Inside PostgreSQL Shared Memory

Inside PostgreSQL Shared Memory

BRUCE MOMJIAN,ENTERPRISEDB

January, 2009

AbstractPOSTGRESQL is an open-source, full-featured relational database.This presentation gives an overview of the shared memorystructures used by Postgres.

Creative Commons Attribution License http://momjian.us/presentations

Outline

1. File storage format

2. Shared memory creation

3. Shared buffers

4. Row value access

5. Locking

6. Other structures

Inside PostgreSQL Shared Memory 1

File System /data

Postgres

Postgres

Postgres

/data


File System /data/base

Postgres

Postgres

Postgres

/data

/pg_clog/pg_multixact/pg_subtrans/pg_tblspc

/pg_xlog

/global

/pg_twophase

/base


File System /data/base/db

Postgres

Postgres

Postgres

/data /base /16385 (production)

/1 (template1)

/17982 (devel)/16821 (test)

/21452 (marketing)


File System /data/base/db/table

Postgres

Postgres

Postgres

/data /base /16385 /24692 (customer)

/27214 (order)/25932 (product)/25952 (employee)/27839 (part)


File System Data Pages

Postgres

Postgres

Postgres

8k 8k 8k 8k

/data /base /16385 /24692


Data Pages

Postgres

Postgres

Postgres

Page Header Item Item Item

Tuple

Tuple Tuple Special

8K

8k 8k 8k 8k

/data /base /16385 /24692


File System Block Tuple

Postgres

Postgres

Postgres


Tuple

Tuple Tuple Special

8K

8k 8k 8k 8k

/data /base /16385 /24692

Tuple


File System Tuple

hoff − length of tuple header

infomask − tuple flags

natts − number of attributes

ctid − tuple id (page / item)

cmax − destruction command id

xmin − creation transaction id

xmax − destruction transaction id

cmin − creation command id

bits − bit map representing NULLs

OID − object id of tuple (optional)

Tuple

Value Value ValueValue Value Value ValueHeader

int4in(’9241’)

textout()

’Martin’


Tuple Header C Structures

typedef struct HeapTupleFields{ TransactionId t_xmin; /* inserting xact ID */ TransactionId t_xmax; /* deleting or locking xact ID */

union { CommandId t_cid; /* inserting or deleting command ID, or both */ TransactionId t_xvac; /* VACUUM FULL xact ID */ } t_field3;} HeapTupleFields;

typedef struct HeapTupleHeaderData{ union { HeapTupleFields t_heap; DatumTupleFields t_datum; } t_choice;

ItemPointerData t_ctid; /* current TID of this or newer tuple */

/* Fields below here must match MinimalTupleData! */

uint16 t_infomask2; /* number of attributes + various flags */

uint16 t_infomask; /* various flag bits, see below */

uint8 t_hoff; /* sizeof header incl. bitmap, padding */

/* ^ − 23 bytes − ^ */

bits8 t_bits [ 1] ; /* bitmap of NULLs −− VARIABLE LENGTH */

/* MORE DATA FOLLOWS AT END OF STRUCT */} HeapTupleHeaderData;


Shared Memory Creation

postmaster postgres postgres

Program (Text)

Data

Program (Text)

Data

Shared Memory

Program (Text)

Data

Shared Memory Shared Memory

Stack Stack Stack

fork()


Shared Memory

Shared Buffers

Proc Array

PROC

Multi−XACT Buffers

Two−Phase Structs

Subtrans Buffers

CLOG Buffers

XLOG Buffers

Shared Invalidation

Lightweight Locks

Lock Hashes

Auto Vacuum

Btree Vacuum

Free Space Map

Buffer Descriptors

Background Writer Synchronized Scan

Semaphores

Statistics

LOCK

PROCLOCK


Shared Buffers


Tuple

Tuple Tuple Special

8KPostgres

Postgres

Postgres

8k 8k 8k 8k

/data /base /16385 /24692

Shared Buffers

LWLock − for page changes

Pin Count − prevent page replacement

read()

write()

8k 8k 8k

Buffer Descriptors


HeapTuples

Shared Buffers

PostgresHeapTuple

C pointer

hoff − length of tuple header

infomask − tuple flags

natts − number of attributes

ctid − tuple id (page / item)

cmax − destruction command id

xmin − creation transaction id

xmax − destruction transaction id

cmin − creation command id

bits − bit map representing NULLs

OID − object id of tuple (optional)

