Continuing work on multi-threading. r/w access to buffer manager getting close, but still buggy.

This commit is contained in:
Sears Russell 2004-07-20 03:40:57 +00:00
parent 490dd86c09
commit 0ce77903fb
11 changed files with 319 additions and 40 deletions

View file

@ -99,8 +99,8 @@ extern int errno;
#define lsn_t long
/*#define DEBUGGING
#define PROFILE_LATCHES*/
/*#define DEBUGGING */
#define PROFILE_LATCHES
#ifdef DEBUGGING
/** @todo Files that use DEBUG have to pull in stdio.h, which is a pain! */

View file

@ -68,6 +68,8 @@ terms specified in this license.
*/
static pthread_mutex_t lastFreepage_mutex;
pthread_mutex_t add_pending_mutex;
static unsigned int lastFreepage = 0;
/**
@ -77,7 +79,7 @@ static unsigned int lastFreepage = 0;
*/
Page * loadPage(int pageid);
pthread_cond_t addPendingOK;
int bufInit() {
@ -89,6 +91,8 @@ int bufInit() {
lastFreepage = 0;
pthread_mutex_init(&lastFreepage_mutex , NULL);
pthread_cond_init(&addPendingOK, NULL);
pthread_mutex_init(&add_pending_mutex, NULL);
return 0;
@ -121,13 +125,13 @@ Page * loadPage (int pageid) {
Page * lastRallocPage = 0;
/** @todo ralloc ignores it's xid parameter; change the interface? */
recordid ralloc(int xid, /*lsn_t lsn,*/ long size) {
recordid ret;
Page * p;
DEBUG("Rallocing record of size %ld\n", (long int)size);
/* DEBUG("Rallocing record of size %ld\n", (long int)size); */
assert(size < BLOB_THRESHOLD_SIZE || size == BLOB_SLOT);
@ -140,7 +144,7 @@ recordid ralloc(int xid, /*lsn_t lsn,*/ long size) {
unlock(p->loadlatch);
pthread_mutex_unlock(&lastFreepage_mutex);
DEBUG("alloced rid = {%d, %d, %ld}\n", ret.page, ret.slot, ret.size);
/* DEBUG("alloced rid = {%d, %d, %ld}\n", ret.page, ret.slot, ret.size); */
return ret;
}
@ -233,18 +237,35 @@ void setSlotType(int pageid, int slot, int type) {
*/
void addPendingEvent(int pageid){
Page * p = loadPage(pageid);
Page * p;
pthread_mutex_lock(&(p->pending_mutex));
p = loadPage(pageid);
assert(!(p->waiting));
pthread_mutex_lock(&add_pending_mutex);
while(p->waiting) {
pthread_mutex_unlock(&add_pending_mutex);
unlock(p->loadlatch);
DEBUG("B");
pthread_mutex_lock(&add_pending_mutex);
pthread_cond_wait(&addPendingOK, &add_pending_mutex);
pthread_mutex_unlock(&add_pending_mutex);
p = loadPage(pageid);
pthread_mutex_lock(&add_pending_mutex);
}
p->pending++;
pthread_mutex_unlock(&(p->pending_mutex));
pthread_mutex_unlock(&add_pending_mutex);
unlock(p->loadlatch);
}
/**
@ -262,12 +283,14 @@ void addPendingEvent(int pageid){
*/
void removePendingEvent(int pageid) {
Page * p = loadPage(pageid);
Page * p;
pthread_mutex_lock(&(p->pending_mutex));
p = loadPage(pageid);
pthread_mutex_lock(&(add_pending_mutex));
p->pending--;
assert(p->id == pageid);
assert(p->pending >= 0);
if(p->waiting && !p->pending) {
@ -275,9 +298,11 @@ void removePendingEvent(int pageid) {
pthread_cond_signal(&(p->noMorePending));
}
pthread_mutex_unlock(&(p->pending_mutex));
pthread_mutex_unlock(&(add_pending_mutex));
unlock(p->loadlatch);
}

