#include "dstm.h"
+#include "ip.h"
#include "clookup.h"
+#include "machinepile.h"
#include "mlookup.h"
#include "llookup.h"
#include "plookup.h"
-#include<pthread.h>
-#include<sys/types.h>
-#include<sys/socket.h>
-#include<netdb.h>
-#include<netinet/in.h>
+#include "prelookup.h"
+#include "threadnotify.h"
+#include "queue.h"
+#include <pthread.h>
+#include <sys/types.h>
+#include <sys/socket.h>
+#include <netdb.h>
+#include <netinet/in.h>
+#include <sys/types.h>
+#include <unistd.h>
+#include <errno.h>
+#include <time.h>
+#include <string.h>
#define LISTEN_PORT 2156
-#define MACHINE_IP "127.0.0.1"
#define RECEIVE_BUFFER_SIZE 2048
+#define NUM_THREADS 10
+#define PREFETCH_CACHE_SIZE 1048576 //1MB
+#define CONFIG_FILENAME "dstm.conf"
+/* Global Variables */
extern int classsize[];
+extern primarypfq_t pqueue; //Shared prefetch queue
+extern mcpileq_t mcqueue; //Shared queue containing prefetch requests sorted by remote machineids
+objstr_t *prefetchcache; //Global Prefetch cache
+pthread_mutex_t prefetchcache_mutex;// Mutex to lock Prefetch Cache
+pthread_mutexattr_t prefetchcache_mutex_attr; /* Attribute for lock to make it a recursive lock */
+extern pthread_mutex_t mainobjstore_mutex;// Mutex to lock main Object store
+extern prehashtable_t pflookup; //Global Prefetch cache's lookup table
+pthread_t wthreads[NUM_THREADS]; //Worker threads for working on the prefetch queue
+pthread_t tPrefetch; /* Primary Prefetch thread that processes the prefetch queue */
+extern objstr_t *mainobjstore;
+unsigned int myIpAddr;
+unsigned int *hostIpAddrs;
+int sizeOfHostArray;
+int numHostsInSystem;
+int myIndexInHostArray;
+unsigned int oidsPerBlock;
+unsigned int oidMin;
+unsigned int oidMax;
+void printhex(unsigned char *, int);
+plistnode_t *createPiles(transrecord_t *);
+
+void printhex(unsigned char *ptr, int numBytes)
+{
+ int i;
+ for (i = 0; i < numBytes; i++)
+ {
+ if (ptr[i] < 16)
+ printf("0%x ", ptr[i]);
+ else
+ printf("%x ", ptr[i]);
+ }
+ printf("\n");
+ return;
+}
+
+inline int arrayLength(int *array) {
+ int i;
+ for(i=0 ;array[i] != -1; i++)
+ ;
+ return i;
+}
+inline int findmax(int *array, int arraylength) {
+ int max, i;
+ max = array[0];
+ for(i = 0; i < arraylength; i++){
+ if(array[i] > max) {
+ max = array[i];
+ }
+ }
+ return max;
+}
+/* This function is a prefetch call generated by the compiler that
+ * populates the shared primary prefetch queue*/
+void prefetch(int ntuples, unsigned int *oids, unsigned short *endoffsets, short *arrayfields) {
+ int qnodesize;
+ int len = 0;
+ int i;
+
+ /* Allocate for the queue node*/
+ char *node;
+ if(ntuples > 0) {
+ qnodesize = sizeof(prefetchqelem_t) + sizeof(int) + ntuples * (sizeof(short) + sizeof(unsigned int)) + endoffsets[ntuples - 1] * sizeof(short);
+ if((node = calloc(1, qnodesize)) == NULL) {
+ printf("Calloc Error %s, %d\n", __FILE__, __LINE__);
+ return;
+ }
+ /* Set queue node values */
+ len = sizeof(prefetchqelem_t);
+ memcpy(node + len, &ntuples, sizeof(int));
+ len += sizeof(int);
+ memcpy(node + len, oids, ntuples*sizeof(unsigned int));
+ len += ntuples * sizeof(unsigned int);
+ memcpy(node + len, endoffsets, ntuples*sizeof(short));
+ len += ntuples * sizeof(short);
+ memcpy(node + len, arrayfields, endoffsets[ntuples-1]*sizeof(short));
+
+ /* Lock and insert into primary prefetch queue */
+ pthread_mutex_lock(&pqueue.qlock);
+ pre_enqueue((prefetchqelem_t *)node);
+ pthread_cond_signal(&pqueue.qcond);
+ pthread_mutex_unlock(&pqueue.qlock);
+ }
+}
+
+/* This function starts up the transaction runtime. */
+int dstmStartup(const char * option) {
+ pthread_t thread_Listen;
+ pthread_attr_t attr;
+ int master=option!=NULL && strcmp(option, "master")==0;
+
+ if (processConfigFile() != 0)
+ return 0; //TODO: return error value, cause main program to exit
+
+ dstmInit();
+ transInit();
+
+ if (master) {
+ pthread_attr_init(&attr);
+ pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED);
+ pthread_create(&thread_Listen, &attr, dstmListen, NULL);
+ return 1;
+ } else {
+ dstmListen();
+ return 0;
+ }
+
+}
+
+//TODO Use this later
+void *pCacheAlloc(objstr_t *store, unsigned int size) {
+ void *tmp;
+ objstr_t *ptr;
+ ptr = store;
+ int success = 0;
+
+ while(ptr->next != NULL) {
+ /* check if store is empty */
+ if(((unsigned int)ptr->top - (unsigned int)ptr - sizeof(objstr_t) + size) <= ptr->size) {
+ tmp = ptr->top;
+ ptr->top += size;
+ success = 1;
+ return tmp;
+ } else {
+ ptr = ptr-> next;
+ }
+ }
+
+ if(success == 0) {
+ return NULL;
+ }
+}
+
+/* This function initiates the prefetch thread
+ * A queue is shared between the main thread of execution
+ * and the prefetch thread to process the prefetch call
+ * Call from compiler populates the shared queue with prefetch requests while prefetch thread
+ * processes the prefetch requests */
+void transInit() {
+ int t, rc;
+ int retval;
+ //Create and initialize prefetch cache structure
+ prefetchcache = objstrCreate(PREFETCH_CACHE_SIZE);
+ //prefetchcache->next = objstrCreate(PREFETCH_CACHE_SIZE);
+ //prefetchcache->next->next = objstrCreate(PREFETCH_CACHE_SIZE);
+
+ /* Initialize attributes for mutex */
+ pthread_mutexattr_init(&prefetchcache_mutex_attr);
+ pthread_mutexattr_settype(&prefetchcache_mutex_attr, PTHREAD_MUTEX_RECURSIVE_NP);
+
+ pthread_mutex_init(&prefetchcache_mutex, &prefetchcache_mutex_attr);
+
+ //Create prefetch cache lookup table
+ if(prehashCreate(HASH_SIZE, LOADFACTOR))
+ return; //Failure
+ //Initialize primary shared queue
+ queueInit();
+ //Initialize machine pile w/prefetch oids and offsets shared queue
+ mcpileqInit();
+ //Create the primary prefetch thread
+
+ do {
+ retval=pthread_create(&tPrefetch, NULL, transPrefetch, NULL);
+ } while(retval!=0);
+
+ pthread_detach(tPrefetch);
+
+ //Create and Initialize a pool of threads
+ /* Threads are active for the entire period runtime is running */
+ for(t = 0; t< NUM_THREADS; t++) {
+ do {
+ rc = pthread_create(&wthreads[t], NULL, mcqProcess, (void *)t);
+ } while(rc!=0);
+ pthread_detach(wthreads[t]);
+ }
+}
+
+/* This function stops the threads spawned */
+void transExit() {
+ int t;
+ pthread_cancel(tPrefetch);
+ for(t = 0; t < NUM_THREADS; t++)
+ pthread_cancel(wthreads[t]);
+
+ return;
+}
+
+/* This functions inserts randowm wait delays in the order of msec
+ * Mostly used when transaction commits retry*/
+void randomdelay()
+{
+ struct timespec req;
+ time_t t;
+
+ t = time(NULL);
+ req.tv_sec = 0;
+ req.tv_nsec = (long)(1000000 + (t%10000000)); //1-11 msec
+ nanosleep(&req, NULL);
+ return;
+}
+
+/* This function initializes things required in the transaction start*/
transrecord_t *transStart()
{
- transrecord_t *tmp = malloc(sizeof(transrecord_t));
+ transrecord_t *tmp = calloc(1, sizeof(transrecord_t));
tmp->cache = objstrCreate(1048576);
tmp->lookupTable = chashCreate(HASH_SIZE, LOADFACTOR);
+#ifdef COMPILER
+ tmp->revertlist=NULL;
+#endif
return tmp;
}
-objheader_t *transRead(transrecord_t *record, unsigned int oid)
-{
+/* This function finds the location of the objects involved in a transaction
+ * and returns the pointer to the object if found in a remote location */
+objheader_t *transRead(transrecord_t *record, unsigned int oid) {
unsigned int machinenumber;
objheader_t *tmp, *objheader;
- void *objcopy;
- int size;
+ objheader_t *objcopy;
+ int size, rc, found = 0;
void *buf;
- //check cache
- if((objheader =(objheader_t *)chashSearch(record->lookupTable, oid)) != NULL){
- return(objheader);
+ struct timespec ts;
+ struct timeval tp;
+
+ if(oid == 0) {
+ return NULL;
+ }
+
+ rc = gettimeofday(&tp, NULL);
+
