postgresql/src/include/access/relscan.h
David Rowley 56788d2156 Allocate consecutive blocks during parallel seqscans
Previously we would allocate blocks to parallel workers during a parallel
sequential scan 1 block at a time.  Since other workers were likely to
request a block before a worker returns for another block number to work
on, this could lead to non-sequential I/O patterns in each worker which
could cause the operating system's readahead to perform poorly or not at
all.

Here we change things so that we allocate consecutive "chunks" of blocks
to workers and have them work on those until they're done, at which time
we allocate another chunk for the worker.  The size of these chunks is
based on the size of the relation.

Initial patch here was by Thomas Munro which showed some good improvements
just having a fixed chunk size of 64 blocks with a simple ramp-down near
the end of the scan. The revisions of the patch to make the chunk size
based on the relation size and the adjusted ramp-down in powers of two was
done by me, along with quite extensive benchmarking to determine the
optimal chunk sizes.

For the most part, benchmarks have shown significant performance
improvements for large parallel sequential scans on Linux, FreeBSD and
Windows using SSDs.  It's less clear how this affects the performance of
cloud providers.  Tests done so far are unable to obtain stable enough
performance to provide meaningful benchmark results.  It is possible that
this could cause some performance regressions on more obscure filesystems,
so we may need to later provide users with some ability to get something
closer to the old behavior.  For now, let's leave that until we see that
it's really required.

Author: Thomas Munro, David Rowley
Reviewed-by: Ranier Vilela, Soumyadeep Chakraborty, Robert Haas
Reviewed-by: Amit Kapila, Kirk Jamison
Discussion: https://postgr.es/m/CA+hUKGJ_EErDv41YycXcbMbCBkztA34+z1ts9VQH+ACRuvpxig@mail.gmail.com
2020-07-26 21:02:45 +12:00