View file

@ -114,7 +114,38 @@ int __lladd_pthread_mutex_destroy(lladd_pthread_mutex_t *mutex) {
*/
int __lladd_pthread_cond_wait(pthread_cond_t *cond, lladd_pthread_mutex_t *mutex,
char * file, int line, char * cond_name, char * mutex_name) {
return pthread_cond_wait(cond, &mutex->mutex);
int ret;
char * location;
int location_length;
profile_tuple * tup = pblHtLookup(mutex->lockpoints, mutex->last_acquired_at, strlen(mutex->last_acquired_at)+1);
released_lock(tup);
released_lock(&(mutex->tup));
free(mutex->last_acquired_at);
ret = pthread_cond_wait(cond, &mutex->mutex);
location_length = asprintf(&location, "%s %d", file, line);
tup = pblHtLookup(mutex->lockpoints, location, location_length+1);
mutex->last_acquired_at = location;
if(!tup) {
tup = malloc(sizeof(profile_tuple));
init_tuple(tup);
pblHtInsert(mutex->lockpoints, location, location_length+1, tup);
}
acquired_lock(&(mutex->tup), 0);
acquired_lock(tup, 0);
return ret;
}
int __lladd_pthread_cond_timedwait(pthread_cond_t *cond, lladd_pthread_mutex_t *mutex, void *abstime,

View file

@ -7,6 +7,8 @@
#include <lladd/bufferManager.h>
#include "../blobManager.h"
/**
@file
Implementation of Talloc() as an operation
This is a bit strange compared to other operations, as it happens
@ -21,6 +23,10 @@
too big of a deal, but it should be fixed someday. A more serious
problem results from crashes during blob allocation.
@ingroup OPERATIONS
$Id$
*/
static int operate(int xid, lsn_t lsn, recordid rid, const void * dat) {

View file

@ -600,7 +600,7 @@ Page *pageAlloc(int id) {
page->rwlatch = initlock();
page->loadlatch = initlock();
pthread_mutex_init(&page->pending_mutex, NULL);
/* pthread_mutex_init(&page->pending_mutex, NULL);*/
pthread_cond_init(&page->noMorePending, NULL);
page->memAddr = malloc(PAGE_SIZE);
@ -614,7 +614,8 @@ Page *pageAlloc(int id) {
pthread_mutex_unlock(&pageAllocMutex);
page->pending = 0;
page->waiting = 0;
return page;
}

View file

@ -125,8 +125,6 @@ typedef struct Page_s {
this properly, and there are no read-only functions for the
pending field. */
pthread_mutex_t pending_mutex; /* pthread_mutex_t */
pthread_cond_t noMorePending; /* pthread_cond_t */
int waiting;
@ -154,6 +152,8 @@ typedef struct Page_s {
int pending;
} Page;
extern pthread_cond_t addPendingOK;
/**
* initializes all the important variables needed in all the
* functions dealing with pages.

View file

@ -213,9 +213,10 @@ Page * getPage(int pageid, int locktype) {
ret = pblHtLookup(activePages, &pageid, sizeof(int));
// Unfortunately, this is a heuristic, as a race condition exists.
// (Until we obtain a readlock on ret, we have no way of knowing if
// we've gotten the correct page.)
/* Unfortunately, this is a heuristic, as a race condition exists.
(Until we obtain a readlock on ret, we have no way of knowing if
we've gotten the correct page.) */
if(ret) {
cacheHitOnPage(ret);
assert(ret->id == -1 || ret->id == pageid);
@ -227,7 +228,7 @@ Page * getPage(int pageid, int locktype) {
ret = dummy_page;
}
writelock(ret->loadlatch, 217);
readlock(ret->loadlatch, 217);
while(ret->id != pageid) { /* Either we got a stale mapping from the HT, or no mapping at all. */
@ -264,7 +265,7 @@ Page * getPage(int pageid, int locktype) {
pthread_mutex_unlock(&loadPagePtr_mutex);
writelock(ret->loadlatch, 217);
readlock(ret->loadlatch, 217);
}

