mirror of
https://github.com/pgvector/pgvector.git
synced 2026-07-22 12:07:34 +08:00
Compare commits
23 Commits
hnsw-build
...
ivfflat-st
| Author | SHA1 | Date | |
|---|---|---|---|
|
|
fc5cce1cb8 | ||
|
|
1749ecb6e7 | ||
|
|
592a711a94 | ||
|
|
139007ea68 | ||
|
|
3f49b95f01 | ||
|
|
ef1bea7163 | ||
|
|
e630efd195 | ||
|
|
b5b912906b | ||
|
|
4b5db94307 | ||
|
|
65e70326b8 | ||
|
|
71641ed84e | ||
|
|
f3dba25036 | ||
|
|
5588ba6410 | ||
|
|
ec9fac5456 | ||
|
|
8085d3e538 | ||
|
|
cae162ffc6 | ||
|
|
62482e3760 | ||
|
|
c81302b835 | ||
|
|
a713e2acaa | ||
|
|
6e1312ddbe | ||
|
|
4ef5bca275 | ||
|
|
1ecf6ada76 | ||
|
|
564a3d45fc |
8
.github/workflows/build.yml
vendored
8
.github/workflows/build.yml
vendored
@@ -8,6 +8,8 @@ jobs:
|
||||
fail-fast: false
|
||||
matrix:
|
||||
include:
|
||||
- postgres: 17
|
||||
os: ubuntu-22.04
|
||||
- postgres: 16
|
||||
os: ubuntu-22.04
|
||||
- postgres: 15
|
||||
@@ -21,7 +23,7 @@ jobs:
|
||||
- postgres: 11
|
||||
os: ubuntu-20.04
|
||||
steps:
|
||||
- uses: actions/checkout@v3
|
||||
- uses: actions/checkout@v4
|
||||
- uses: ankane/setup-postgres@v1
|
||||
with:
|
||||
postgres-version: ${{ matrix.postgres }}
|
||||
@@ -43,7 +45,7 @@ jobs:
|
||||
runs-on: macos-latest
|
||||
if: ${{ !startsWith(github.ref_name, 'windows') }}
|
||||
steps:
|
||||
- uses: actions/checkout@v3
|
||||
- uses: actions/checkout@v4
|
||||
- uses: ankane/setup-postgres@v1
|
||||
with:
|
||||
postgres-version: 14
|
||||
@@ -65,7 +67,7 @@ jobs:
|
||||
runs-on: windows-latest
|
||||
if: ${{ !startsWith(github.ref_name, 'mac') }}
|
||||
steps:
|
||||
- uses: actions/checkout@v3
|
||||
- uses: actions/checkout@v4
|
||||
- uses: ankane/setup-postgres@v1
|
||||
with:
|
||||
postgres-version: 14
|
||||
|
||||
@@ -1,6 +1,7 @@
|
||||
## 0.5.1 (unreleased)
|
||||
## 0.5.1 (2023-10-10)
|
||||
|
||||
- Improved performance of index scans for IVFFlat after updates and deletes
|
||||
- Improved performance of HNSW index builds
|
||||
- Added check for MVCC-compliant snapshot for index scans
|
||||
|
||||
## 0.5.0 (2023-08-28)
|
||||
|
||||
|
||||
@@ -2,7 +2,7 @@
|
||||
"name": "vector",
|
||||
"abstract": "Open-source vector similarity search for Postgres",
|
||||
"description": "Supports L2 distance, inner product, and cosine distance",
|
||||
"version": "0.5.0",
|
||||
"version": "0.5.1",
|
||||
"maintainer": [
|
||||
"Andrew Kane <andrew@ankane.org>"
|
||||
],
|
||||
@@ -20,7 +20,7 @@
|
||||
"vector": {
|
||||
"file": "sql/vector.sql",
|
||||
"docfile": "README.md",
|
||||
"version": "0.5.0",
|
||||
"version": "0.5.1",
|
||||
"abstract": "Open-source vector similarity search for Postgres"
|
||||
}
|
||||
},
|
||||
|
||||
2
Makefile
2
Makefile
@@ -1,5 +1,5 @@
|
||||
EXTENSION = vector
|
||||
EXTVERSION = 0.5.0
|
||||
EXTVERSION = 0.5.1
|
||||
|
||||
MODULE_big = vector
|
||||
DATA = $(wildcard sql/*--*.sql)
|
||||
|
||||