+ /* 1ms delay */
+ tp.tv_usec += 1000;
+ if (tp.tv_usec >= 1000000)
+ {
+ tp.tv_usec -= 1000000;
+ tp.tv_sec += 1;
+ }
+ /* Convert from timeval to timespec */
+ ts.tv_sec = tp.tv_sec;
+ ts.tv_nsec = tp.tv_usec * 1000;
+
+ /* Search local transaction cache */
+ if((objheader = (objheader_t *)chashSearch(record->lookupTable, oid)) != NULL){
+#ifdef COMPILER
+ return &objheader[1];
+#else
+ return objheader;
+#endif
} else if ((objheader = (objheader_t *) mhashSearch(oid)) != NULL) {
- //Look up in Machine lookup table and found
- printf("oid not found in local cache\n");
- tmp = mhashSearch(oid);
- size = sizeof(objheader_t)+classsize[tmp->type];
- //Copy into cache
+ /* Look up in machine lookup table and copy into cache*/
+ GETSIZE(size, objheader);
+ size += sizeof(objheader_t);
+ objcopy = objstrAlloc(record->cache, size);
+ memcpy(objcopy, (void *)objheader, size);
+ /* Insert into cache's lookup table */
+ chashInsert(record->lookupTable, OID(objheader), objcopy);
+#ifdef COMPILER
+ return &objcopy[1];
+#else
+ return objcopy;
+#endif
+ } else if((tmp = (objheader_t *) prehashSearch(oid)) != NULL) { /* Look up in prefetch cache */
+ GETSIZE(size, tmp);
+ size+=sizeof(objheader_t);
objcopy = objstrAlloc(record->cache, size);
memcpy(objcopy, (void *)tmp, size);
- //Insert into cache's lookup table
- chashInsert(record->lookupTable, objheader->oid, objcopy);
- return(objcopy);
+ /* Insert into cache's lookup table */
+ chashInsert(record->lookupTable, OID(tmp), objcopy);
+#ifdef COMPILER
+ return &objcopy[1];
+#else
+ return objcopy;
+#endif
} else {
- //Get the object from the remote location
- //printf("oid not found in local machine lookup\n");
+ /*If object not found in prefetch cache then block until object appears in the prefetch cache */
+ pthread_mutex_lock(&pflookup.lock);
+ while(!found) {
+ rc = pthread_cond_timedwait(&pflookup.cond, &pflookup.lock, &ts);
+ /* Check Prefetch cache again */
+ if((tmp =(objheader_t *) prehashSearch(oid)) != NULL) {
+ found = 1;
+ GETSIZE(size,tmp);
+ size+=sizeof(objheader_t);
+ objcopy = objstrAlloc(record->cache, size);
+ memcpy(objcopy, (void *)tmp, size);
+ chashInsert(record->lookupTable, OID(tmp), objcopy);
+ pthread_mutex_unlock(&pflookup.lock);
+#ifdef COMPILER
+ return &objcopy[1];
+#else
+ return objcopy;
+#endif
+ } else if (rc == ETIMEDOUT) {
+ printf("Wait timed out\n");
+ pthread_mutex_unlock(&pflookup.lock);
+ break;
+ }
+ }
+
+ /* Get the object from the remote location */
machinenumber = lhashSearch(oid);
objcopy = getRemoteObj(record, machinenumber, oid);
- if(objcopy == NULL)
- //If object is not found in Remote location
- printf("Object not found in Machine %d\n", machinenumber);
- else
- return(objcopy);
- }
+ if(objcopy == NULL) {
+ printf("Object not found in Remote location %s, %d\n", __FILE__, __LINE__);
+ return NULL;
+ } else {
+#ifdef COMPILER
+ return &objcopy[1];
+#else
+ return objcopy;
+#endif
+ }
+ }
}
-objheader_t *transCreateObj(transrecord_t *record, unsigned short type)
+/* This function creates objects in the transaction record */
+objheader_t *transCreateObj(transrecord_t *record, unsigned int size)
{
- objheader_t *tmp = (objheader_t *) objstrAlloc(record->cache, (sizeof(objheader_t) + classsize[type]));
- tmp->oid = getNewOID();
- tmp->type = type;
- tmp->version = 1;
- tmp->rcount = 0; //? not sure how to handle this yet
- tmp->status = 0;
- tmp->status |= NEW;
- chashInsert(record->lookupTable, tmp->oid, tmp);
- return tmp;
+ objheader_t *tmp = (objheader_t *) objstrAlloc(record->cache, (sizeof(objheader_t) + size));
+ tmp->notifylist = NULL;
+ OID(tmp) = getNewOID();
+ tmp->version = 1;
+ tmp->rcount = 1;
+ STATUS(tmp) = NEW;
+ chashInsert(record->lookupTable, OID(tmp), tmp);
+
+#ifdef COMPILER
+ return &tmp[1]; //want space after object header
+#else
+ return tmp;
+#endif
}
-//TODO Change the values inside write() to exact size of the message being sent
-//int decideResponse(thread_data_array_t *tdata, char *control, int sd) {
-int decideResponse(thread_data_array_t *tdata, int sd) {
- int i, j, n, N, sum, oidcount = 0, transagree = 0, transdisagree = 0, transsoftabort = 0, transmiss = 0;
- char ctrl, control, *ptr;
- unsigned int *oidnotfound[tdata->pilecount];
- objheader_t *header;
- //Check common data structure
- for (i = 0 ; i < tdata->pilecount ; i++) {
- //Switch case
- control = tdata->recvmsg[i].rcv_status;
- switch(control) {
- case TRANS_DISAGREE:
- printf("DEBUG-> Inside TRANS_DISAGREE\n");
- transdisagree++;
- //Free transaction records
- objstrDelete(tdata->rec->cache);
- chashDelete(tdata->rec->lookupTable);
- free(tdata->rec);
- //send Abort
- ctrl = TRANS_ABORT;
- if (write(sd, &ctrl, sizeof(char)) < 0) {
- perror("Error sending ctrl message for participant\n");
- return 1;
- }
- break;
-
- case TRANS_AGREE:
- printf("DEBUG-> Inside TRANS_AGREE\n");
- transagree++;
- break;
-
- case TRANS_SOFT_ABORT:
- printf("DEBUG-> Inside TRANS_SOFT_ABORT\n");
- transsoftabort++;
- //Read list of objects missing
- if(read(sd, &oidcount, sizeof(int)) != 0) {
- if (oidcount == 0) {
- sleep(2);
- break;
- }
- N = oidcount * sizeof(unsigned int);
- if((oidnotfound[i] = calloc(oidcount, sizeof(unsigned int))) == NULL) {
- printf("Calloc error %s, %d\n", __FILE__, __LINE__);
- }
- ptr = (char *) oidnotfound[i];
- do {
- n = read(sd, ptr+sum, N-sum);
- sum += n;
- } while(sum < N && n !=0);
- }
+/* This function creates machine piles based on all machines involved in a
+ * transaction commit request */
+plistnode_t *createPiles(transrecord_t *record) {
+ int i = 0;
+ unsigned int size;/* Represents number of bins in the chash table */
+ chashlistnode_t *curr, *ptr, *next;
+ plistnode_t *pile = NULL;
+ unsigned int machinenum;
+ void *localmachinenum;
+ objheader_t *headeraddr;
+
+ ptr = record->lookupTable->table;
+ size = record->lookupTable->size;
+
+ for(i = 0; i < size ; i++) {
+ curr = &ptr[i];
+ /* Inner loop to traverse the linked list of the cache lookupTable */
+ while(curr != NULL) {
+ //if the first bin in hash table is empty
+ if(curr->key == 0) {
break;
-
- case TRANS_AGREE_BUT_MISSING_OBJECTS:
- printf("DEBUG-> Inside TRANS_AGREE_BUT_MISSING_OBJECTS\n");
- transmiss++;
- //Read list of objects missing
- read(sd, &oidcount, sizeof(int));
- N = oidcount * sizeof(unsigned int);
- if((oidnotfound[i] = calloc(oidcount, sizeof(unsigned int))) == NULL) {
- printf("Calloc error %s, %d\n", __FILE__, __LINE__);
- }
- ptr = (char *) oidnotfound[i];
- do {
- n = read(sd, ptr+sum, N-sum);
- sum += n;
- } while(sum < N && n !=0);
+ }
+ next = curr->next;
+
+ if ((headeraddr = chashSearch(record->lookupTable, curr->key)) == NULL) {
+ printf("Error: No such oid %s, %d\n", __FILE__, __LINE__);
+ return NULL;
+ }
+ //Get machine location for object id (and whether local or not)
+ if (STATUS(headeraddr) & NEW || mhashSearch(curr->key) != NULL) {
+ machinenum = myIpAddr;
+ } else if ((machinenum = lhashSearch(curr->key)) == 0) {
+ printf("Error: No such machine %s, %d\n", __FILE__, __LINE__);
+ return NULL;
+ }
- break;
- default:
- printf("Participant sent unknown message\n");
- }
- }
-
- //For Debug purposes
- for(i=0 ; i< tdata->pilecount; i++) {
- for(j=0 ; j < oidcount; j++) {
- printf("DEBUG-> Oid %d missing for pilecount: %d\n", oidnotfound[j], i+1);
+ //Make machine groups
+ if ((pile = pInsert(pile, headeraddr, machinenum, record->lookupTable->numelements)) == NULL) {
+ printf("pInsert error %s, %d\n", __FILE__, __LINE__);
+ return NULL;
+ }
+
+ curr = next;
}
}
+ return pile;
+}
+
+/* This function initiates the transaction commit process