188 lines
6.6 KiB
C

/*-------------------------------------------------------------------------
*
* relscan.h
* POSTGRES relation scan descriptor definitions.
*
*
* Portions Copyright (c) 1996-2020, PostgreSQL Global Development Group
* Portions Copyright (c) 1994, Regents of the University of California
*
* src/include/access/relscan.h
*
*-------------------------------------------------------------------------
*/
#ifndef RELSCAN_H
#define RELSCAN_H
#include "access/htup_details.h"
#include "access/itup.h"
#include "port/atomics.h"
#include "storage/buf.h"
#include "storage/spin.h"
#include "utils/relcache.h"
struct ParallelTableScanDescData;
/*
* Generic descriptor for table scans. This is the base-class for table scans,
* which needs to be embedded in the scans of individual AMs.
*/
typedef struct TableScanDescData
{
/* scan parameters */
Relation rs_rd; /* heap relation descriptor */
struct SnapshotData *rs_snapshot; /* snapshot to see */
int rs_nkeys; /* number of scan keys */
struct ScanKeyData *rs_key; /* array of scan key descriptors */
/*
* Information about type and behaviour of the scan, a bitmask of members
* of the ScanOptions enum (see tableam.h).
*/
uint32 rs_flags;
void *rs_private; /* per-worker private memory for AM to use */
struct ParallelTableScanDescData *rs_parallel; /* parallel scan
* information */
} TableScanDescData;
typedef struct TableScanDescData *TableScanDesc;
/*
* Shared state for parallel table scan.
*
* Each backend participating in a parallel table scan has its own
* TableScanDesc in backend-private memory, and those objects all contain a
* pointer to this structure. The information here must be sufficient to
* properly initialize each new TableScanDesc as workers join the scan, and it
* must act as a information what to scan for those workers.
*/
typedef struct ParallelTableScanDescData
{
Oid phs_relid; /* OID of relation to scan */
bool phs_syncscan; /* report location to syncscan logic? */
bool phs_snapshot_any; /* SnapshotAny, not phs_snapshot_data? */
Size phs_snapshot_off; /* data for snapshot */
} ParallelTableScanDescData;
typedef struct ParallelTableScanDescData *ParallelTableScanDesc;
/*
* Shared state for parallel table scans, for block oriented storage.
*/
typedef struct ParallelBlockTableScanDescData
{
ParallelTableScanDescData base;
BlockNumber phs_nblocks; /* # blocks in relation at start of scan */
slock_t phs_mutex; /* mutual exclusion for setting startblock */
BlockNumber phs_startblock; /* starting block number */
pg_atomic_uint64 phs_nallocated; /* number of blocks allocated to
* workers so far. */
} ParallelBlockTableScanDescData;
typedef struct ParallelBlockTableScanDescData *ParallelBlockTableScanDesc;
/*
* Per backend state for parallel table scan, for block-oriented storage.
*/
typedef struct ParallelBlockTableScanWorkerData
{
uint64 phsw_nallocated; /* Current # of blocks into the scan */
uint32 phsw_chunk_remaining; /* # blocks left in this chunk */
uint32 phsw_chunk_size; /* The number of blocks to allocate in
* each I/O chunk for the scan */
} ParallelBlockTableScanWorkerData;
typedef struct ParallelBlockTableScanWorkerData *ParallelBlockTableScanWorker;
/*
* Base class for fetches from a table via an index. This is the base-class
* for such scans, which needs to be embedded in the respective struct for
* individual AMs.
*/
typedef struct IndexFetchTableData
{
Relation rel;
} IndexFetchTableData;
/*
* We use the same IndexScanDescData structure for both amgettuple-based
* and amgetbitmap-based index scans. Some fields are only relevant in
* amgettuple-based scans.
*/
typedef struct IndexScanDescData
{
/* scan parameters */
Relation heapRelation; /* heap relation descriptor, or NULL */
Relation indexRelation; /* index relation descriptor */
struct SnapshotData *xs_snapshot; /* snapshot to see */
int numberOfKeys; /* number of index qualifier conditions */
int numberOfOrderBys; /* number of ordering operators */
struct ScanKeyData *keyData; /* array of index qualifier descriptors */
struct ScanKeyData *orderByData; /* array of ordering op descriptors */
bool xs_want_itup; /* caller requests index tuples */
bool xs_temp_snap; /* unregister snapshot at scan end? */
/* signaling to index AM about killing index tuples */
bool kill_prior_tuple; /* last-returned tuple is dead */
bool ignore_killed_tuples; /* do not return killed entries */
bool xactStartedInRecovery; /* prevents killing/seeing killed
* tuples */
/* index access method's private state */
void *opaque; /* access-method-specific info */
/*
* In an index-only scan, a successful amgettuple call must fill either
* xs_itup (and xs_itupdesc) or xs_hitup (and xs_hitupdesc) to provide the
* data returned by the scan. It can fill both, in which case the heap
* format will be used.
*/
IndexTuple xs_itup; /* index tuple returned by AM */
struct TupleDescData *xs_itupdesc; /* rowtype descriptor of xs_itup */
HeapTuple xs_hitup; /* index data returned by AM, as HeapTuple */
struct TupleDescData *xs_hitupdesc; /* rowtype descriptor of xs_hitup */
ItemPointerData xs_heaptid; /* result */
bool xs_heap_continue; /* T if must keep walking, potential
* further results */
IndexFetchTableData *xs_heapfetch;
bool xs_recheck; /* T means scan keys must be rechecked */
/*
* When fetching with an ordering operator, the values of the ORDER BY
* expressions of the last returned tuple, according to the index. If
* xs_recheckorderby is true, these need to be rechecked just like the
* scan keys, and the values returned here are a lower-bound on the actual
* values.
*/
Datum *xs_orderbyvals;
bool *xs_orderbynulls;
bool xs_recheckorderby;
/* parallel index scan information, in shared memory */
struct ParallelIndexScanDescData *parallel_scan;
} IndexScanDescData;
/* Generic structure for parallel scans */
typedef struct ParallelIndexScanDescData
{
Oid ps_relid;
Oid ps_indexid;
Size ps_offset; /* Offset in bytes of am specific structure */
char ps_snapshot_data[FLEXIBLE_ARRAY_MEMBER];
} ParallelIndexScanDescData;
struct TupleTableSlot;
/* Struct for storage-or-index scans of system tables */
typedef struct SysScanDescData
{
Relation heap_rel; /* catalog being scanned */
Relation irel; /* NULL if doing heap scan */
struct TableScanDescData *scan; /* only valid in storage-scan case */
struct IndexScanDescData *iscan; /* only valid in index-scan case */
struct SnapshotData *snapshot; /* snapshot to unregister at end of scan */
struct TupleTableSlot *slot;
} SysScanDescData;
#endif /* RELSCAN_H */