database/mariadb-columnstore-engine
1
0
Fork 0
mirror of https://github.com/mariadb-corporation/mariadb-columnstore-engine.git synced 2025-08-01 06:46:55 +03:00
Code Activity

MCOL-4580 extent elimination for dictionary-based text/varchar types

Browse Source 
The idea is relatively simple - encode prefixes of collated strings as
integers and use them to compute extents' ranges. Then we can eliminate
extents with strings.

The actual patch does have all the code there but miss one important
step: we do not keep collation index, we keep charset index. Because of
this, some of the tests in the bugfix suite fail and thus main
functionality is turned off.

The reason of this patch to be put into PR at all is that it contains
changes that made CHAR/VARCHAR columns unsigned. This change is needed in
vectorization work.
This commit is contained in:
Serguey Zefirov
2022-02-04 11:55:09 +00:00
parent a66a8dfabf
commit 53b9a2a0f9
54 changed files with 698 additions and 227 deletions
Download Patch File Download Diff File Expand all files Collapse all files

79
dbcon/joblist/columncommand-jl.cpp
View File

@ -130,6 +130,59 @@ ColumnCommandJL::ColumnCommandJL(const pColStep& step)
fFilesPerColumnPartition = cf->uFromText(fpc);
}
ColumnCommandJL::ColumnCommandJL(const ColumnCommandJL& prevCmd, const DictStepJL& dictWithFilters)
{
BRM::DBRM dbrm;
/* grab necessary vars from scan */
traceFlags = prevCmd.traceFlags;
// we should call this constructor only when paired with dictionary
// and in that case previous command should not have any filters and
// should be "dict" (tokens) column command.
idbassert(dictWithFilters.getFilterCount() == 0 || prevCmd.filterCount == 0);
idbassert(prevCmd.fIsDict);
// need to reencode filters.
filterString = dictWithFilters.reencodedFilterString();
// we have a limitation here.
// consider this: textcol IS NULL AND textcol IN ('a', 'b')
// XXX: should check.
if (filterString.length() > 0 && (BOP = dictWithFilters.getBop() || prevCmd.filterString.length() < 1))
{
filterCount = dictWithFilters.getFilterCount();
BOP = dictWithFilters.getBop();
fContainsRanges = true;
}
else
{
filterCount = prevCmd.filterCount;
filterString = prevCmd.filterString;
BOP = prevCmd.BOP;
}
isScan = prevCmd.isScan;
colType = prevCmd.colType;
extents = prevCmd.extents;
OID = prevCmd.OID;
colName = prevCmd.colName;
rpbShift = prevCmd.rpbShift;
fIsDict = prevCmd.fIsDict;
fLastLbid = prevCmd.fLastLbid;
lbid = prevCmd.lbid;
traceFlags = prevCmd.traceFlags;
dbroot = prevCmd.dbroot;
numDBRoots = prevCmd.numDBRoots;
/* I think modmask isn't necessary for scans */
divShift = prevCmd.divShift;
modMask = (1 << divShift) - 1;
// @Bug 2889. Drop partition enhancement. Read FilesPerColumnPartition and ExtentsPerSegmentFile for use
// in RID calculation.
fFilesPerColumnPartition = prevCmd.fFilesPerColumnPartition;
// MCOL-4685 remove the option to set more than 2 extents per file (ExtentsPreSegmentFile).
fExtentsPerSegmentFile = prevCmd.fExtentsPerSegmentFile;
}
ColumnCommandJL::~ColumnCommandJL()
{
}
@ -141,9 +194,22 @@ void ColumnCommandJL::createCommand(ByteStream& bs) const
colType.serialize(bs);
bs << (uint8_t)isScan;
bs << traceFlags;
bs << filterString;
bs << BOP;
bs << filterCount;
if (isDict() && fContainsRanges)
{
// XXX: we should discern here between IS (NOT) NULL and other filters.
ByteStream empty;
auto zeroFC = filterCount;
bs << empty;
bs << BOP;
zeroFC = 0;
bs << zeroFC;
}
else
{
bs << filterString;
bs << BOP;
bs << filterCount;
}
serializeInlineVector(bs, fLastLbid);
CommandJL::createCommand(bs);
@ -250,7 +316,7 @@ string ColumnCommandJL::toString()
{
ostringstream ret;
ret << "ColumnCommandJL: " << filterCount << " filters colwidth=" << colType.colWidth << " oid=" << OID
ret << "ColumnCommandJL: " << filterCount << " filters, BOP=" << ((int)BOP) << ", colwidth=" << colType.colWidth << " oid=" << OID
<< " name=" << colName;
if (isScan)
@ -286,4 +352,9 @@ void ColumnCommandJL::reloadExtents()
sort(extents.begin(), extents.end(), BRM::ExtentSorter());
}
bool ColumnCommandJL::getIsDict()
{
return fIsDict;
}
}; // namespace joblist