+ * Spawns threads for each of the new connections with Participants
+ * and creates new piles by calling the createPiles(),
+ * Sends a transrequest() to each remote machines for objects found remotely
+ * and calls handleLocalReq() to process objects found locally */
+int transCommit(transrecord_t *record) {
+ unsigned int tot_bytes_mod, *listmid;
+ plistnode_t *pile, *pile_ptr;
+ int i, j, rc, val;
+ int pilecount, offset, threadnum, trecvcount;
+ char buffer[RECEIVE_BUFFER_SIZE],control;
+ char transid[TID_LEN];
+ trans_req_data_t *tosend;
+ trans_commit_data_t transinfo;
+ static int newtid = 0;
+ char treplyctrl = 0, treplyretry = 0; /* keeps track of the common response that needs to be sent */
+ char localstat = 0;
+
+ do {
+ trecvcount = 0;
+ threadnum = 0;
+ treplyretry = 0;
+
+ /* Look through all the objects in the transaction record and make piles
+ * for each machine involved in the transaction*/
+ pile_ptr = pile = createPiles(record);
+
+ /* Create the packet to be sent in TRANS_REQUEST */
+
+ /* Count the number of participants */
+ pilecount = pCount(pile);
- //Decide what control message to send to Participant
- if(transagree == tdata->pilecount){
- //Send Commit
- ctrl = TRANS_COMMIT;
- printf("Sending TRANS_COMMIT\n");
- if (write(sd, &ctrl, sizeof(char)) < 0) {
- perror("Error sending ctrl message for participant\n");
+ /* Create a list of machine ids(Participants) involved in transaction */
+ if((listmid = calloc(pilecount, sizeof(unsigned int))) == NULL) {
+ printf("Calloc error %s, %d\n", __FILE__, __LINE__);
+ //free(record);
return 1;
- }
- }
+ }
+ pListMid(pile, listmid);
- if(transsoftabort > 0 && transdisagree == 0) {
- //Send abort but retry commit
- ctrl = TRANS_ABORT_BUT_RETRY_COMMIT;
- printf("Sending TRANS_ABORT_BUT_RETRY_COMMIT\n");
- if (write(sd, &ctrl, sizeof(char)) < 0) {
- perror("Error sending ctrl message for participant\n");
+
+ /* Initialize thread variables,
+ * Spawn a thread for each Participant involved in a transaction */
+ pthread_t thread[pilecount];
+ pthread_attr_t attr;
+ pthread_cond_t tcond;
+ pthread_mutex_t tlock;
+ pthread_mutex_t tlshrd;
+
+ thread_data_array_t *thread_data_array;
+ if((thread_data_array = (thread_data_array_t *) calloc(pilecount, sizeof(thread_data_array_t))) == NULL) {
+ printf("Malloc error %s, %d\n", __FILE__, __LINE__);
+ pthread_cond_destroy(&tcond);
+ pthread_mutex_destroy(&tlock);
+ pDelete(pile_ptr);
+ free(listmid);
+ //free(record);
return 1;
}
- //TODO
- //Relookup all missing objects
- //set up a new connection readClientReq()
- //rebuilt the pile and llookup table
- //i.e. wait at the participant end and then resend either agree or disagree
- }
- if(transmiss > 0 && transsoftabort == 0 && transdisagree == 0) {
- //Send abort but retry commit
- ctrl = TRANS_ABORT_BUT_RETRY_COMMIT_WITH_RELOCATING;
- printf("Sending TRANS_ABORT_BUT_RETRY_COMMIT_WITH_RELOCATING\n");
- if (write(sd, &ctrl, sizeof(char)) < 0) {
- perror("Error sending ctrl message for participant\n");
+
+ local_thread_data_array_t *ltdata;
+ if((ltdata = calloc(1, sizeof(local_thread_data_array_t))) == NULL) {
+ printf("Calloc error %s, %d\n", __FILE__, __LINE__);
+ pthread_cond_destroy(&tcond);
+ pthread_mutex_destroy(&tlock);
+ pDelete(pile_ptr);
+ free(listmid);
+ free(thread_data_array);
+ //free(record);
return 1;
}
- //TODO
- //Relookup all missing objects
- //set up a new connection readClientReq()
- //rebuilt the pile and llookup table
- //send missing mising object/ objects
- }
-
- //Free pointers
- for(i=0 ; i< tdata->pilecount; i++) {
- free(oidnotfound[i]);
- }
-
+ thread_response_t rcvd_control_msg[pilecount]; /* Shared thread array that keeps track of responses of participants */
+
+ /* Initialize and set thread detach attribute */
+ pthread_attr_init(&attr);
+ pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_JOINABLE);
+ pthread_mutex_init(&tlock, NULL);
+ pthread_cond_init(&tcond, NULL);
+
+ /* Process each machine pile */
+ while(pile != NULL) {
+ //Create transaction id
+ newtid++;
+ if ((tosend = calloc(1, sizeof(trans_req_data_t))) == NULL) {
+ printf("Calloc error %s, %d\n", __FILE__, __LINE__);
+ pthread_cond_destroy(&tcond);
+ pthread_mutex_destroy(&tlock);
+ pDelete(pile_ptr);
+ free(listmid);
+ free(thread_data_array);
+ free(ltdata);
+ //free(record);
+ return 1;
+ }
+ tosend->f.control = TRANS_REQUEST;
+ sprintf(tosend->f.trans_id, "%x_%d", pile->mid, newtid);
+ tosend->f.mcount = pilecount;
+ tosend->f.numread = pile->numread;
+ tosend->f.nummod = pile->nummod;
+ tosend->f.numcreated = pile->numcreated;
+ tosend->f.sum_bytes = pile->sum_bytes;
+ tosend->listmid = listmid;
+ tosend->objread = pile->objread;
+ tosend->oidmod = pile->oidmod;
+ tosend->oidcreated = pile->oidcreated;
+ thread_data_array[threadnum].thread_id = threadnum;
+ thread_data_array[threadnum].mid = pile->mid;
+ thread_data_array[threadnum].buffer = tosend;
+ thread_data_array[threadnum].recvmsg = rcvd_control_msg;
+ thread_data_array[threadnum].threshold = &tcond;
+ thread_data_array[threadnum].lock = &tlock;
+ thread_data_array[threadnum].count = &trecvcount;
+ thread_data_array[threadnum].replyctrl = &treplyctrl;
+ thread_data_array[threadnum].replyretry = &treplyretry;
+ thread_data_array[threadnum].rec = record;
+ /* If local do not create any extra connection */
+ if(pile->mid != myIpAddr) { /* Not local */
+ do {
+ rc = pthread_create(&thread[threadnum], &attr, transRequest, (void *) &thread_data_array[threadnum]);
+ } while(rc!=0);
+ if(rc) {
+ perror("Error in pthread create\n");
+ pthread_cond_destroy(&tcond);
+ pthread_mutex_destroy(&tlock);
+ pDelete(pile_ptr);
+ free(listmid);
+ for (i = 0; i < threadnum; i++)
+ free(thread_data_array[i].buffer);
+ free(thread_data_array);
+ free(ltdata);
+ //free(record);
+ return 1;
+ }
+ } else { /*Local*/
+ ltdata->tdata = &thread_data_array[threadnum];
+ ltdata->transinfo = &transinfo;
+ do {
+ val = pthread_create(&thread[threadnum], &attr, handleLocalReq, (void *) ltdata);
+ } while(val!=0);
+ if(val) {
+ perror("Error in pthread create\n");
+ pthread_cond_destroy(&tcond);
+ pthread_mutex_destroy(&tlock);
+ pDelete(pile_ptr);
+ free(listmid);
+ for (i = 0; i < threadnum; i++)
+ free(thread_data_array[i].buffer);
+ free(thread_data_array);
+ free(ltdata);
+ //free(record);
+ return 1;
+ }
+ }
+
+ threadnum++;
+ pile = pile->next;
+ }
+
+ /* Free attribute and wait for the other threads */
+ pthread_attr_destroy(&attr);
+
+ for (i = 0; i < pilecount; i++) {
+ rc = pthread_join(thread[i], NULL);
+ if(rc)
+ {
+ printf("ERROR return code from pthread_join() is %d\n", rc);
+ pthread_cond_destroy(&tcond);
+ pthread_mutex_destroy(&tlock);
+ pDelete(pile_ptr);
+ free(listmid);
+ for (j = i; j < pilecount; j++)
+ free(thread_data_array[j].buffer);
+ free(thread_data_array);
+ free(ltdata);
+ //free(record);
+ return 1;
+ }
+ free(thread_data_array[i].buffer);
+ }
+
+ /* Free resources */
+ pthread_cond_destroy(&tcond);
+ pthread_mutex_destroy(&tlock);
+ free(listmid);
+ pDelete(pile_ptr);
+ free(thread_data_array);
+ free(ltdata);
+
+ /* wait a random amount of time before retrying to commit transaction*/
+ if(treplyretry == 1) {
+ randomdelay();
+ }
+
+ /* Retry trans commit procedure if not sucessful in the first try */
+ } while (treplyretry == 1);
+
+ /* Free Resources */
+ objstrDelete(record->cache);
+ chashDelete(record->lookupTable);
+ free(record);
return 0;
}
+/* This function sends information involved in the transaction request
+ * to participants and accepts a response from particpants.
