWID fast 
Fixes #11
 2018-06-24 15:07:27 -04:00
Colin Dellow b774973852 Avoid row filter check when no constraints 
The function call overhead is expensive!

This makes count(*) on the census data 175ms instead
of 225ms, while not significantly impacting other use cases.
 2018-06-24 14:51:54 -04:00
Colin Dellow 84a22e6e77 link to blog 2018-06-24 11:39:44 -04:00
Colin Dellow 16cdd70f2b Short-circuit row group evaluation 
We can avoid eagerly computing bitmasks for other constraints this way.

Possible future work - order the constraints such that we evaluate the
one that is cheapest/most likely to prune a row group first.

This reduces the cyclist query from ~65ms to ~60ms
 2018-06-24 11:08:56 -04:00
Colin Dellow fd87c44ccd Add link to csv2parquet 2018-06-23 23:58:13 -04:00
Colin Dellow e1a86954e5 Revert "Don't eagerly evaluate constraints" 
This reverts commit cbde3c73b6.

This regresses:

```
WITH inputs AS (
  SELECT
    geo_name,
    CASE WHEN profile_id = 1930 THEN 'total' ELSE 'cyclist' END AS mode,
    female,
    male
  FROM census
  WHERE profile_id IN ( '1930', '1935') AND
    csd_type_name = 'CY' AND
    geo_name IN ('Victoria', 'Dawson Creek', 'Kitchener')
)
SELECT
  total.geo_name,
  cyclist.male,
  cyclist.female,
  100.0 * cyclist.male / total.male,
  100.0 * cyclist.female / total.female
FROM inputs AS total
JOIN inputs AS cyclist USING (geo_name)
WHERE total.mode = 'total' AND cyclist.mode = 'cyclist';
```

while improving:

```
select count(*) from census where geo_name in ('Dawson Creek', 'Kitchener', 'Victoria') and csd_type_name = 'CY' and profile_id = '1930';
```

which seems like a bad tradeoff.
 2018-06-23 20:48:39 -04:00
Colin Dellow 603153c36c avoid looking up physical type 2018-06-23 20:42:38 -04:00
Colin Dellow cbde3c73b6 Don't eagerly evaluate constraints 
...to avoid decompressing columns when we know from previous
columns that the row can't match.

Fixes #10
 2018-06-23 20:31:03 -04:00
Colin Dellow d7c5002cee Move some code out of ensureColumn 
Saves ~4% on the cold census needle query (~425ms -> ~405ms)
 2018-06-23 19:10:23 -04:00
Colin Dellow b9c58bd97e persist row group clauses on EOF 
...not on close. Fixes #9
 2018-06-23 16:25:56 -04:00
Colin Dellow 6d4be61261 tweak Makefile 2018-06-23 16:13:18 -04:00