@@ -1,5 +1,5 @@
|
||||
EXTENSION = vector
|
||||
EXTVERSION = 0.5.0
|
||||
EXTVERSION = 0.5.1
|
||||
|
||||
OBJS = src\hnsw.obj src\hnswbuild.obj src\hnswinsert.obj src\hnswscan.obj src\hnswutils.obj src\hnswvacuum.obj src\ivfbuild.obj src\ivfflat.obj src\ivfinsert.obj src\ivfkmeans.obj src\ivfscan.obj src\ivfutils.obj src\ivfvacuum.obj src\vector.obj
|
||||
HEADERS = src\vector.h
|
||||
|
||||
@@ -18,7 +18,7 @@ Compile and install the extension (supports Postgres 11+)
|
||||
|
||||
```sh
|
||||
cd /tmp
|
||||
git clone --branch v0.5.0 https://github.com/pgvector/pgvector.git
|
||||
git clone --branch v0.5.1 https://github.com/pgvector/pgvector.git
|
||||
cd pgvector
|
||||
make
|
||||
make install # may need sudo
|
||||
@@ -509,7 +509,7 @@ Then use `nmake` to build:
|
||||
|
||||
```cmd
|
||||
set "PGROOT=C:\Program Files\PostgreSQL\15"
|
||||
git clone --branch v0.5.0 https://github.com/pgvector/pgvector.git
|
||||
git clone --branch v0.5.1 https://github.com/pgvector/pgvector.git
|
||||
cd pgvector
|
||||
nmake /F Makefile.win
|
||||
nmake /F Makefile.win install
|
||||
@@ -530,7 +530,7 @@ This adds pgvector to the [Postgres image](https://hub.docker.com/_/postgres) (r
|
||||
You can also build the image manually:
|
||||
|
||||
```sh
|
||||
git clone --branch v0.5.0 https://github.com/pgvector/pgvector.git
|
||||
git clone --branch v0.5.1 https://github.com/pgvector/pgvector.git
|
||||
cd pgvector
|
||||
docker build --build-arg PG_MAJOR=15 -t myuser/pgvector .
|
||||
```
|
||||
|
||||
2
sql/vector--0.5.0--0.5.1.sql
Normal file
2
sql/vector--0.5.0--0.5.1.sql
Normal file
@@ -0,0 +1,2 @@
|
||||
-- complain if script is sourced in psql, rather than via CREATE EXTENSION
|
||||
\echo Use "ALTER EXTENSION vector UPDATE TO '0.5.1'" to load this file. \quit
|
||||
@@ -57,6 +57,8 @@
|
||||
/* PROGRESS_CREATEIDX_SUBPHASE_INITIALIZE is 1 */
|
||||
#define PROGRESS_HNSW_PHASE_LOAD 2
|
||||
|
||||
#define HNSW_MAX_SIZE (BLCKSZ - MAXALIGN(SizeOfPageHeaderData) - MAXALIGN(sizeof(HnswPageOpaqueData)) - sizeof(ItemIdData))
|
||||
|
||||
#define HNSW_ELEMENT_TUPLE_SIZE(_dim) MAXALIGN(offsetof(HnswElementTupleData, vec) + VECTOR_SIZE(_dim))
|
||||
#define HNSW_NEIGHBOR_TUPLE_SIZE(level, m) MAXALIGN(offsetof(HnswNeighborTupleData, indextids) + ((level) + 2) * (m) * sizeof(ItemPointerData))
|
||||
|
||||
@@ -110,13 +112,14 @@ typedef struct HnswCandidate
|
||||
{
|
||||
HnswElement element;
|
||||
float distance;
|
||||
bool closer;
|
||||
} HnswCandidate;
|
||||
|
||||
typedef struct HnswNeighborArray
|
||||
{
|
||||
int length;
|
||||
bool closerSet;
|
||||
HnswCandidate *items;
|
||||
HnswElement firstPruned;
|
||||
} HnswNeighborArray;
|
||||
|
||||
typedef struct HnswPairingHeapNode
|
||||
|
||||
@@ -117,12 +117,12 @@ CreateElementPages(HnswBuildState * buildstate)
|
||||
ListCell *lc;
|
||||
|
||||
/* Calculate sizes */
|
||||