+ * It calls decideresponse() to decide on what control message
+ * to send next to participants and sends the message using sendResponse()*/
void *transRequest(void *threadarg) {
int sd, i, n;
struct sockaddr_in serv_addr;
thread_data_array_t *tdata;
objheader_t *headeraddr;
char buffer[RECEIVE_BUFFER_SIZE], control, recvcontrol;
+ char machineip[16], retval;
tdata = (thread_data_array_t *) threadarg;
- printf("DEBUG -> New thread id %d\n", tdata->thread_id);
- //Send Trans Request
+
+ /* Send Trans Request */
if ((sd = socket(AF_INET, SOCK_STREAM, 0)) < 0) {
- perror("Error in socket for TRANS_REQUEST");
- return NULL;
+ perror("Error in socket for TRANS_REQUEST\n");
+ pthread_exit(NULL);
}
bzero((char*) &serv_addr, sizeof(serv_addr));
serv_addr.sin_family = AF_INET;
serv_addr.sin_port = htons(LISTEN_PORT);
- serv_addr.sin_addr.s_addr = inet_addr(MACHINE_IP);
- //serv_addr.sin_addr.s_addr = inet_addr(tdata->mid);
-
+ midtoIP(tdata->mid,machineip);
+ machineip[15] = '\0';
+ serv_addr.sin_addr.s_addr = inet_addr(machineip);
+ /* Open Connection */
if (connect(sd, (struct sockaddr *) &serv_addr, sizeof(struct sockaddr)) < 0) {
- perror("Error in connect for TRANS_REQUEST");
- return NULL;
+ perror("Error in connect for TRANS_REQUEST\n");
+ close(sd);
+ pthread_exit(NULL);
}
- //Multiple writes for sending packets of data
- //Send first few fixed bytes of the TRANS_REQUEST protocol
- printf("DEBUG -> Start sending commit data... %d\n", tdata->buffer->f.control);
-// printf("DEBUG-> Bytes sent in first write: %d\n", sizeof(fixed_data_t));
-// printf("Machine count = %d\tnumread = %d\tnummod = %d\tsum_bytes = %d\n", tdata->buffer->f.mcount, tdata->buffer->f.numread, tdata->buffer->f.nummod, tdata->buffer->f.sum_bytes);
- if (write(sd, &(tdata->buffer->f), (sizeof(fixed_data_t))) < 0) {
- perror("Error sending fixed bytes for thread");
- return NULL;
+ /* Send bytes of data with TRANS_REQUEST control message */
+ if (send(sd, &(tdata->buffer->f), sizeof(fixed_data_t),MSG_NOSIGNAL) < sizeof(fixed_data_t)) {
+ perror("Error sending fixed bytes for thread\n");
+ close(sd);
+ pthread_exit(NULL);
}
- //Send list of machines involved in the transaction
-// printf("DEBUG-> Bytes sent in second write: %d\n", sizeof(unsigned int) * tdata->pilecount);
- if (write(sd, tdata->buffer->listmid, (sizeof(unsigned int) * tdata->pilecount )) < 0) {
- perror("Error sending list of machines for thread");
- return NULL;
+ /* Send list of machines involved in the transaction */
+ {
+ int size=sizeof(unsigned int)*tdata->buffer->f.mcount;
+ if (send(sd, tdata->buffer->listmid, size, MSG_NOSIGNAL) < size) {
+ perror("Error sending list of machines for thread\n");
+ close(sd);
+ pthread_exit(NULL);
+ }
}
- //Send oids and version number tuples for objects that are read
-// printf("DEBUG-> Bytes sent in the third write: %d\n", (sizeof(unsigned int) + sizeof(short)) * tdata->buffer->f.numread);
-// printf(" DEBUG->Read oids are %d %x %d %d\n", *(tdata->buffer->objread), (tdata->buffer->objread + 6), *(tdata->buffer->objread + 12), *(tdata->buffer->objread +18));
- if (write(sd, tdata->buffer->objread, ((sizeof(unsigned int) + sizeof(short)) * tdata->buffer->f.numread )) < 0) {
- perror("Error sending tuples for thread");
- return NULL;
+ /* Send oids and version number tuples for objects that are read */
+ {
+ int size=(sizeof(unsigned int)+sizeof(short))*tdata->buffer->f.numread;
+ if (send(sd, tdata->buffer->objread, size, MSG_NOSIGNAL) < size) {
+ perror("Error sending tuples for thread\n");
+ close(sd);
+ pthread_exit(NULL);
+ }
}
- //Send objects that are modified
+ /* Send objects that are modified */
for(i = 0; i < tdata->buffer->f.nummod ; i++) {
+ int size;
headeraddr = chashSearch(tdata->rec->lookupTable, tdata->buffer->oidmod[i]);
-// printf("DEBUG -> Bytes sent for oid = %d modified %d\n", *((int *)headeraddr), sizeof(objheader_t) + classsize[headeraddr->type]);
- if (write(sd, headeraddr, sizeof(objheader_t) + classsize[headeraddr->type]) < 0) {
- perror("Error sending obj modified for thread");
- return NULL;
+ GETSIZE(size,headeraddr);
+ size+=sizeof(objheader_t);
+ if (send(sd, headeraddr, size, MSG_NOSIGNAL) < size) {
+ perror("Error sending obj modified for thread\n");
+ close(sd);
+ pthread_exit(NULL);
}
}
-
- //Read message control message from participant side
- n = read(sd, &control, sizeof(char));
+
+ /* Read control message from Participant */
+ if((n = read(sd, &control, sizeof(char))) <= 0) {
+ perror("Error in reading control message from Participant\n");
+ close(sd);
+ pthread_exit(NULL);
+ }
recvcontrol = control;
- printf("DEBUG -> After TRANS_REQUEST, message control recv is %d\n", recvcontrol);
-// while(n != 0) {
-// n = read(sd, buffer, sizeof(buffer));
-// }
-
- //Update common data structure and increment count
+
+ /* Update common data structure and increment count */
tdata->recvmsg[tdata->thread_id].rcv_status = recvcontrol;
- //Lock and update count
- //Thread sleeps until all messages from pariticipants are received by coordinator
+
+ /* Lock and update count */
+ /* Thread sleeps until all messages from pariticipants are received by coordinator */
pthread_mutex_lock(tdata->lock);
- (*(tdata->count))++;
-
- if(*(tdata->count) == tdata->pilecount) {
+
+ (*(tdata->count))++; /* keeps track of no of messages received by the coordinator */
+
+ /* Wake up the threads and invoke decideResponse (once) */
+ if(*(tdata->count) == tdata->buffer->f.mcount) {
+ decideResponse(tdata);
pthread_cond_broadcast(tdata->threshold);
- //process the participant's request
- if (decideResponse(tdata, sd) != 0) {
- printf("decideResponse returned error %s. %d\n", __FILE__, __LINE__);
- return NULL;
- }
} else {
pthread_cond_wait(tdata->threshold, tdata->lock);
- }
+ }
pthread_mutex_unlock(tdata->lock);
+
+ /* Send the final response such as TRANS_COMMIT or TRANS_ABORT t
+ * to all participants in their respective socket */
+ if (sendResponse(tdata, sd) == 0) {
+ printf("sendResponse returned error %s,%d\n", __FILE__, __LINE__);
+ close(sd);
+ pthread_exit(NULL);
+ }
+
+ /* Close connection */
close(sd);
pthread_exit(NULL);
}
-int transCommit(transrecord_t *record) {
- chashlistnode_t *curr, *ptr, *next;
- unsigned int size;//Represents number of bins in the chash table
- unsigned int machinenum, tot_bytes_mod, *listmid;
- objheader_t *headeraddr;
- plistnode_t *tmp, *pile = NULL;
- int i, rc;
- int pilecount = 0, offset, numthreads = 0, trecvcount = 0;
- char buffer[RECEIVE_BUFFER_SIZE],control;
- char transid[TID_LEN];
- static int newtid = 0;
+/* This function decides the reponse that needs to be sent to
+ * all Participant machines after the TRANS_REQUEST protocol */
+void decideResponse(thread_data_array_t *tdata) {
+ char control;
+ int i, transagree = 0, transdisagree = 0, transsoftabort = 0; /* Counters to formulate decision of what
+ message to send */
- ptr = record->lookupTable->table;
- size = record->lookupTable->size;
- //Look through all the objects in the cache and make piles
- for(i = 0; i < size ;i++) {
- curr = &ptr[i];
- //Inner loop to traverse the linked list of the cache lookupTable
- while(curr != NULL) {
- //if the first bin in hash table is empty
- if(curr->key == 0) {
+ for (i = 0 ; i < tdata->buffer->f.mcount; i++) {
+ control = tdata->recvmsg[i].rcv_status; /* tdata: keeps track of all participant responses
+ written onto the shared array */
+ switch(control) {
+ default:
+ printf("Participant sent unknown message in %s, %d\n", __FILE__, __LINE__);
+ /* treat as disagree, pass thru */
+ case TRANS_DISAGREE:
+ transdisagree++;
break;
- }
- next = curr->next;
- //Get machine location for object id
- /*
- if ((machinenum = lhashSearch(curr->key)) == 0) {
- printf("Error: No such machine\n");
- return 1;
- }
- */
- //TODO only for debug
- machinenum = 1;
- if ((headeraddr = chashSearch(record->lookupTable, curr->key)) == NULL) {
- printf("Error: No such oid\n");
- return 1;
- }
- //Make machine groups
- if ((pile = pInsert(pile, headeraddr, machinenum, record->lookupTable->numelements)) == NULL) {
- printf("pInsert error %s, %d\n", __FILE__, __LINE__);
- return 1;
- }
- curr = next;
- }
- }
- //Create the packet to be sent in TRANS_REQUEST
- tmp = pile;
- pilecount = pCount(pile); //Keeps track of the number of participants
-
- //Thread related variables
- pthread_t thread[pilecount]; //Create threads for each participant
- pthread_attr_t attr;
- pthread_cond_t tcond;
- pthread_mutex_t tlock;
- thread_data_array_t thread_data_array[pilecount];
-
- thread_response_t rcvd_control_msg[pilecount]; //Shared thread array that keeps track of responses of participants
-
- //Initialize and set thread detach attribute
- pthread_attr_init(&attr);
- pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_JOINABLE);
- pthread_mutex_init(&tlock, NULL);
- pthread_cond_init(&tcond, NULL);
-
- //Keep track of list of machine ids per transaction
- if((listmid = calloc(pilecount, sizeof(unsigned int))) == NULL) {
- printf("Calloc error %s, %d\n", __FILE__, __LINE__);
- return 1;
- }
-
- pListMid(pile, listmid);
- //Process each machine group
- while(tmp != NULL) {
- printf("DEBUG -> Created thread %d... \n", numthreads);
- //Create transaction id
- newtid++;
- trans_req_data_t *tosend;
- if ((tosend = calloc(1, sizeof(trans_req_data_t))) == NULL) {
- printf("Calloc error %s, %d\n", __FILE__, __LINE__);
- return 1;
- }
- tosend->f.control = TRANS_REQUEST;
- sprintf(tosend->f.trans_id, "%x_%d", tmp->mid, newtid);
- tosend->f.mcount = pilecount;
- tosend->f.numread = tmp->numread;
- tosend->f.nummod = tmp->nummod;
- tosend->f.sum_bytes = tmp->sum_bytes;
- tosend->listmid = listmid;
- tosend->objread = tmp->objread;
- tosend->oidmod = tmp->oidmod;
- thread_data_array[numthreads].thread_id = numthreads;
- thread_data_array[numthreads].mid = tmp->mid;
- thread_data_array[numthreads].pilecount = pilecount;
- thread_data_array[numthreads].buffer = tosend;
- thread_data_array[numthreads].recvmsg = rcvd_control_msg;
- thread_data_array[numthreads].threshold = &tcond;
- thread_data_array[numthreads].lock = &tlock;
- thread_data_array[numthreads].count = &trecvcount;
- thread_data_array[numthreads].rec = record;
-
- rc = pthread_create(&thread[numthreads], NULL, transRequest, (void *) &thread_data_array[numthreads]);
- if (rc) {
- perror("Error in pthread create");
- return 1;
- }
- numthreads++;
- //TODO frees
- free(tosend);
- tmp = tmp->next;
- }
+ case TRANS_AGREE:
+ transagree++;
+ break;
- // Free attribute and wait for the other threads
- pthread_attr_destroy(&attr);
- for (i = 0 ;i < pilecount ; i++) {
- rc = pthread_join(thread[i], NULL);
- if (rc)
- {
- printf("ERROR return code from pthread_join() is %d\n", rc);
- return 1;
+ case TRANS_SOFT_ABORT:
+ transsoftabort++;
+ break;
}
}
-
- //Free resources
- pthread_cond_destroy(&tcond);
- pthread_mutex_destroy(&tlock);
- free(listmid);
- pDelete(pile);
- return 0;
-}
-int transSoftAbort(transrecord_t *record){
+ if(transdisagree > 0) {
+ /* Send Abort */
+ *(tdata->replyctrl) = TRANS_ABORT;
+ *(tdata->replyretry) = 0;
+ /* clear objects from prefetch cache */
+ for (i = 0; i < tdata->buffer->f.numread; i++)
+ prehashRemove(*(unsigned int *)(tdata->buffer->objread + (sizeof(unsigned int) + sizeof(unsigned short))*i));
+ for (i = 0; i < tdata->buffer->f.nummod; i++)
+ prehashRemove(tdata->buffer->oidmod[i]);
+ } else if(transagree == tdata->buffer->f.mcount){
+ /* Send Commit */
+ *(tdata->replyctrl) = TRANS_COMMIT;
+ *(tdata->replyretry) = 0;
+ } else {
+ /* Send Abort in soft abort case followed by retry commiting transaction again*/
+ *(tdata->replyctrl) = TRANS_ABORT;
+ *(tdata->replyretry) = 1;
+ }
+ return;
}
+/* This function sends the final response to remote machines per thread in their respective socket id
+ * It returns a char that is only needed to check the correctness of execution of this function inside
+ * transRequest()*/
+char sendResponse(thread_data_array_t *tdata, int sd) {
+ int n, N, sum, oidcount = 0;
+ char *ptr, retval = 0;
+ unsigned int *oidnotfound;
-int transAbort(transrecord_t *record){
+ /* If the decided response is due to a soft abort and missing objects at the Participant's side */
+ if(tdata->recvmsg[tdata->thread_id].rcv_status == TRANS_SOFT_ABORT) {
+ /* Read list of objects missing */
+ if((read(sd, &oidcount, sizeof(int)) != 0) && (oidcount != 0)) {
+ N = oidcount * sizeof(unsigned int);
+ if((oidnotfound = calloc(oidcount, sizeof(unsigned int))) == NULL) {
+ printf("Calloc error %s, %d\n", __FILE__, __LINE__);
+ return 0;
+ }
+ ptr = (char *) oidnotfound;
+ do {
+ n = read(sd, ptr+sum, N-sum);
+ sum += n;
+ } while(sum < N && n !=0);
+ }
+ retval = TRANS_SOFT_ABORT;
+ }
+ /* If the decided response is TRANS_ABORT */
+ if(*(tdata->replyctrl) == TRANS_ABORT) {
+ retval = TRANS_ABORT;
+ } else if(*(tdata->replyctrl) == TRANS_COMMIT) { /* If the decided response is TRANS_COMMIT */
+ retval = TRANS_COMMIT;
+ }
+ if (send(sd, tdata->replyctrl, sizeof(char),MSG_NOSIGNAL) < sizeof(char)) {
+ perror("Error sending ctrl message for participant\n");
+ }
+ return retval;
}
-//mnun will be used to represent the machine IP address later
+/* This function opens a connection, places an object read request to the
+ * remote machine, reads the control message and object if available and
+ * copies the object and its header to the local cache.