View file

@ -12,6 +12,10 @@
#include "logger/logWriter.h"
static FILE * stable = NULL;
/** Defined in bufferManager.c */
extern pthread_mutex_t add_pending_mutex;
/**
This function blocks until there are no events pending for this page.
@ -21,15 +25,16 @@ static FILE * stable = NULL;
void finalize(Page * p) {
pthread_mutex_lock(&(p->pending_mutex));
pthread_mutex_lock(&(add_pending_mutex));
p->waiting++;
while(p->pending) {
pthread_cond_wait(&(p->noMorePending), &(p->pending_mutex));
DEBUG("A");
pthread_cond_wait(&(p->noMorePending), &(add_pending_mutex));
}
pthread_mutex_unlock(&(p->pending_mutex));
pthread_mutex_unlock(&(add_pending_mutex));
pthread_cond_broadcast(&addPendingOK);
return;
}
@ -90,6 +95,8 @@ void pageWrite(Page * ret) {
long pageoffset = ret->id * PAGE_SIZE;
long offset ;
assert(ret->pending == 0);
if(flushedLSN() < pageReadLSN(ret)) {
DEBUG("pageWrite is calling syncLog()!\n");
syncLog();

View file

@ -1,6 +1,6 @@
#include <config.h>
#include <lladd/common.h>
#include "latches.h"
#include <lladd/transactional.h>
#include <lladd/recovery.h>
@ -14,6 +14,16 @@
TransactionLog XactionTable[MAX_TRANSACTIONS];
int numActiveXactions = 0;
int xidCount = 0;
/**
Locking for transactional2.c works as follows:
numActiveXactions, xidCount are protected, XactionTable is not.
This implies that we do not support multi-threaded transactions,
at least for now.
*/
pthread_mutex_t transactional_2_mutex;
#define INVALID_XTABLE_XID -1
/** Needed for debugging -- sometimes we don't want to run all of Tinit() */
@ -33,6 +43,8 @@ void setupOperationsTable() {
int Tinit() {
pthread_mutex_init(&transactional_2_mutex, NULL);
setupOperationsTable();
/* pageInit(); */
@ -42,6 +54,7 @@ int Tinit() {
InitiateRecovery();
return 0;
}
@ -49,6 +62,9 @@ int Tinit() {
int Tbegin() {
int i, index = 0;
int xidCount_tmp;
pthread_mutex_lock(&transactional_2_mutex);
if( numActiveXactions == MAX_TRANSACTIONS )
return EXCEED_MAX_TRANSACTIONS;
@ -63,16 +79,26 @@ int Tbegin() {
}
}
xidCount_tmp = xidCount;
/* Don't want to block while we're logging... */
pthread_mutex_unlock(&transactional_2_mutex);
assert( i < MAX_TRANSACTIONS );
XactionTable[index] = LogTransBegin(xidCount);
XactionTable[index] = LogTransBegin(xidCount_tmp);
return XactionTable[index].xid;
}
void Tupdate(int xid, recordid rid, const void *dat, int op) {
LogEntry * e;
#ifdef DEBUGGING
pthread_mutex_lock(&transactional_2_mutex);
assert(numActiveXactions <= MAX_TRANSACTIONS);
pthread_mutex_unlock(&transactional_2_mutex);
#endif
e = LogUpdate(&XactionTable[xid % MAX_TRANSACTIONS], rid, op, dat);
assert(XactionTable[xid % MAX_TRANSACTIONS].prevLSN == e->LSN);
@ -89,12 +115,20 @@ void Tread(int xid, recordid rid, void * dat) {
int Tcommit(int xid) {
lsn_t lsn;
#ifdef DEBUGGING
pthread_mutex_lock(&transactional_2_mutex);
assert(numActiveXactions <= MAX_TRANSACTIONS);
pthread_mutex_unlock(&transactional_2_mutex);
#endif
lsn = LogTransCommit(&XactionTable[xid % MAX_TRANSACTIONS]);
bufTransCommit(xid, lsn); /* unlocks pages */
XactionTable[xid%MAX_TRANSACTIONS].xid = INVALID_XTABLE_XID;
pthread_mutex_lock(&transactional_2_mutex);
numActiveXactions--;
assert( numActiveXactions >= 0 );
pthread_mutex_unlock(&transactional_2_mutex);
return 0;
}
@ -107,9 +141,11 @@ int Tabort(int xid) {
bufTransAbort(xid, lsn);
XactionTable[xid%MAX_TRANSACTIONS].xid = INVALID_XTABLE_XID;
numActiveXactions--;
pthread_mutex_lock(&transactional_2_mutex);
numActiveXactions--;
assert( numActiveXactions >= 0 );
pthread_mutex_unlock(&transactional_2_mutex);
return 0;
}
@ -141,10 +177,14 @@ void Trevive(int xid, long lsn) {
} else {
XactionTable[index].xid = xid;
XactionTable[index].prevLSN = lsn;
pthread_mutex_lock(&transactional_2_mutex);
numActiveXactions++;
pthread_mutex_unlock(&transactional_2_mutex);
}
}
void TsetXIDCount(int xid) {
pthread_mutex_lock(&transactional_2_mutex);
xidCount = xid;
pthread_mutex_unlock(&transactional_2_mutex);
}

