5480de7fb6
Compile w/static linkages for parquet
...
Fixes #4 . A stock Ubuntu 14.04 can now install sqlite3:amd64 and
libboost-all-dev, then use this module to read the test parquet file.
2018-03-20 19:06:39 -04:00
8bf890ab66
Fix incorrect row pruning for non-text BYTE_ARRAY
2018-03-18 19:43:09 -04:00
e2af2a07a4
Make rowid start from 1, not 0
...
Unclear whether this is strictly required, but I'm going to start using
SQLite as an oracle, and it'll be simpler if our rowids match theirs.
2018-03-18 17:03:46 -04:00
1f3ffce560
Row group filtering for BYTE_ARRAY
2018-03-18 15:03:08 -04:00
7b302a0eb2
Bail on rowId constraint when non-int
2018-03-18 14:31:23 -04:00
3b557f7fb0
Add explicit test for file not found
...
...caching the metadata moved where ParquetTable did I/O,
which introduced a segfault on not found
2018-03-18 11:58:23 -04:00
4cbde9fc09
Row filtering for doubles
2018-03-17 16:09:57 -04:00
86e09b111e
Add row filtering for int32/64/96/boolean
2018-03-17 16:05:38 -04:00
a3af16eb54
Row-filtering for other string ops
2018-03-17 15:28:51 -04:00
03a20a9432
LIKE row group filtering
...
~1.7s -> ~1.0s for the census data set on `LIKE 'Dawson %'`
2018-03-17 00:11:38 -04:00
01e8ffaba7
Row group filtering for double/float
2018-03-16 16:30:05 -04:00
9c22fd1f57
Row group filters for strings, int32/64/96, bools
2018-03-16 16:07:41 -04:00
1f4cebe2a6
Don't use accessors
...
This drops the `= 'Dawson Creek'` query from 210ms to 145ms.
Maybe inlining would have been an option here? I'm not familiar enough
with g++ to know. :(
2018-03-15 23:04:11 -04:00
8ba13f44d5
Remove unnecessary copy
...
Now the `== 'Dawson Creek'` query is ~210ms, which is approx the
same as a `count(*)` query. This seems maybe OK, since the row group
filter is only excluding 30% of records.
2018-03-15 22:10:45 -04:00
f7f1ed03d1
add row filter for string ==
...
This gets the census `== 'Dawson Creek'` query down to ~410ms from
~650ms.
That still seems much slower than it should be. Am I accidentally
doing a copy? Now to go learn how to profile C++ code...
2018-03-15 21:37:52 -04:00
6648ff5968
add string == row group filter
...
For the statscan census set filtering on `== 'Dawson Creek'`, the query
goes from 980ms to 660ms.
This is expected, since the data isn't sorted by that column.
I'll try adding some scaffolding to do filtering at the row level, too.
We could also try unpacking the dictionary and testing the individual
values, although we may want some heuristics to decide whether it's
worth doing -- eg if < 10% of the rows have a unique value.
Ideally, this should be like a ~1ms query.
2018-03-15 20:40:21 -04:00
dc431aee20
Dispatch row group filtering based on parquet type
2018-03-15 20:25:02 -04:00
92ba5f94e0
reuse FileMetaData
...
For the statscan dataset, parsing the file metadata takes ~30-40ms,
so stash it away for future re-use.
2018-03-15 19:57:38 -04:00
769060dbcb
Add stub row group filters for text/int/dbl
...
Checkpointing to investigate why min/max stats for text aren't
present
2018-03-12 23:07:41 -04:00
110e3e3668
row group skipping for is [not] null queries
2018-03-12 21:09:00 -04:00
95748a5192
Remove bool from Constraint
2018-03-12 20:50:30 -04:00
acc15256ec
Add rowgroup filtering for rowid
2018-03-12 20:42:50 -04:00
1f938a005d
More tests cases to deal with affinity
...
I'm not sure how these manifest - whether SQLite retypes them based on
column affinity before we see them, or whether they're provided as is.
2018-03-11 19:18:44 -04:00
095b576cc2
Scaffolding for row group filters, tests
...
rowid is special since its column index is -1, so add
explicit tests around it
2018-03-11 15:44:51 -04:00
5559a7b563
Fix when last rowgroup is not same size as first
...
...change test data to use 99 rows, so that when we have
rowgroup size 10 we exercise this code.
2018-03-11 15:15:27 -04:00
830053c1fc
Scaffolding for in-extension filtering
...
Supports IS NULL and IS NOT NULL checks
2018-03-11 13:58:10 -04:00
d28ae86d15
Test unusable constraints
2018-03-10 13:38:34 -05:00
96fcafcd2f
Add test cases
2018-03-10 13:25:13 -05:00
210f322a1c
Code to pretty print constraints
2018-03-10 10:59:53 -05:00
2bc054a2cf
Add crappy Makefile
2018-03-10 10:46:10 -05:00
824a416f51
better debug logs for xBestIndex
2018-03-08 13:21:33 -05:00
0d4806ca6f
Rejig parquet generation
...
- "fixed_size_binary" -> "binary_10"
- make null parquet use rowgroups of sie 10: first rowgroup
has no nulls, 2nd has all null, 3rd-10th have alternating
nulls
This is prep for making a Postgres layer to use as an oracle
for generating test cases so that we have good coverage before
implementing advanced `xBestIndex` and `xFilter` modes.
2018-03-06 21:02:26 -05:00
67005623df
`ensureColumn` catches up when rows are skipped
2018-03-04 22:29:35 -05:00
bb3a9440f7
Add query test framework, fix xFilter
2018-03-04 21:05:26 -05:00
4c54ab89ae
Don't segfault on full table scan
2018-03-04 17:49:19 -05:00
7edb5e472f
Support BLOBs
2018-03-04 17:20:59 -05:00
67b0d96967
float support
2018-03-03 20:57:09 -05:00
18f07f4c43
More defensive, add caveats
2018-03-03 20:30:46 -05:00
eb0b48f867
Boolean, INT96, INT64
2018-03-03 20:00:50 -05:00
1de843fca8
Very rough first cut
...
supports int32, double, strings.
2018-03-03 15:44:01 -05:00
f8599f8d3e
Rename some references to CSVs
...
...some nonsensical things, like "first row of Parquet",
but we'll tidy them up later.
2018-03-02 19:18:36 -05:00
552da5a647
Initial checkin of CSV table
...
parquet.cc is a fork of the sample CSV virtual table at
https://www.sqlite.org/src/artifact?ci=trunk&filename=ext/misc/csv.c
So far the only changes are those needed to make it compile cleanly in
C++11 mode.
2018-03-02 18:59:34 -05:00
Colin Dellow