+ * */
+
void *getRemoteObj(transrecord_t *record, unsigned int mnum, unsigned int oid) {
- int sd, size;
+ int sd, size, val;
struct sockaddr_in serv_addr;
struct hostent *server;
- char buffer[RECEIVE_BUFFER_SIZE],control;
+ char control;
+ char machineip[16];
objheader_t *h;
void *objcopy;
if ((sd = socket(AF_INET, SOCK_STREAM, 0)) < 0) {
- perror("Error in socket");
+ perror("Error in socket\n");
return NULL;
}
+
bzero((char*) &serv_addr, sizeof(serv_addr));
serv_addr.sin_family = AF_INET;
serv_addr.sin_port = htons(LISTEN_PORT);
- serv_addr.sin_addr.s_addr = inet_addr(MACHINE_IP);
+ midtoIP(mnum,machineip);
+ machineip[15] = '\0';
+ serv_addr.sin_addr.s_addr = inet_addr(machineip);
+ /* Open connection */
if (connect(sd, (struct sockaddr *) &serv_addr, sizeof(struct sockaddr)) < 0) {
- perror("Error in connect");
+ perror("Error in connect\n");
return NULL;
}
- bzero((char *)buffer,sizeof(buffer));
- control = READ_REQUEST;
- buffer[0] = control;
- memcpy(buffer+1, &oid, sizeof(int));
- if (write(sd, buffer, sizeof(int) + 1) < 0) {
- perror("Error sending message");
+ char readrequest[sizeof(char)+sizeof(unsigned int)];
+ readrequest[0] = READ_REQUEST;
+ *((unsigned int *)(&readrequest[1])) = oid;
+ if (send(sd, &readrequest, sizeof(readrequest), MSG_NOSIGNAL) < sizeof(readrequest)) {
+ perror("Error sending message\n");
return NULL;
}
-#ifdef DEBUG1
- printf("DEBUG -> ready to rcv ...\n");
-#endif
- //Read response from the Participant
- read(sd, &control, sizeof(char));
+ /* Read response from the Participant */
+ if((val = read(sd, &control, sizeof(char))) <= 0) {
+ perror("No control response for getRemoteObj sent\n");
+ return NULL;
+ }
switch(control) {
case OBJECT_NOT_FOUND:
return NULL;
- break;
case OBJECT_FOUND:
- read(sd, &size, sizeof(int));
+ /* Read object if found into local cache */
+ if((val = read(sd, &size, sizeof(int))) <= 0) {
+ perror("No size is read from the participant\n");
+ return NULL;
+ }
objcopy = objstrAlloc(record->cache, size);
- read(sd, objcopy, size);
+ if((val = read(sd, objcopy, size)) <= 0) {
+ perror("No objects are read from the remote participant\n");
+ return NULL;
+ }
+ /* Insert into cache's lookup table */
+ chashInsert(record->lookupTable, oid, objcopy);
break;
default:
- printf("Error in recv request from participant on a READ_REQUEST\n");
+ printf("Error in recv request from participant on a READ_REQUEST %s, %d\n",__FILE__, __LINE__);
return NULL;
}
-
-#if 0
- read(sd, buffer, sizeof(buffer));
+ /* Close connection */
close(sd);
- if (buffer[0] == OBJECT_NOT_FOUND) {
- return NULL;
+ return objcopy;
+}
+
+/* This function handles the local objects involved in a transaction commiting process.
+ * It also makes a decision if this local machine sends AGREE or DISAGREE or SOFT_ABORT to coordinator.
+ * Note Coordinator = local machine
+ * It wakes up the other threads from remote participants that are waiting for the coordinator's decision and
+ * based on common agreement it either commits or aborts the transaction.
+ * It also frees the memory resources */
+void *handleLocalReq(void *threadarg) {
+ unsigned int *oidnotfound = NULL, *oidlocked = NULL;
+ local_thread_data_array_t *localtdata;
+ int objnotfound = 0, objlocked = 0;
+ int v_nomatch = 0, v_matchlock = 0, v_matchnolock = 0;
+ int numread, i;
+ unsigned int oid;
+ unsigned short version;
+ void *mobj;
+ objheader_t *headptr;
+
+ localtdata = (local_thread_data_array_t *) threadarg;
+
+ /* Counters and arrays to formulate decision on control message to be sent */
+ oidnotfound = (unsigned int *) calloc((localtdata->tdata->buffer->f.numread + localtdata->tdata->buffer->f.nummod), sizeof(unsigned int));
+ oidlocked = (unsigned int *) calloc((localtdata->tdata->buffer->f.numread + localtdata->tdata->buffer->f.nummod), sizeof(unsigned int));
+
+ numread = localtdata->tdata->buffer->f.numread;
+ /* Process each oid in the machine pile/ group per thread */
+ for (i = 0; i < localtdata->tdata->buffer->f.numread + localtdata->tdata->buffer->f.nummod; i++) {
+ if (i < localtdata->tdata->buffer->f.numread) {
+ int incr = sizeof(unsigned int) + sizeof(short);// Offset that points to next position in the objread array
+ incr *= i;
+ oid = *((unsigned int *)(localtdata->tdata->buffer->objread + incr));
+ version = *((short *)(localtdata->tdata->buffer->objread + incr + sizeof(unsigned int)));
+ } else { // Objects Modified
+ int tmpsize;
+ headptr = (objheader_t *) chashSearch(localtdata->tdata->rec->lookupTable, localtdata->tdata->buffer->oidmod[i-numread]);
+ if (headptr == NULL) {
+ printf("Error: handleLocalReq() returning NULL\n");
+ return NULL;
+ }
+ oid = OID(headptr);
+ version = headptr->version;
+ }
+ /* Check if object is still present in the machine since the beginning of TRANS_REQUEST */
+
+ /* Save the oids not found and number of oids not found for later use */
+ if ((mobj = mhashSearch(oid)) == NULL) { /* Obj not found */
+ /* Save the oids not found and number of oids not found for later use */
+ oidnotfound[objnotfound] = oid;
+ objnotfound++;
+ } else { /* If Obj found in machine (i.e. has not moved) */
+ /* Check if Obj is locked by any previous transaction */
+ if ((STATUS(((objheader_t *)mobj)) & LOCK) == LOCK) {
+ if (version == ((objheader_t *)mobj)->version) { /* If locked then match versions */
+ v_matchlock++;
+ } else {/* If versions don't match ...HARD ABORT */
+ v_nomatch++;
+ /* Send TRANS_DISAGREE to Coordinator */
+ localtdata->tdata->recvmsg[localtdata->tdata->thread_id].rcv_status = TRANS_DISAGREE;
+ }
+ } else {/* If Obj is not locked then lock object */
+ STATUS(((objheader_t *)mobj)) |= LOCK;
+ /* Save all object oids that are locked on this machine during this transaction request call */
+ oidlocked[objlocked] = OID(((objheader_t *)mobj));
+ objlocked++;
+ if (version == ((objheader_t *)mobj)->version) { /* Check if versions match */
+ v_matchnolock++;
+ } else { /* If versions don't match ...HARD ABORT */
+ v_nomatch++;
+ /* Send TRANS_DISAGREE to Coordinator */
+ localtdata->tdata->recvmsg[localtdata->tdata->thread_id].rcv_status = TRANS_DISAGREE;
+ }
+ }
+ }
+ } // End for
+ /* Condition to send TRANS_AGREE */
+ if(v_matchnolock == localtdata->tdata->buffer->f.numread + localtdata->tdata->buffer->f.nummod) {
+ localtdata->tdata->recvmsg[localtdata->tdata->thread_id].rcv_status = TRANS_AGREE;
+ }
+ /* Condition to send TRANS_SOFT_ABORT */
+ if((v_matchlock > 0 && v_nomatch == 0) || (objnotfound > 0 && v_nomatch == 0)) {
+ localtdata->tdata->recvmsg[localtdata->tdata->thread_id].rcv_status = TRANS_SOFT_ABORT;
+ }
+
+ /* Fill out the trans_commit_data_t data structure. This is required for a trans commit process
+ * if Participant receives a TRANS_COMMIT */
+ localtdata->transinfo->objlocked = oidlocked;
+ localtdata->transinfo->objnotfound = oidnotfound;
+ localtdata->transinfo->modptr = NULL;
+ localtdata->transinfo->numlocked = objlocked;
+ localtdata->transinfo->numnotfound = objnotfound;
+ /* Lock and update count */
+ //Thread sleeps until all messages from pariticipants are received by coordinator
+ pthread_mutex_lock(localtdata->tdata->lock);