maxSize = BLCKSZ - MAXALIGN(SizeOfPageHeaderData) - MAXALIGN(sizeof(HnswPageOpaqueData));
|
||||
maxSize = HNSW_MAX_SIZE;
|
||||
etupSize = HNSW_ELEMENT_TUPLE_SIZE(dimensions);
|
||||
|
||||
/* Allocate once */
|
||||
etup = palloc0(etupSize);
|
||||
ntup = palloc0(maxSize);
|
||||
ntup = palloc0(BLCKSZ);
|
||||
|
||||
/* Prepare first page */
|
||||
buf = HnswNewBuffer(index, forkNum);
|
||||
|
||||
@@ -135,7 +135,7 @@ WriteNewElementPages(Relation index, HnswElement e, int m, BlockNumber insertPag
|
||||
etupSize = HNSW_ELEMENT_TUPLE_SIZE(dimensions);
|
||||
ntupSize = HNSW_NEIGHBOR_TUPLE_SIZE(e->level, m);
|
||||
combinedSize = etupSize + ntupSize + sizeof(ItemIdData);
|
||||
maxSize = BLCKSZ - MAXALIGN(SizeOfPageHeaderData) - MAXALIGN(sizeof(HnswPageOpaqueData));
|
||||
maxSize = HNSW_MAX_SIZE;
|
||||
minCombinedSize = etupSize + HNSW_NEIGHBOR_TUPLE_SIZE(0, m) + sizeof(ItemIdData);
|
||||
|
||||
/* Prepare element tuple */
|
||||
|
||||
@@ -160,6 +160,11 @@ hnswgettuple(IndexScanDesc scan, ScanDirection dir)
|
||||
if (scan->orderByData == NULL)
|
||||
elog(ERROR, "cannot scan hnsw index without order");
|
||||
|
||||
/* Requires MVCC-compliant snapshot as not able to maintain a pin */
|
||||
/* https://www.postgresql.org/docs/current/index-locking.html */
|
||||
if (!IsMVCCSnapshot(scan->xs_snapshot))
|
||||
elog(ERROR, "non-MVCC snapshots are not supported with hnsw");
|
||||
|
||||
/* Get scan value */
|
||||
value = GetScanValue(scan);
|
||||
|
||||
@@ -201,15 +206,6 @@ hnswgettuple(IndexScanDesc scan, ScanDirection dir)
|
||||
scan->xs_ctup.t_self = *heaptid;
|
||||
#endif
|
||||
|
||||
/*
|
||||
* Typically, an index scan must maintain a pin on the index page
|
||||
* holding the item last returned by amgettuple. However, this is not
|
||||
* needed with the current vacuum strategy, which ensures scans do not
|
||||
* visit tuples in danger of being marked as deleted.
|
||||
*
|
||||
* https://www.postgresql.org/docs/current/index-locking.html
|
||||
*/
|
||||
|
||||
scan->xs_recheckorderby = false;
|
||||
return true;
|
||||
}
|
||||
|
||||
127
src/hnswutils.c
127
src/hnswutils.c
@@ -139,7 +139,7 @@ HnswInitNeighbors(HnswElement element, int m)
|
||||
a = &element->neighbors[lc];
|
||||
a->length = 0;
|
||||
a->items = palloc(sizeof(HnswCandidate) * lm);
|
||||
a->firstPruned = NULL;
|
||||
a->closerSet = false;
|
||||
}
|
||||
}
|
||||
|
||||
@@ -693,6 +693,34 @@ HnswSearchLayer(Datum q, List *ep, int ef, int lc, Relation index, FmgrInfo *pro
|
||||
return w;
|
||||
}
|
||||
|
||||
/*
|
||||
* Compare candidate distances
|
||||
*/
|
||||
static int
|
||||
#if PG_VERSION_NUM >= 130000
|
||||
CompareCandidateDistances(const ListCell *a, const ListCell *b)
|
||||
#else
|
||||
CompareCandidateDistances(const void *a, const void *b)
|
||||
#endif
|
||||
{
|
||||
HnswCandidate *hca = lfirst((ListCell *) a);
|
||||
