postgresql/contrib/pg_buffercache/pg_buffercache_pages.c
Andres Freund 48354581a4 Allow Pin/UnpinBuffer to operate in a lockfree manner.
Pinning/Unpinning a buffer is a very frequent operation; especially in
read-mostly cache resident workloads. Benchmarking shows that in various
scenarios the spinlock protecting a buffer header's state becomes a
significant bottleneck. The problem can be reproduced with pgbench -S on
larger machines, but can be considerably worse for queries which touch
the same buffers over and over at a high frequency (e.g. nested loops
over a small inner table).

To allow atomic operations to be used, cram BufferDesc's flags,
usage_count, buf_hdr_lock, refcount into a single 32bit atomic variable;
that allows to manipulate them together using 32bit compare-and-swap
operations. This requires reducing MAX_BACKENDS to 2^18-1 (which could
be lifted by using a 64bit field, but it's not a realistic configuration
atm).

As not all operations can easily implemented in a lockfree manner,
implement the previous buf_hdr_lock via a flag bit in the atomic
variable. That way we can continue to lock the header in places where
it's needed, but can get away without acquiring it in the more frequent
hot-paths.  There's some additional operations which can be done without
the lock, but aren't in this patch; but the most important places are
covered.

As bufmgr.c now essentially re-implements spinlocks, abstract the delay
logic from s_lock.c into something more generic. It now has already two
users, and more are coming up; there's a follupw patch for lwlock.c at
least.

This patch is based on a proof-of-concept written by me, which Alexander
Korotkov made into a fully working patch; the committed version is again
revised by me.  Benchmarking and testing has, amongst others, been
provided by Dilip Kumar, Alexander Korotkov, Robert Haas.

On a large x86 system improvements for readonly pgbench, with a high
client count, of a factor of 8 have been observed.

Author: Alexander Korotkov and Andres Freund
Discussion: 2400449.GjM57CE0Yg@dinodell
2016-04-10 20:12:32 -07:00