View file

@ -17,45 +17,104 @@
#define LOG_NAME "check_bufferMananger.log"
#define NUM_PAGES 1000
#define THREAD_COUNT 5
#define READS_PER_THREAD 50000
#define THREAD_COUNT 25
#define READS_PER_THREAD 10000
#define RECORDS_PER_THREAD 10000
#define RECORD_THREAD_COUNT 25
void initializePages() {
int i;
printf("Initialization starting\n"); fflush(NULL);
for(i = 0 ; i < NUM_PAGES; i++) {
Page * p;
recordid rid;
rid.page = i;
rid.slot = 0;
rid.size = sizeof(int);
p = loadPage(rid.page);
assert(p->id != -1);
pageSlotRalloc(p, 0, rid);
/* addPendingEvent(rid.page); */
writeRecord(1, 1, rid, &i);
/* removePendingEvent(rid.page); */
assert(p->pending == 0);
unlock(p->loadlatch);
}
printf("Initialization complete.\n"); fflush(NULL);
}
void * workerThread(void * p) {
int i;
for(i = 0 ; i < READS_PER_THREAD; i++) {
recordid rid;
int j;
int k = (int) (((double)NUM_PAGES)*rand()/(RAND_MAX+1.0));
if(! (i % 5000) ) {
printf("%d", i / 5000); fflush(NULL);
if(! (i % 500) ) {
printf("%d", i / 500); fflush(NULL);
}
rid.page = k;
rid.slot = 0;
rid.size = sizeof(int);
addPendingEvent(rid.page);
readRecord(1, rid, &j);
assert(rid.page == k);
removePendingEvent(rid.page);
assert(k == j);
}
return NULL;
}
void * workerThreadWriting(void * p) {
int offset = *(int*)p;
recordid rids[RECORDS_PER_THREAD];
for(int i = 0 ; i < RECORDS_PER_THREAD; i++) {
rids[i] = ralloc(1, sizeof(int));
}
for(int i = 0; i < RECORDS_PER_THREAD; i++) {
int val = i + offset;
int oldpage = rids[i].page;
addPendingEvent(rids[i].page);
writeRecord(1, 0, rids[i], &val);
assert(oldpage == rids[i].page);
removePendingEvent(rids[i].page);
if(! (i % 1000) ) {
printf("W%d", i / 1000); fflush(NULL);
}
}
for(int i = 0; i < RECORDS_PER_THREAD; i++) {
int val;
addPendingEvent(rids[i].page);
readRecord(1, rids[i], &val);
if(! (i % 1000) ) {
printf("R%d", i / 1000); fflush(NULL);
}
assert(val == i+offset);
removePendingEvent(rids[i].page);
}
return NULL;
}
START_TEST(pageSingleThreadTest) {
Tinit();
@ -101,15 +160,43 @@ START_TEST(pageLoadTest) {
Tdeinit();
} END_TEST
START_TEST(pageSingleThreadWriterTest) {
int i = 100;
Tinit();
workerThreadWriting(&i);
Tdeinit();
}END_TEST
START_TEST(pageThreadedWritersTest) {
pthread_t workers[RECORD_THREAD_COUNT];
int i;
Tinit();
for(i = 0; i < RECORD_THREAD_COUNT; i++) {
pthread_create(&workers[i], NULL, workerThreadWriting, &i);
}
for(i = 0; i < RECORD_THREAD_COUNT; i++) {
pthread_join(workers[i], NULL);
}
Tdeinit();
}END_TEST
Suite * check_suite(void) {
Suite *s = suite_create("logWriter");
Suite *s = suite_create("bufferManager");
/* Begin a new test */
TCase *tc = tcase_create("writeNew");
TCase *tc = tcase_create("multithreaded");
/* Sub tests are added, one per line, here */
/*tcase_add_test(tc, pageSingleThreadTest); */
tcase_add_test(tc, pageSingleThreadTest);
tcase_add_test(tc, pageLoadTest);
tcase_add_test(tc, pageSingleThreadWriterTest);
tcase_add_test(tc, pageThreadedWritersTest);
/* --------------------------------------------- */