HnswCandidate *hcb = lfirst((ListCell *) b);
|
||||
|
||||
if (hca->distance < hcb->distance)
|
||||
return 1;
|
||||
|
||||
if (hca->distance > hcb->distance)
|
||||
return -1;
|
||||
|
||||
if (hca->element < hcb->element)
|
||||
return 1;
|
||||
|
||||
if (hca->element > hcb->element)
|
||||
return -1;
|
||||
|
||||
return 0;
|
||||
}
|
||||
|
||||
/*
|
||||
* Calculate the distance between elements
|
||||
*/
|
||||
@@ -749,61 +777,76 @@ CheckElementCloser(HnswCandidate * e, List *r, int lc, FmgrInfo *procinfo, Oid c
|
||||
* Algorithm 4 from paper
|
||||
*/
|
||||
static List *
|
||||
SelectNeighbors(List *c, int m, int lc, FmgrInfo *procinfo, Oid collation, HnswElement e2, HnswCandidate * newCandidate, HnswCandidate * *pruned)
|
||||
SelectNeighbors(List *c, int m, int lc, FmgrInfo *procinfo, Oid collation, HnswElement e2, HnswCandidate * newCandidate, HnswCandidate * *pruned, bool sortCandidates)
|
||||
{
|
||||
List *r = NIL;
|
||||
List *w = list_copy(c);
|
||||
pairingheap *wd;
|
||||
bool mustCalculate = e2->neighbors[lc].firstPruned == NULL;
|
||||
bool mustCalculate = !e2->neighbors[lc].closerSet;
|
||||
List *added = NIL;
|
||||
bool removedAny = false;
|
||||
|
||||
if (list_length(w) <= m)
|
||||
return w;
|
||||
|
||||
wd = pairingheap_allocate(CompareNearestCandidates, NULL);
|
||||
|
||||
/* Ensure order of candidates is deterministic for closer caching */
|
||||
if (sortCandidates)
|
||||
list_sort(w, CompareCandidateDistances);
|
||||
|
||||
while (list_length(w) > 0 && list_length(r) < m)
|
||||
{
|
||||
/* Assumes w is already ordered desc */
|
||||
HnswCandidate *e = llast(w);
|
||||
bool closer;
|
||||
|
||||
w = list_delete_last(w);
|
||||
|
||||
/*
|
||||
* r and wd will be the same as previous calls until the new
|
||||
* candidate, so can skip distance calculations for as many candidates
|
||||
* as there is state for
|
||||
*/
|
||||
/* Use previous state of r and wd to skip work when possible */
|
||||
if (mustCalculate)
|
||||
closer = CheckElementCloser(e, r, lc, procinfo, collation);
|
||||
else if (e->element == e2->neighbors[lc].firstPruned)
|
||||
e->closer = CheckElementCloser(e, r, lc, procinfo, collation);
|
||||
else if (list_length(added) > 0)
|
||||
{
|
||||
closer = false;
|
||||
|
||||
/*
|
||||
* Could store multiple pruned and only calculate when exhausted
|
||||
* (or store full state) at the expense of memory
|
||||
* If the current candidate was closer, we only need to compare it
|
||||
* with the other candidates that we have added.
|
||||
*/
|
||||
mustCalculate = true;
|
||||
if (e->closer)
|
||||
{
|
||||
e->closer = CheckElementCloser(e, added, lc, procinfo, collation);
|
||||
|
||||
if (!e->closer)
|
||||
removedAny = true;
|
||||
}
|
||||
else
|
||||
{
|
||||
/*
|
||||
* If we have removed any candidates from closer, a candidate
|
||||
* that was not closer earlier might now be.