253 lines
7.7 KiB
C

/*-------------------------------------------------------------------------
*
* pg_buffercache_pages.c
* display some contents of the buffer cache
*
* contrib/pg_buffercache/pg_buffercache_pages.c
*-------------------------------------------------------------------------
*/
#include "postgres.h"
#include "access/htup_details.h"
#include "catalog/pg_type.h"
#include "funcapi.h"
#include "storage/buf_internals.h"
#include "storage/bufmgr.h"
#define NUM_BUFFERCACHE_PAGES_MIN_ELEM 8
#define NUM_BUFFERCACHE_PAGES_ELEM 9
PG_MODULE_MAGIC;
/*
* Record structure holding the to be exposed cache data.
*/
typedef struct
{
uint32 bufferid;
Oid relfilenode;
Oid reltablespace;
Oid reldatabase;
ForkNumber forknum;
BlockNumber blocknum;
bool isvalid;
bool isdirty;
uint16 usagecount;
/*
* An int32 is sufficiently large, as MAX_BACKENDS prevents a buffer from
* being pinned by too many backends and each backend will only pin once
* because of bufmgr.c's PrivateRefCount infrastructure.
*/
int32 pinning_backends;
} BufferCachePagesRec;
/*
* Function context for data persisting over repeated calls.
*/
typedef struct
{
TupleDesc tupdesc;
BufferCachePagesRec *record;
} BufferCachePagesContext;
/*
* Function returning data from the shared buffer cache - buffer number,
* relation node/tablespace/database/blocknum and dirty indicator.
*/
PG_FUNCTION_INFO_V1(pg_buffercache_pages);
Datum
pg_buffercache_pages(PG_FUNCTION_ARGS)
{
FuncCallContext *funcctx;
Datum result;
MemoryContext oldcontext;
BufferCachePagesContext *fctx; /* User function context. */
TupleDesc tupledesc;
TupleDesc expected_tupledesc;
HeapTuple tuple;
if (SRF_IS_FIRSTCALL())
{
int i;
funcctx = SRF_FIRSTCALL_INIT();
/* Switch context when allocating stuff to be used in later calls */
oldcontext = MemoryContextSwitchTo(funcctx->multi_call_memory_ctx);
/* Create a user function context for cross-call persistence */
fctx = (BufferCachePagesContext *) palloc(sizeof(BufferCachePagesContext));
/*
* To smoothly support upgrades from version 1.0 of this extension
* transparently handle the (non-)existence of the pinning_backends
* column. We unfortunately have to get the result type for that... -
* we can't use the result type determined by the function definition
* without potentially crashing when somebody uses the old (or even
* wrong) function definition though.
*/
if (get_call_result_type(fcinfo, NULL, &expected_tupledesc) != TYPEFUNC_COMPOSITE)
elog(ERROR, "return type must be a row type");
if (expected_tupledesc->natts < NUM_BUFFERCACHE_PAGES_MIN_ELEM ||
expected_tupledesc->natts > NUM_BUFFERCACHE_PAGES_ELEM)
elog(ERROR, "incorrect number of output arguments");
/* Construct a tuple descriptor for the result rows. */
tupledesc = CreateTemplateTupleDesc(expected_tupledesc->natts, false);
TupleDescInitEntry(tupledesc, (AttrNumber) 1, "bufferid",
INT4OID, -1, 0);
TupleDescInitEntry(tupledesc, (AttrNumber) 2, "relfilenode",
OIDOID, -1, 0);
TupleDescInitEntry(tupledesc, (AttrNumber) 3, "reltablespace",
OIDOID, -1, 0);
TupleDescInitEntry(tupledesc, (AttrNumber) 4, "reldatabase",
OIDOID, -1, 0);
TupleDescInitEntry(tupledesc, (AttrNumber) 5, "relforknumber",
INT2OID, -1, 0);
TupleDescInitEntry(tupledesc, (AttrNumber) 6, "relblocknumber",
INT8OID, -1, 0);
TupleDescInitEntry(tupledesc, (AttrNumber) 7, "isdirty",
BOOLOID, -1, 0);
TupleDescInitEntry(tupledesc, (AttrNumber) 8, "usage_count",
INT2OID, -1, 0);
if (expected_tupledesc->natts == NUM_BUFFERCACHE_PAGES_ELEM)
TupleDescInitEntry(tupledesc, (AttrNumber) 9, "pinning_backends",
INT4OID, -1, 0);
fctx->tupdesc = BlessTupleDesc(tupledesc);
/* Allocate NBuffers worth of BufferCachePagesRec records. */
fctx->record = (BufferCachePagesRec *) palloc(sizeof(BufferCachePagesRec) * NBuffers);
/* Set max calls and remember the user function context. */
funcctx->max_calls = NBuffers;
funcctx->user_fctx = fctx;
/* Return to original context when allocating transient memory */
MemoryContextSwitchTo(oldcontext);
/*
* To get a consistent picture of the buffer state, we must lock all
* partitions of the buffer map. Needless to say, this is horrible
* for concurrency. Must grab locks in increasing order to avoid
* possible deadlocks.
*/
for (i = 0; i < NUM_BUFFER_PARTITIONS; i++)
LWLockAcquire(BufMappingPartitionLockByIndex(i), LW_SHARED);
/*
* Scan through all the buffers, saving the relevant fields in the
* fctx->record structure.
*/
for (i = 0; i < NBuffers; i++)
{
BufferDesc *bufHdr;
uint32 buf_state;
bufHdr = GetBufferDescriptor(i);
/* Lock each buffer header before inspecting. */
buf_state = LockBufHdr(bufHdr);
fctx->record[i].bufferid = BufferDescriptorGetBuffer(bufHdr);
fctx->record[i].relfilenode = bufHdr->tag.rnode.relNode;
fctx->record[i].reltablespace = bufHdr->tag.rnode.spcNode;
fctx->record[i].reldatabase = bufHdr->tag.rnode.dbNode;
fctx->record[i].forknum = bufHdr->tag.forkNum;
fctx->record[i].blocknum = bufHdr->tag.blockNum;
fctx->record[i].usagecount = BUF_STATE_GET_USAGECOUNT(buf_state);
fctx->record[i].pinning_backends = BUF_STATE_GET_REFCOUNT(buf_state);
if (buf_state & BM_DIRTY)
fctx->record[i].isdirty = true;
else
fctx->record[i].isdirty = false;
/* Note if the buffer is valid, and has storage created */
if ((buf_state & BM_VALID) && (buf_state & BM_TAG_VALID))
fctx->record[i].isvalid = true;
else
fctx->record[i].isvalid = false;
UnlockBufHdr(bufHdr, buf_state);
}
/*
* And release locks. We do this in reverse order for two reasons:
* (1) Anyone else who needs more than one of the locks will be trying
* to lock them in increasing order; we don't want to release the
* other process until it can get all the locks it needs. (2) This
* avoids O(N^2) behavior inside LWLockRelease.
*/
for (i = NUM_BUFFER_PARTITIONS; --i >= 0;)
LWLockRelease(BufMappingPartitionLockByIndex(i));
}
funcctx = SRF_PERCALL_SETUP();
/* Get the saved state */
fctx = funcctx->user_fctx;
if (funcctx->call_cntr < funcctx->max_calls)
{
uint32 i = funcctx->call_cntr;
Datum values[NUM_BUFFERCACHE_PAGES_ELEM];
bool nulls[NUM_BUFFERCACHE_PAGES_ELEM];
values[0] = Int32GetDatum(fctx->record[i].bufferid);
nulls[0] = false;
/*
* Set all fields except the bufferid to null if the buffer is unused
* or not valid.
*/
if (fctx->record[i].blocknum == InvalidBlockNumber ||
fctx->record[i].isvalid == false)
{
nulls[1] = true;
nulls[2] = true;
nulls[3] = true;
nulls[4] = true;
nulls[5] = true;
nulls[6] = true;
nulls[7] = true;
/* unused for v1.0 callers, but the array is always long enough */
nulls[8] = true;
}
else
{
values[1] = ObjectIdGetDatum(fctx->record[i].relfilenode);
nulls[1] = false;
values[2] = ObjectIdGetDatum(fctx->record[i].reltablespace);
nulls[2] = false;
values[3] = ObjectIdGetDatum(fctx->record[i].reldatabase);
nulls[3] = false;
values[4] = ObjectIdGetDatum(fctx->record[i].forknum);
nulls[4] = false;
values[5] = Int64GetDatum((int64) fctx->record[i].blocknum);
nulls[5] = false;
values[6] = BoolGetDatum(fctx->record[i].isdirty);
nulls[6] = false;
values[7] = Int16GetDatum(fctx->record[i].usagecount);
nulls[7] = false;
/* unused for v1.0 callers, but the array is always long enough */
values[8] = Int32GetDatum(fctx->record[i].pinning_backends);
nulls[8] = false;
}
/* Build and return the tuple. */
tuple = heap_form_tuple(fctx->tupdesc, values, nulls);
result = HeapTupleGetDatum(tuple);
SRF_RETURN_NEXT(funcctx, result);
}
else
SRF_RETURN_DONE(funcctx);
}