View file

@ -39,12 +39,63 @@ permission to use and distribute the software in accordance with the
terms specified in this license.
---*/
#include <config.h>
#include <lladd/common.h>
#include <../../src/lladd/latches.h>
#include <check.h>
#include <assert.h>
#include <lladd/transactional.h>
#include "../check_includes.h"
#define LOG_NAME "check_transactional2.log"
#define THREAD_COUNT 25
#define RECORDS_PER_THREAD 10000
/** Allocate a bunch of stuff, set it, read it, commit it, and read it again. */
void * writingWorkerThread ( void * v ) {
int offset = * (int *) v;
recordid * rids = malloc(RECORDS_PER_THREAD * sizeof(recordid));
int xid = Tbegin();
for(int i = 0; i < RECORDS_PER_THREAD; i++) {
rids[i] = Talloc(xid, sizeof(int));
if(! (i %100)) {
printf("A%d", i/100);fflush(NULL);
}
}
for(int i = 0; i < RECORDS_PER_THREAD; i++) {
int tmp = i + offset;
Tset(xid, rids[i], &tmp);
if(! (i %100)) {
printf("W%d", i/100); fflush(NULL);
}
}
for(int i = 0; i < RECORDS_PER_THREAD; i++) {
int j;
Tread(xid, rids[i], &j);
assert(i + offset == j);
if(! (i %100)) {
printf("R%d", i/100);fflush(NULL);
}
}
Tcommit(xid);
xid = Tbegin();
for(int i = 0; i < RECORDS_PER_THREAD; i++) {
int j;
Tread(xid, rids[i], &j);
assert(i + offset == j);
}
return NULL;
}
/**
Assuming that the Tset() operation is implemented correctly, checks
that doUpdate, redoUpdate and undoUpdate are working correctly, for
@ -142,6 +193,33 @@ START_TEST(transactional_blobSmokeTest) {
}
END_TEST
START_TEST(transactional_nothreads_commit) {
int five = 5;
Tinit();
writingWorkerThread(&five);
Tdeinit();
} END_TEST
START_TEST(transactional_threads_commit) {
pthread_t workers[THREAD_COUNT];
int i;
Tinit();
for(i = 0; i < THREAD_COUNT; i++) {
int arg = i + 100;
pthread_create(&workers[i], NULL, writingWorkerThread, &arg);
}
for(i = 0; i < THREAD_COUNT; i++) {
pthread_join(workers[i], NULL);
}
Tdeinit();
} END_TEST
/**
Add suite declarations here
*/
@ -153,6 +231,9 @@ Suite * check_suite(void) {
/* Sub tests are added, one per line, here */
tcase_add_test(tc, transactional_smokeTest);
tcase_add_test(tc, transactional_blobSmokeTest);
tcase_add_test(tc, transactional_nothreads_commit);
tcase_add_test(tc, transactional_threads_commit);
/** @todo still need to make blobs reentrant! */
/* --------------------------------------------- */
tcase_add_checked_fixture(tc, setup, teardown);
suite_add_tcase(s, tc);