|
||||
*/
|
||||
if (removedAny)
|
||||
{
|
||||
e->closer = CheckElementCloser(e, r, lc, procinfo, collation);
|
||||
if (e->closer)
|
||||
added = lappend(added, e);
|
||||
}
|
||||
}
|
||||
}
|
||||
else if (e == newCandidate)
|
||||
{
|
||||
closer = CheckElementCloser(e, r, lc, procinfo, collation);
|
||||
mustCalculate = true;
|
||||
e->closer = CheckElementCloser(e, r, lc, procinfo, collation);
|
||||
if (e->closer)
|
||||
added = lappend(added, e);
|
||||
}
|
||||
else
|
||||
closer = true;
|
||||
|
||||
if (closer)
|
||||
if (e->closer)
|
||||
r = lappend(r, e);
|
||||
else
|
||||
pairingheap_add(wd, &(CreatePairingHeapNode(e)->ph_node));
|
||||
}
|
||||
|
||||
/* Save first pruned */
|
||||
/* OK to leave previous element if empty */
|
||||
if (!pairingheap_is_empty(wd))
|
||||
e2->neighbors[lc].firstPruned = ((HnswPairingHeapNode *) pairingheap_first(wd))->inner->element;
|
||||
/* Cached value can only be used in future if sorted deterministically */
|
||||
e2->neighbors[lc].closerSet = sortCandidates;
|
||||
|
||||
/* Keep pruned connections */
|
||||
while (!pairingheap_is_empty(wd) && list_length(r) < m)
|
||||
@@ -858,28 +901,6 @@ AddConnections(HnswElement element, List *neighbors, int m, int lc)
|
||||
a->items[a->length++] = *((HnswCandidate *) lfirst(lc2));
|
||||
}
|
||||
|
||||
/*
|
||||
* Compare candidate distances
|
||||
*/
|
||||
static int
|
||||
#if PG_VERSION_NUM >= 130000
|
||||
CompareCandidateDistances(const ListCell *a, const ListCell *b)
|
||||
#else
|
||||
CompareCandidateDistances(const void *a, const void *b)
|
||||
#endif
|
||||
{
|
||||
HnswCandidate *hca = lfirst((ListCell *) a);
|
||||
HnswCandidate *hcb = lfirst((ListCell *) b);
|
||||
|
||||
if (hca->distance < hcb->distance)
|
||||
return 1;
|
||||
|
||||
if (hca->distance > hcb->distance)
|
||||
return -1;
|
||||
|
||||
return 0;
|
||||
}
|
||||
|
||||
/*
|
||||
* Update connections
|
||||
*/
|
||||
@@ -933,13 +954,12 @@ HnswUpdateConnection(HnswElement element, HnswCandidate * hc, int m, int lc, int
|
||||
{
|
||||
List *c = NIL;
|
||||
|
||||
/* Add and sort candidates */
|
||||
/* Add candidates */
|
||||
for (int i = 0; i < currentNeighbors->length; i++)
|
||||
c = lappend(c, ¤tNeighbors->items[i]);
|
||||
c = lappend(c, &hc2);
|
||||
list_sort(c, CompareCandidateDistances);
|
||||
|
||||
SelectNeighbors(c, m, lc, procinfo, collation, hc->element, &hc2, &pruned);
|
||||
SelectNeighbors(c, m, lc, procinfo, collation, hc->element, &hc2, &pruned, true);
|
||||
|
||||
/* Should not happen */
|
||||
if (pruned == NULL)
|
||||
@@ -1038,7 +1058,12 @@ HnswInsertElement(HnswElement element, HnswElement entryPoint, Relation index, F
|
||||
else
|
||||
lw = w;
|
||||
|
||||
neighbors = SelectNeighbors(lw, lm, lc, procinfo, collation, element, NULL, NULL);
|
||||
/*
|
||||
* Candidates are sorted, but not deterministically. Could set
|
||||
* sortCandidates to true for in-memory builds to enable closer
|
||||
* caching, but there does not seem to be a difference in performance.