+ (*(localtdata->tdata->count))++; /* keeps track of no of messages received by the coordinator */
+
+ /* Wake up the threads and invoke decideResponse (once) */
+ if(*(localtdata->tdata->count) == localtdata->tdata->buffer->f.mcount) {
+ decideResponse(localtdata->tdata);
+ pthread_cond_broadcast(localtdata->tdata->threshold);
} else {
+ pthread_cond_wait(localtdata->tdata->threshold, localtdata->tdata->lock);
+ }
+ pthread_mutex_unlock(localtdata->tdata->lock);
+ if(*(localtdata->tdata->replyctrl) == TRANS_ABORT){
+ if(transAbortProcess(localtdata) != 0) {
+ printf("Error in transAbortProcess() %s,%d\n", __FILE__, __LINE__);
+ pthread_exit(NULL);
+ }
+ } else if(*(localtdata->tdata->replyctrl) == TRANS_COMMIT) {
+ if(transComProcess(localtdata) != 0) {
+ printf("Error in transComProcess() %s,%d\n", __FILE__, __LINE__);
+ pthread_exit(NULL);
+ }
+ }
+ /* Free memory */
+ if (localtdata->transinfo->objlocked != NULL) {
+ free(localtdata->transinfo->objlocked);
+ }
+ if (localtdata->transinfo->objnotfound != NULL) {
+ free(localtdata->transinfo->objnotfound);
+ }
- h = (objheader_t *) buffer+1;
- size = sizeof(objheader_t) + sizeof(classsize[h->type]);
-#ifdef DEBUG1
- printf("DEBUG -> Received: oid = %d, type = %d\n", h->oid, h->type);
-#endif
+ pthread_exit(NULL);
+}
+
+/* This function completes the ABORT process if the transaction is aborting */
+int transAbortProcess(local_thread_data_array_t *localtdata) {
+ int i, numlocked;
+ unsigned int *objlocked;
+ void *header;
+
+ numlocked = localtdata->transinfo->numlocked;
+ objlocked = localtdata->transinfo->objlocked;
+
+ for (i = 0; i < numlocked; i++) {
+ if((header = mhashSearch(objlocked[i])) == NULL) {
+ printf("mhashsearch returns NULL at %s, %d\n", __FILE__, __LINE__);
+ return 1;
+ }
+ STATUS(((objheader_t *)header)) &= ~(LOCK);
+ }
+
+ printf("TRANS_ABORTED\n");
+
+ return 0;
+}
+
+/*This function completes the COMMIT process is the transaction is commiting*/
+int transComProcess(local_thread_data_array_t *localtdata) {
+ objheader_t *header, *tcptr;
+ int i, nummod, tmpsize, numcreated, numlocked;
+ unsigned int *oidmod, *oidcreated, *oidlocked;
+ void *ptrcreate;
+
+ nummod = localtdata->tdata->buffer->f.nummod;
+ oidmod = localtdata->tdata->buffer->oidmod;
+ numcreated = localtdata->tdata->buffer->f.numcreated;
+ oidcreated = localtdata->tdata->buffer->oidcreated;
+ numlocked = localtdata->transinfo->numlocked;
+ oidlocked = localtdata->transinfo->objlocked;
+ for (i = 0; i < nummod; i++) {
+ if((header = (objheader_t *) mhashSearch(oidmod[i])) == NULL) {
+ printf("Error: transComProcess() mhashsearch returns NULL at %s, %d\n", __FILE__, __LINE__);
+ return 1;
+ }
+ /* Copy from transaction cache -> main object store */
+ if ((tcptr = ((objheader_t *) chashSearch(localtdata->tdata->rec->lookupTable, oidmod[i]))) == NULL) {
+ printf("Error: transComProcess() chashSearch returned NULL at %s, %d\n", __FILE__, __LINE__);
+ return 1;
+ }
+ GETSIZE(tmpsize, header);
+
+ pthread_mutex_lock(&mainobjstore_mutex);
+ memcpy(header, tcptr, tmpsize + sizeof(objheader_t));
+ header->version += 1;
+ /* If threads are waiting on this object to be updated, notify them */
+ if(header->notifylist != NULL) {
+ notifyAll(&header->notifylist, OID(header), header->version);
+ }
+
+ pthread_mutex_unlock(&mainobjstore_mutex);
+ }
+ /* If object is newly created inside transaction then commit it */
+ for (i = 0; i < numcreated; i++) {
+ if ((header = ((objheader_t *) chashSearch(localtdata->tdata->rec->lookupTable, oidcreated[i]))) == NULL) {
+ printf("Error: transComProcess() chashSearch returned NULL at %s, %d\n", __FILE__, __LINE__);
+ return 1;
+ }
+ GETSIZE(tmpsize, header);
+ tmpsize += sizeof(objheader_t);
+ pthread_mutex_lock(&mainobjstore_mutex);
+ if ((ptrcreate = objstrAlloc(mainobjstore, tmpsize)) == NULL) {
+ printf("Error: transComProcess() failed objstrAlloc\n");
+ pthread_mutex_unlock(&mainobjstore_mutex);
+ return 1;
+ }
+ pthread_mutex_unlock(&mainobjstore_mutex);
+ memcpy(ptrcreate, header, tmpsize + sizeof(objheader_t));
+ mhashInsert(oidcreated[i], ptrcreate);
+ lhashInsert(oidcreated[i], myIpAddr);
+ }
+ /* Unlock locked objects */
+ for(i = 0; i < numlocked; i++) {
+ if((header = (objheader_t *) mhashSearch(oidlocked[i])) == NULL) {
+ printf("mhashsearch returns NULL at %s, %d\n", __FILE__, __LINE__);
+ return 1;
+ }
+ STATUS(header) &= ~(LOCK);
+ }
+
+ return 0;
+}
+
+/* This function checks if the prefetch oids are same and have same offsets
+ * for case x.a.b and y.a.b where x and y have same oid's
+ * or if a.b.c is a subset of x.b.c.d*/
+/* check for case where the generated request a.y.z or x.y.z.g then
+ * prefetch needs to be generated for x.y.z.g if oid of a and x are same*/
+void checkPrefetchTuples(prefetchqelem_t *node) {
+ int i,j, count,k, sindex, index;
+ char *ptr, *tmp;
+ int ntuples, slength;
+ unsigned int *oid;
+ short *endoffsets, *arryfields;
+
+ /* Check for the case x.y.z and a.b.c are same oids */
+ ptr = (char *) node;
+ ntuples = *(GET_NTUPLES(ptr));
+ oid = GET_PTR_OID(ptr);
+ endoffsets = GET_PTR_EOFF(ptr, ntuples);
+ arryfields = GET_PTR_ARRYFLD(ptr, ntuples);
+
+ /* Find offset length for each tuple */
+ int numoffset[ntuples];
+ numoffset[0] = endoffsets[0];
+ for(i = 1; i<ntuples; i++) {
+ numoffset[i] = endoffsets[i] - endoffsets[i-1];
+ }
+ /* Check for redundant tuples by comparing oids of each tuple */
+ for(i = 0; i < ntuples; i++) {
+ if(oid[i] == 0)
+ continue;
+ for(j = i+1 ; j < ntuples; j++) {
+ if(oid[j] == 0)
+ continue;
+ /*If oids of tuples match */
+ if (oid[i] == oid[j]) {
+ /* Find the smallest offset length of two tuples*/
+ if(numoffset[i] > numoffset[j]){
+ slength = numoffset[j];
+ sindex = j;
+ }
+ else {
+ slength = numoffset[i];
+ sindex = i;
+ }
+
+ /* Compare the offset values based on the current indices
+ * break if they do not match
+ * if all offset values match then pick the largest tuple*/
+
+ if(i == 0) {
+ k = 0;
+ } else {
+ k = endoffsets[i-1];
+ }
+ index = endoffsets[j -1];
+ for(count = 0; count < slength; count ++) {
+ if (arryfields[k] != arryfields[index]) {
+ break;
+ }
+ index++;
+ k++;
+ }
+ if(slength == count) {
+ oid[sindex] = 0;
+ }
+ }
+ }
}
- objcopy = objstrAlloc(record->cache, size);
- memcpy(objcopy, (void *)buffer+1, size);
+}
+/* This function makes machine piles to be added into the machine pile queue for each prefetch call */
+prefetchpile_t *makePreGroups(prefetchqelem_t *node, int *numoffset) {
+ char *ptr;
+ int ntuples, i, machinenum, count=0;
+ unsigned int *oid;
+ short *endoffsets, *arryfields, *offset;
+ prefetchpile_t *head = NULL;
+
+ /* Check for the case x.y.z and a.b.c are same oids */
+ ptr = (char *) node;
+ ntuples = *(GET_NTUPLES(ptr));
+ oid = GET_PTR_OID(ptr);
+ endoffsets = GET_PTR_EOFF(ptr, ntuples);
+ arryfields = GET_PTR_ARRYFLD(ptr, ntuples);
+
+ if((head = (prefetchpile_t *) calloc(1, sizeof(prefetchpile_t))) == NULL) {
+ printf("Calloc error: %s %d\n", __FILE__, __LINE__);
+ return NULL;
+ }
+
+ /* Check for redundant tuples by comparing oids of each tuple */
+ for(i = 0; i < ntuples; i++) {
+ if(oid[i] == 0)
+ continue;
+ /* For each tuple make piles */