Tuple

Value Value ValueValue Value Value ValueHeader

int4in(’9241’)

textout()

’Martin’


Tuple

Tuple Tuple Special

8K

8k 8k 8k


Finding A Tuple Value in CDatumnocachegetattr ( HeapTuple tuple, int attnum, TupleDesc tupleDesc, bool * isnull ){ HeapTupleHeader tup = tuple −>t_data; Form_pg_attribute * att = tupleDesc −>attrs;

{ int i;

/* * Note − This loop is a little tricky. For each non−null attribute, * we have to first account for alignment padding before the attr, * then advance over the attr based on its length. Nulls have no * storage and no alignment padding either. We can use/set * attcacheoff until we reach either a null or a var−width attribute. */ off = 0; for ( i = 0;; i ++) /* loop exit is at "break" */ { if ( HeapTupleHasNulls ( tuple ) && att_isnull ( i, bp )) continue; /* this cannot be the target att */

if ( att [ i ]−> attlen == −1) off = att_align_pointer ( off, att [ i ]−> attalign, −1, tp + off ) ; else /* not varlena, so safe to use att_align_nominal */ off = att_align_nominal ( off, att [ i ]−> attalign ) ;

if ( i == attnum ) break;

off = att_addlength_pointer ( off, att [ i ]−> attlen, tp + off ) ; } }

return fetchatt ( att [ attnum ] , tp + off ) ;}


Value Access in C

#define fetch_att(T,attbyval,attlen) $ \ (attbyval) ? \ ( \ (attlen) == (int) sizeof(int32) ? \ Int32GetDatum(*((int32 *)(T))) \ : \ ( \ (attlen) == (int) sizeof(int16) ? \ Int16GetDatum(*((int16 *)(T))) \ : \ ( \ AssertMacro((attlen) == 1), \ CharGetDatum(*((char *)(T))) \ ) \ ) \ ) \ : \ PointerGetDatum((char *) (T)) $


Test And Set LockCan Succeed Or Fail

0/1

1

0

Success

Was 0 on exchange

Lock already taken

Was 1 on exchange

Failure

1

1


Test And Set Lockx86 Assembler

static __inline__ inttas ( volatile slock_t * lock ){ register slock_t _res = 1;

/* * Use a non−locking test before asserting the bus lock. Note that the * extra test appears to be a small loss on some x86 platforms and a small * win on others; it’s by no means clear that we should keep it. */ __asm__ __volatile__ ( " cmpb $0,%1 \n" " jne 1f \n" " lock \n" " xchgb %0,%1 \n" "1: \n": "+q"( _res ) , "+m"(* lock ):: "memory", "cc") ; return ( int) _res;}


Spin LockAlways Succeeds

0/1

1

Success

Was 0 on exchange

Failure

Was 1 on exchange

Lock already taken

Sleep of increasing duration

0 1

1

Spinlocks are designed for short-lived locking operations, like access tocontrol structures. They are not be used to protect code that makeskernel calls or other heavy operations.Inside PostgreSQL Shared Memory 19

Light Weight Locks

Shared Buffers

Proc Array

PROC


Two−Phase Structs

Subtrans Buffers

CLOG Buffers

XLOG Buffers

Shared Invalidation

Lightweight Locks

Lock Hashes

Auto Vacuum

Btree Vacuum

Free Space Map


Semaphores

Statistics

LOCK

PROCLOCK

Buffer Descriptors

Sleep On Lock

Light weight locks attempt to acquire the lock, and go to sleep on asemaphore if the lock request fails. Spinlocks control access to the lightweight lock control structure.


Database Object Locks

PROCLOCKPROC LOCK

Lock Hashes


Proc

Proc Array

PROC

used usedusedempty empty empty


Other Shared Memory Structures

Shared Buffers

Proc Array

PROC


Two−Phase Structs

Subtrans Buffers

CLOG Buffers

XLOG Buffers

Shared Invalidation

Lightweight Locks

Lock Hashes

Auto Vacuum

Btree Vacuum

Free Space Map

Buffer Descriptors


Semaphores

Statistics

LOCK

PROCLOCK


Conclusion

Pink Floyd: Wish You Were HereInside PostgreSQL Shared Memory 24

Technology

Inside PostgreSQL Shared Memory