|
||||
*/
|
||||
neighbors = SelectNeighbors(lw, lm, lc, procinfo, collation, element, NULL, NULL, false);
|
||||
|
||||
AddConnections(element, neighbors, lm, lc);
|
||||
|
||||
|
||||
@@ -287,7 +287,7 @@ InsertTuples(Relation index, IvfflatBuildState * buildstate, ForkNumber forkNum)
|
||||
#else
|
||||
TupleTableSlot *slot = MakeSingleTupleTableSlot(buildstate->tupdesc);
|
||||
#endif
|
||||
TupleDesc tupdesc = RelationGetDescr(index);
|
||||
TupleDesc tupdesc = IvfflatTupleDesc(index);
|
||||
|
||||
UpdateProgress(PROGRESS_CREATEIDX_SUBPHASE, PROGRESS_IVFFLAT_PHASE_LOAD);
|
||||
|
||||
@@ -339,6 +339,8 @@ InsertTuples(Relation index, IvfflatBuildState * buildstate, ForkNumber forkNum)
|
||||
/* Set the start and insert pages */
|
||||
IvfflatUpdateList(index, buildstate->listInfo[i], insertPage, InvalidBlockNumber, startPage, forkNum);
|
||||
}
|
||||
|
||||
FreeTupleDesc(tupdesc);
|
||||
}
|
||||
|
||||
/*
|
||||
|
||||
@@ -246,8 +246,6 @@ typedef struct IvfflatScanOpaqueData
|
||||
int probes;
|
||||
int dimensions;
|
||||
bool first;
|
||||
Buffer buf;
|
||||
ItemPointerData heaptid;
|
||||
|
||||
/* Sorting */
|
||||
Tuplesortstate *sortstate;
|
||||
@@ -289,6 +287,7 @@ void IvfflatInitPage(Buffer buf, Page page);
|
||||
void IvfflatInitRegisterPage(Relation index, Buffer *buf, Page *page, GenericXLogState **state);
|
||||
void IvfflatInit(void);
|
||||
PGDLLEXPORT void IvfflatParallelBuildMain(dsm_segment *seg, shm_toc *toc);
|
||||
TupleDesc IvfflatTupleDesc(Relation index);
|
||||
|
||||
/* Index access methods */
|
||||
IndexBuildResult *ivfflatbuild(Relation heap, Relation index, IndexInfo *indexInfo);
|
||||
|
||||
@@ -94,12 +94,12 @@ InsertTuple(Relation index, Datum *values, bool *isnull, ItemPointer heap_tid, R
|
||||
originalInsertPage = insertPage;
|
||||
|
||||
/* Form tuple */
|
||||
itup = index_form_tuple(RelationGetDescr(index), &value, isnull);
|
||||
itup = index_form_tuple(IvfflatTupleDesc(index), &value, isnull);
|
||||
itup->t_tid = *heap_tid;
|
||||
|
||||
/* Get tuple size */
|
||||
itemsz = MAXALIGN(IndexTupleSize(itup));
|
||||
Assert(itemsz <= BLCKSZ - MAXALIGN(SizeOfPageHeaderData) - MAXALIGN(sizeof(IvfflatPageOpaqueData)));
|
||||
Assert(itemsz <= BLCKSZ - MAXALIGN(SizeOfPageHeaderData) - MAXALIGN(sizeof(IvfflatPageOpaqueData)) - sizeof(ItemIdData));
|
||||
|
||||
/* Find a page to insert the item */
|
||||
for (;;)
|
||||
|
||||
@@ -143,10 +143,6 @@ GetScanItems(IndexScanDesc scan, Datum value)
|
||||
bool isnull;
|
||||
ItemId itemid = PageGetItemId(page, offno);
|
||||
|
||||
/* Skip dead tuples */
|
||||
if (scan->ignore_killed_tuples && ItemIdIsDead(itemid))
|
||||
continue;
|
||||
|
||||
itup = (IndexTuple) PageGetItem(page, itemid);
|
||||
datum = index_getattr(itup, 1, tupdesc, &isnull);
|
||||
|
||||
@@ -161,8 +157,6 @@ GetScanItems(IndexScanDesc scan, Datum value)
|
||||
slot->tts_isnull[0] = false;
|
||||
slot->tts_values[1] = PointerGetDatum(&itup->t_tid);
|
||||
slot->tts_isnull[1] = false;
|
||||
slot->tts_values[2] = Int32GetDatum((int) searchPage);
|
||||
slot->tts_isnull[2] = false;
|
||||
ExecStoreVirtualTuple(slot);
|
||||
|
||||
tuplesort_puttupleslot(so->sortstate, slot);
|
||||
@@ -187,55 +181,6 @@ GetScanItems(IndexScanDesc scan, Datum value)
|
||||
tuplesort_performsort(so->sortstate);
|
||||
}
|
||||
|
||||
/*
|
||||
* Mark prior tuple as dead
|
||||
*/
|
||||
static void
|
||||
MarkPriorTupleDead(IndexScanDesc scan)
|
||||
{
|
||||
IvfflatScanOpaque so = (IvfflatScanOpaque) scan->opaque;
|
||||
Buffer buf = so->buf;
|
||||
Page page;
|
||||
OffsetNumber maxoffno;
|
||||
|
||||
/* Safety check */
|
||||
if (!BufferIsValid(so->buf) || !ItemPointerIsValid(&so->heaptid))
|
||||
return;
|
||||
|
||||
/* Only a shared locked is needed for ItemIdMarkDead */
|
||||
LockBuffer(buf, BUFFER_LOCK_SHARE);
|
||||
page = BufferGetPage(buf);
|
||||
maxoffno = PageGetMaxOffsetNumber(page);
|
||||
|
||||
for (OffsetNumber offno = FirstOffsetNumber; offno <= maxoffno; offno = OffsetNumberNext(offno))
|
||||
{
|
||||
ItemId itemid = PageGetItemId(page, offno);
|
||||
IndexTuple itup = (IndexTuple) PageGetItem(page, itemid);
|
||||
|
||||
/*
|
||||
* Find tuple. Since buffer has been pinned, tuple cannot have been
|
||||
* vacuumed (and heap TID reused).