+ if ((machinenum = lhashSearch(oid[i])) == 0) {
+ printf("Error: No such Machine %s, %d\n", __FILE__, __LINE__);
+ return NULL;
+ }
+ /* Insert into machine pile */
+ if(i == 0){
+ offset = &arryfields[0];
+ } else {
+ offset = &arryfields[endoffsets[i-1]];
+ }
+
+ if((head = insertPile(machinenum, oid[i], numoffset[i], offset, head)) == NULL){
+ printf("Error: Couldn't create a pile %s, %d\n", __FILE__, __LINE__);
+ return NULL;
+ }
+ }
+
+ return head;
+}
+
+prefetchpile_t *foundLocal(prefetchqelem_t *node) {
+ int ntuples,i, j, k, oidnfound = 0, arryfieldindex,nextarryfieldindex, flag = 0, val;
+ unsigned int *oid;
+ unsigned int objoid;
+ int isArray = 0;
+ char *ptr, *tmp;
+ objheader_t *objheader;
+ short *endoffsets, *arryfields;
+ prefetchpile_t *head = NULL;
+
+ ptr = (char *) node;
+ ntuples = *(GET_NTUPLES(ptr));
+ oid = GET_PTR_OID(ptr);
+ endoffsets = GET_PTR_EOFF(ptr, ntuples);
+ arryfields = GET_PTR_ARRYFLD(ptr, ntuples);
+
+ /* Find offset length for each tuple */
+ int numoffset[ntuples];//Number of offsets for each tuple
+ numoffset[0] = endoffsets[0];
+ for(i = 1; i<ntuples; i++) {
+ numoffset[i] = endoffsets[i] - endoffsets[i-1];
+ }
+ for(i = 0; i < ntuples; i++) {
+ if(oid[i] == 0){
+ if(i == 0) {
+ arryfieldindex = 0;
+ nextarryfieldindex = endoffsets[0];
+ }else {
+ arryfieldindex = endoffsets[i-1];
+ nextarryfieldindex = endoffsets[i];
+ }
+ numoffset[i] = 0;
+ endoffsets[0] = val = numoffset[0];
+ for(k = 1; k < ntuples; k++) {
+ val = val + numoffset[k];
+ endoffsets[k] = val;
+ }
+
+ for(k = 0; k<endoffsets[ntuples-1]; k++) {
+ arryfields[arryfieldindex+k] = arryfields[nextarryfieldindex+k];
+ }
+ continue;
+ }
+ /* If object found locally */
+ if((objheader = (objheader_t*) mhashSearch(oid[i])) != NULL) {
+ tmp = (char *) objheader;
+ int orgnumoffset = numoffset[i];
+ if(i == 0) {
+ arryfieldindex = 0;
+ }else {
+ arryfieldindex = endoffsets[i-1];
+ }
+
+ for(j = 0; j<orgnumoffset; j++) {
+ /* Check for arrays */
+ if(TYPE(objheader) > NUMCLASSES) {
+ isArray = 1;
+ }
+ if(isArray == 1) {
+ int elementsize = classsize[TYPE(objheader)];
+ objoid = *((unsigned int *)(tmp + sizeof(objheader_t) + sizeof(struct ArrayObject) + (elementsize*arryfields[arryfieldindex])));
+ } else {
+ objoid = *((unsigned int *)(tmp + sizeof(objheader_t) + arryfields[arryfieldindex]));
+ }
+ //Update numoffset array
+ numoffset[i] = numoffset[i] - 1;
+ //Update oid array
+ oid[i] = objoid;
+ //Update endoffset array
+ endoffsets[0] = val = numoffset[0];
+ for(k = 1; k < ntuples; k++) {
+ val = val + numoffset[k];
+ endoffsets[k] = val;
+ }
+ //Update arrayfields array
+ for(k = 0; k < endoffsets[ntuples-1]; k++) {
+ arryfields[arryfieldindex+k] = arryfields[arryfieldindex+k+1];
+ }
+ if((objheader = (objheader_t*) mhashSearch(oid[i])) == NULL) {
+ flag = 1;
+ checkPreCache(node, numoffset, oid[i], i);
+ break;
+ }
+ tmp = (char *) objheader;
+ isArray = 0;
+ }
+ /*If all offset oids are found locally,make the prefetch tuple invalid */
+ if(flag == 0) {
+ oid[i] = 0;
+ }
+ } else {
+ /* Look in Prefetch cache */
+ checkPreCache(node, numoffset, oid[i],i);
+ }
+ }
+
+ /* Make machine groups */
+ if((head = makePreGroups(node, numoffset)) == NULL) {
+ printf("Error in makePreGroups() %s %d\n", __FILE__, __LINE__);
+ return NULL;
+ }
+
+ return head;
+}
+
+void checkPreCache(prefetchqelem_t *node, int *numoffset, unsigned int objoid, int index) {
+ char *ptr, *tmp;
+ int ntuples, i, k, flag=0, isArray =0, arryfieldindex, val;
+ unsigned int * oid;
+ short *endoffsets, *arryfields;
+ objheader_t *header;
+
+ ptr = (char *) node;
+ ntuples = *(GET_NTUPLES(ptr));
+ oid = GET_PTR_OID(ptr);
+ endoffsets = GET_PTR_EOFF(ptr, ntuples);
+ arryfields = GET_PTR_ARRYFLD(ptr, ntuples);
+
+ if((header = (objheader_t *) prehashSearch(objoid)) == NULL) {
+ return;
+ } else { //Found in Prefetch Cache
+ //TODO Decide if object is too old, if old remove from cache
+ tmp = (char *) header;
+ int loopcount = numoffset[index];
+ if(index == 0)
+ arryfieldindex = 0;
+ else
+ arryfieldindex = endoffsets[(index - 1)];
+ // Check if any of the offset oid is available in the Prefetch cache
+ for(i = 0; i < loopcount; i++) {
+ /* Check for arrays */
+ if(TYPE(header) > NUMCLASSES) {
+ isArray = 1;
+ }
+ if(isArray == 1) {
+ int elementsize = classsize[TYPE(header)];
+ objoid = *((unsigned int *)(tmp + sizeof(objheader_t) + sizeof(struct ArrayObject) + (elementsize*arryfields[arryfieldindex])));
+ } else {
+ objoid = *((unsigned int *)(tmp + sizeof(objheader_t) + arryfields[arryfieldindex]));
+ }
+ //Update numoffset array
+ numoffset[index] = numoffset[index] - 1;
+ //Update oid array
+ oid[index] = objoid;
+ //Update endoffset array
+ endoffsets[0] = val = numoffset[0];
+ for(k = 1; k < ntuples; k++) {
+ val = val + numoffset[k];
+ endoffsets[k] = val;
+ }
+ //Update arrayfields array
+ for(k = 0; k < endoffsets[ntuples-1]; k++) {
+ arryfields[arryfieldindex+k] = arryfields[arryfieldindex+k+1];
+ }
+ if((header = (objheader_t *)prehashSearch(oid[index])) != NULL) {
+ tmp = (char *) header;
+ isArray = 0;
+ } else {
+ flag = 1;
+ break;
+ }
+ }
+ }
+ //Found in the prefetch cache
+ if(flag == 0 && (numoffset[index] == 0)) {
+ oid[index] = 0;
+ }
+}
+
+
+
+/* This function is called by the thread calling transPrefetch */
+void *transPrefetch(void *t) {
+ prefetchqelem_t *qnode;
+ prefetchpile_t *pilehead = NULL;
+ prefetchpile_t *ptr = NULL, *piletail = NULL;
+
+ while(1) {
+ /* lock mutex of primary prefetch queue */
+ pthread_mutex_lock(&pqueue.qlock);
+ /* while primary queue is empty, then wait */
+ while((pqueue.front == NULL) && (pqueue.rear == NULL)) {
+ pthread_cond_wait(&pqueue.qcond, &pqueue.qlock);
+ }
+
+ /* dequeue node to create a machine piles and finally unlock mutex */
+ if((qnode = pre_dequeue()) == NULL) {
+ printf("Error: No node returned %s, %d\n", __FILE__, __LINE__);
+ pthread_mutex_unlock(&pqueue.qlock);
+ pthread_exit(NULL);
+ }
+ pthread_mutex_unlock(&pqueue.qlock);
+
+ /* Reduce redundant prefetch requests */
+ checkPrefetchTuples(qnode);
+ /* Check if the tuples are found locally, if yes then reduce them further*/
+ /* and group requests by remote machine ids by calling the makePreGroups() */
+ if((pilehead = foundLocal(qnode)) == NULL) {
+ printf("Error: No node created for serving prefetch request %s %d\n", __FILE__, __LINE__);
+ pthread_exit(NULL);
+ }
+
+ ptr = pilehead;
+ while(ptr != NULL) {
+ if(ptr->next == NULL) {
+ piletail = ptr;
+ }
+ ptr = ptr->next;
+ }
+
+ /* Lock mutex of pool queue */
+ pthread_mutex_lock(&mcqueue.qlock);
+ /* Update the pool queue with the new remote machine piles generated per prefetch call */
+ mcpileenqueue(pilehead, piletail);
+ /* Broadcast signal on machine pile queue */
+ pthread_cond_broadcast(&mcqueue.qcond);
+ /* Unlock mutex of machine pile queue */
+ pthread_mutex_unlock(&mcqueue.qlock);
+ /* Deallocate the prefetch queue pile node */
+ predealloc(qnode);
+ pthread_exit(NULL);
+ }
+}
+
+/* Each thread in the pool of threads calls this function to establish connection with
+ * remote machines, send the prefetch requests and process the reponses from
+ * the remote machines .