|
||||
*/
|
||||
if (ItemPointerEquals(&itup->t_tid, &so->heaptid))
|
||||
{
|
||||
/*
|
||||
* Make sure tuple has not already been marked dead to avoid extra
|
||||
* WAL if wal_log_hints or data checksums enabled
|
||||
*/
|
||||
if (!ItemIdIsDead(itemid))
|
||||
{
|
||||
ItemIdMarkDead(itemid);
|
||||
MarkBufferDirtyHint(buf, true);
|
||||
}
|
||||
|
||||
break;
|
||||
}
|
||||
}
|
||||
|
||||
/* Unlock buffer */
|
||||
LockBuffer(buf, BUFFER_LOCK_UNLOCK);
|
||||
}
|
||||
|
||||
/*
|
||||
* Prepare for an index scan
|
||||
*/
|
||||
@@ -261,9 +206,7 @@ ivfflatbeginscan(Relation index, int nkeys, int norderbys)
|
||||
probes = lists;
|
||||
|
||||
so = (IvfflatScanOpaque) palloc(offsetof(IvfflatScanOpaqueData, lists) + probes * sizeof(IvfflatScanList));
|
||||
so->buf = InvalidBuffer;
|
||||
so->first = true;
|
||||
ItemPointerSetInvalid(&so->heaptid);
|
||||
so->probes = probes;
|
||||
so->dimensions = dimensions;
|
||||
|
||||
@@ -274,13 +217,12 @@ ivfflatbeginscan(Relation index, int nkeys, int norderbys)
|
||||
|
||||
/* Create tuple description for sorting */
|
||||
#if PG_VERSION_NUM >= 120000
|
||||
so->tupdesc = CreateTemplateTupleDesc(3);
|
||||
so->tupdesc = CreateTemplateTupleDesc(2);
|
||||
#else
|
||||
so->tupdesc = CreateTemplateTupleDesc(3, false);
|
||||
so->tupdesc = CreateTemplateTupleDesc(2, false);
|
||||
#endif
|
||||
TupleDescInitEntry(so->tupdesc, (AttrNumber) 1, "distance", FLOAT8OID, -1, 0);
|
||||
TupleDescInitEntry(so->tupdesc, (AttrNumber) 2, "heaptid", TIDOID, -1, 0);
|
||||
TupleDescInitEntry(so->tupdesc, (AttrNumber) 3, "indexblkno", INT4OID, -1, 0);
|
||||
|
||||
/* Prep sort */
|
||||
so->sortstate = tuplesort_begin_heap(so->tupdesc, 1, attNums, sortOperators, sortCollations, nullsFirstFlags, work_mem, NULL, false);
|
||||
@@ -312,7 +254,6 @@ ivfflatrescan(IndexScanDesc scan, ScanKey keys, int nkeys, ScanKey orderbys, int
|
||||
#endif
|
||||
|
||||
so->first = true;
|
||||
ItemPointerSetInvalid(&so->heaptid);
|
||||
pairingheap_reset(so->listQueue);
|
||||
|
||||
if (keys && scan->numberOfKeys > 0)
|
||||
@@ -347,6 +288,11 @@ ivfflatgettuple(IndexScanDesc scan, ScanDirection dir)
|
||||
if (scan->orderByData == NULL)
|
||||
elog(ERROR, "cannot scan ivfflat index without order");
|
||||
|
||||
/* Requires MVCC-compliant snapshot as not able to pin during sorting */
|
||||
/* https://www.postgresql.org/docs/current/index-locking.html */
|
||||
if (!IsMVCCSnapshot(scan->xs_snapshot))
|
||||
elog(ERROR, "non-MVCC snapshots are not supported with ivfflat");
|
||||
|
||||
if (scan->orderByData->sk_flags & SK_ISNULL)
|
||||
value = PointerGetDatum(InitVector(so->dimensions));
|
||||
else
|
||||
@@ -370,17 +316,10 @@ ivfflatgettuple(IndexScanDesc scan, ScanDirection dir)
|
||||
if (value != scan->orderByData->sk_argument)
|
||||
pfree(DatumGetPointer(value));
|
||||
}
|
||||
else
|
||||
{
|
||||
/* Mark prior tuple as dead */
|
||||
if (scan->kill_prior_tuple)
|
||||
MarkPriorTupleDead(scan);
|
||||
}
|
||||
|
||||
if (tuplesort_gettupleslot(so->sortstate, true, false, so->slot, NULL))
|
||||
{
|
||||
ItemPointer heaptid = (ItemPointer) DatumGetPointer(slot_getattr(so->slot, 2, &so->isnull));
|
||||
BlockNumber indexblkno = DatumGetInt32(slot_getattr(so->slot, 3, &so->isnull));
|
||||
|
||||
#if PG_VERSION_NUM >= 120000
|
||||
scan->xs_heaptid = *heaptid;
|
||||
@@ -388,21 +327,6 @@ ivfflatgettuple(IndexScanDesc scan, ScanDirection dir)
|
||||
scan->xs_ctup.t_self = *heaptid;
|
||||
#endif
|
||||
|
||||
/* Keep track of info needed to mark tuple as dead */
|
||||
so->heaptid = *heaptid;
|
||||
|
||||
/* Unpin buffer */
|
||||
if (BufferIsValid(so->buf))
|
||||
ReleaseBuffer(so->buf);
|
||||
|
||||
/*
|
||||
* An index scan must maintain a pin on the index page holding the
|
||||
* item last returned by amgettuple
|
||||
*
|
||||
* https://www.postgresql.org/docs/current/index-locking.html
|
||||
*/
|
||||
so->buf = ReadBuffer(scan->indexRelation, indexblkno);
|
||||
|
||||
scan->xs_recheckorderby = false;
|
||||
return true;
|
||||
}
|
||||
@@ -418,10 +342,6 @@ ivfflatendscan(IndexScanDesc scan)
|
||||
{
|
||||
IvfflatScanOpaque so = (IvfflatScanOpaque) scan->opaque;
|
||||
|
||||
/* Release pin */
|
||||
if (BufferIsValid(so->buf))
|
||||
ReleaseBuffer(so->buf);
|
||||
|
||||
pairingheap_free(so->listQueue);
|
||||
tuplesort_end(so->sortstate);
|
||||
|
||||
|
||||
@@ -4,6 +4,10 @@
|
||||
#include "storage/bufmgr.h"
|
||||
#include "vector.h"
|
||||
|
||||
#if PG_VERSION_NUM < 130000
|
||||
#define TYPSTORAGE_PLAIN 'p'
|
||||
#endif
|
||||
|
||||
/*
|
||||
* Allocate a vector array
|
||||
*/
|
||||
@@ -238,3 +242,17 @@ IvfflatUpdateList(Relation index, ListInfo listInfo,
|
||||
UnlockReleaseBuffer(buf);
|
||||
}
|
||||
}
|
||||
|
||||
/*
|
||||
* Get the tuple descriptor
|
||||
*/
|
||||
TupleDesc
|
||||
IvfflatTupleDesc(Relation index)
|
||||
{
|
||||
TupleDesc tupdesc = CreateTupleDescCopyConstr(RelationGetDescr(index));
|
||||
|
||||
/* Prevent compression */
|
||||
TupleDescAttr(tupdesc, 0)->attstorage = TYPSTORAGE_PLAIN;
|
||||
|
||||
return tupdesc;
|
||||
}
|
||||
|
||||
@@ -1,4 +1,4 @@
|
||||
comment = 'vector data type and ivfflat and hnsw access methods'
|
||||
default_version = '0.5.0'
|
||||
default_version = '0.5.1'
|
||||
module_pathname = '$libdir/vector'
|
||||
relocatable = true
|
||||
|
||||
Reference in New Issue
Block a user