+ * The thread is active throughout the period of runtime */
+
+void *mcqProcess(void *threadid) {
+ int tid;
+ prefetchpile_t *mcpilenode;
+
+ tid = (int) threadid;
+ while(1) {
+ /* Lock mutex of mc pile queue */
+ pthread_mutex_lock(&mcqueue.qlock);
+ /* When mc pile queue is empty, wait */
+ while((mcqueue.front == NULL) && (mcqueue.rear == NULL)) {
+ pthread_cond_wait(&mcqueue.qcond, &mcqueue.qlock);
+ }
+ /* Dequeue node to send remote machine connections*/
+ if((mcpilenode = mcpiledequeue()) == NULL) {
+ printf("Dequeue Error: No node returned %s %d\n", __FILE__, __LINE__);
+ pthread_mutex_unlock(&mcqueue.qlock);
+ pthread_exit(NULL);
+ }
+ /* Unlock mutex */
+ pthread_mutex_unlock(&mcqueue.qlock);
+
+ /*Initiate connection to remote host and send request */
+ /* Process Request */
+ sendPrefetchReq(mcpilenode, tid);
+ /* Deallocate the machine queue pile node */
+ mcdealloc(mcpilenode);
+ pthread_exit(NULL);
+ }
+}
+
+void sendPrefetchReq(prefetchpile_t *mcpilenode, int threadid) {
+ int sd, i, off, len, endpair, count = 0;
+ struct sockaddr_in serv_addr;
+ struct hostent *server;
+ char machineip[16], control;
+ objpile_t *tmp;
+
+ /* Send Trans Prefetch Request */
+ if ((sd = socket(AF_INET, SOCK_STREAM, 0)) < 0) {
+ perror("Error in socket for SEND_PREFETCH_REQUEST\n");
+ return;
+ }
+
+ bzero((char*) &serv_addr, sizeof(serv_addr));
+ serv_addr.sin_family = AF_INET;
+ serv_addr.sin_port = htons(LISTEN_PORT);
+ midtoIP(mcpilenode->mid ,machineip);
+ machineip[15] = '\0';
+ serv_addr.sin_addr.s_addr = inet_addr(machineip);
+
+ /* Open Connection */
+ if (connect(sd, (struct sockaddr *) &serv_addr, sizeof(struct sockaddr)) < 0) {
+ perror("Error in connect for SEND_PREFETCH_REQUEST\n");
+ close(sd);
+ return;
+ }
+
+ /* Send TRANS_PREFETCH control message */
+ control = TRANS_PREFETCH;
+ if(send(sd, &control, sizeof(char), MSG_NOSIGNAL) < sizeof(char)) {
+ perror("Error in sending prefetch control\n");
+ close(sd);
+ return;
+ }
+
+ /* Send Oids and offsets in pairs */
+ tmp = mcpilenode->objpiles;
+ while(tmp != NULL) {
+ off = 0;
+ count++; /* Keeps track of the number of oid and offset tuples sent per remote machine */
+ len = sizeof(int) + sizeof(unsigned int) + ((tmp->numoffset) * sizeof(short));
+ char oidnoffset[len];
+ bzero(oidnoffset, len);
+ memcpy(oidnoffset, &len, sizeof(int));
+ off = sizeof(int);
+ memcpy(oidnoffset + off, &tmp->oid, sizeof(unsigned int));
+ off += sizeof(unsigned int);
+ for(i = 0; i < tmp->numoffset; i++) {
+ memcpy(oidnoffset + off, &(tmp->offset[i]), sizeof(short));
+ off+=sizeof(short);
+ }
+ if (send(sd, &oidnoffset, sizeof(oidnoffset),MSG_NOSIGNAL) < sizeof(oidnoffset)) {
+ perror("Error sending fixed bytes for thread\n");
+ close(sd);
+ return;
+ }
+
+ tmp = tmp->next;
+ }
+
+ /* Send a special char -1 to represent the end of sending oids + offset pair to remote machine */
+ endpair = -1;
+ if (send(sd, &endpair, sizeof(int), MSG_NOSIGNAL) < sizeof(int)) {
+ perror("Error sending fixed bytes for thread\n");
+ close(sd);
+ return;
+ }
+
+ /* Get Response from the remote machine */
+ getPrefetchResponse(count,sd);
+ close(sd);
+ return;
+}
+
+void getPrefetchResponse(int count, int sd) {
+ int i = 0, val, n, N, sum, index, objsize;
+ unsigned int bufsize,oid;
+ char buffer[RECEIVE_BUFFER_SIZE], control;
+ char *ptr;
+ void *modptr, *oldptr;
+
+
+ /* Read prefetch response from the Remote machine */
+ if((val = read(sd, &control, sizeof(char))) <= 0) {
+ perror("No control response for Prefetch request sent\n");
+ return;
+ }
+
+ if(control == TRANS_PREFETCH_RESPONSE) {
+ /*For each oid and offset tuple sent as prefetch request to remote machine*/
+ while(i < count) {
+ sum = 0;
+ index = 0;
+ /* Read the size of buffer to be received */
+ if((N = read(sd, buffer, sizeof(unsigned int))) <= 0) {
+ perror("Size of buffer not recv\n");
+ return;
+ }
+ bufsize = *((unsigned int *) buffer);
+ ptr = buffer + sizeof(unsigned int);
+ /* Keep receiving the buffer containing oid info */
+ do {
+ n = recv((int)sd, (void *)ptr+sum, bufsize-sum, 0);
+ sum +=n;
+ } while(sum < bufsize && n != 0);
+
+ /* Decode the contents of the buffer */
+ index = sizeof(unsigned int);
+ while(index < (bufsize - sizeof(unsigned int))) {
+ if(buffer[index] == OBJECT_FOUND) {
+ /* Increment it to get the object */
+ index += sizeof(char);
+ oid = *((unsigned int *)(buffer+index));
+ index += sizeof(unsigned int);
+ /* For each object found add to Prefetch Cache */
+ objsize = *((int *)(buffer+index));
+ index+=sizeof(int);
+ pthread_mutex_lock(&prefetchcache_mutex);
+ if ((modptr = objstrAlloc(prefetchcache, objsize)) == NULL) {
+ printf("objstrAlloc error for copying into prefetch cache %s, %d\n", __FILE__, __LINE__);
+ pthread_mutex_unlock(&prefetchcache_mutex);
+ return;
+ }
+ pthread_mutex_unlock(&prefetchcache_mutex);
+ memcpy(modptr, buffer+index, objsize);
+ index += objsize;
+ /* Insert the oid and its address into the prefetch hash lookup table */
+ /* Do a version comparison if the oid exists */
+ if((oldptr = prehashSearch(oid)) != NULL) {
+ /* If older version then update with new object ptr */
+ if(((objheader_t *)oldptr)->version < ((objheader_t *)modptr)->version) {
+ prehashRemove(oid);
+ prehashInsert(oid, modptr);
+ } else if(((objheader_t *)oldptr)->version == ((objheader_t *)modptr)->version) {
+ /* Add the new object ptr to hash table */
+ prehashRemove(oid);
+ prehashInsert(oid, modptr);
+ } else { /* Do nothing: TODO modptr should be reference counted */
+ ;
+ }
+ } else {/*If doesn't no match found in hashtable, add the object ptr to hash table*/
+ prehashInsert(oid, modptr);
+ }
+ /* Lock the Prefetch Cache look up table*/
+ pthread_mutex_lock(&pflookup.lock);
+ /* Broadcast signal on prefetch cache condition variable */
+ pthread_cond_broadcast(&pflookup.cond);
+ /* Unlock the Prefetch Cache look up table*/
+ pthread_mutex_unlock(&pflookup.lock);
+ } else if(buffer[index] == OBJECT_NOT_FOUND) {
+ /* Increment it to get the object */
+ /* TODO: For each object not found query DHT for new location and retrieve the object */
+ index += sizeof(char);
+ oid = *((unsigned int *)(buffer + index));
+ index += sizeof(unsigned int);
+ /* Throw an error */
+ printf("OBJECT NOT FOUND.... THIS SHOULD NOT HAPPEN...TERMINATE PROGRAM\n");
+ exit(-1);
+ } else {
+ printf("Error in decoding the index value %s, %d\n",__FILE__, __LINE__);
+ return;
+ }
+ }
+
+ i++;
+ }
+ } else
+ printf("Error in receving response for prefetch request %s, %d\n",__FILE__, __LINE__);
+ return;
+}
+
+unsigned short getObjType(unsigned int oid)
+{
+ objheader_t *objheader;
+ unsigned short numoffset[] ={0};
+ short fieldoffset[] ={};
+
+ if ((objheader = (objheader_t *) mhashSearch(oid)) == NULL)
+ {
+ if ((objheader = (objheader_t *) prehashSearch(oid)) == NULL)
+ {
+ prefetch(1, &oid, numoffset, fieldoffset);
+ pthread_mutex_lock(&pflookup.lock);
+ while ((objheader = (objheader_t *) prehashSearch(oid)) == NULL)
+ {
+ pthread_cond_wait(&pflookup.cond, &pflookup.lock);
+ }
+ pthread_mutex_unlock(&pflookup.lock);
+ }
+ }
+
+ return TYPE(objheader);
+}
+
+int startRemoteThread(unsigned int oid, unsigned int mid)
+{
+ int sock;
+ struct sockaddr_in remoteAddr;
+ char msg[1 + sizeof(unsigned int)];
+ int bytesSent;
+ int status;
+
+ if ((sock = socket(AF_INET, SOCK_STREAM, 0)) < 0)
+ {
+ perror("startRemoteThread():socket()");
+ return -1;
+ }
+
+ bzero(&remoteAddr, sizeof(remoteAddr));
+ remoteAddr.sin_family = AF_INET;
+ remoteAddr.sin_port = htons(LISTEN_PORT);
+ remoteAddr.sin_addr.s_addr = htonl(mid);
+
+ if (connect(sock, (struct sockaddr *)&remoteAddr, sizeof(remoteAddr)) < 0)
+ {
+ printf("startRemoteThread():error %d connecting to %s:%d\n", errno,
+ inet_ntoa(remoteAddr.sin_addr), LISTEN_PORT);
+ status = -1;
+ }
+ else
+ {
+ msg[0] = START_REMOTE_THREAD;
+ memcpy(&msg[1], &oid, sizeof(unsigned int));
+
+ bytesSent = send(sock, msg, 1 + sizeof(unsigned int), 0);
+ if (bytesSent < 0)
+ {
+ perror("startRemoteThread():send()");
+ status = -1;
+ }
+ else if (bytesSent != 1 + sizeof(unsigned int))
+ {
+ printf("startRemoteThread(): error, sent %d bytes\n", bytesSent);
+ status = -1;
+ }
+ else
+ {
+ status = 0;
+ }
+ }
+
+ close(sock);
+ return status;
+}
+
+//TODO: when reusing oids, make sure they are not already in use!
+unsigned int getNewOID(void) {
+ static unsigned int id = 0xFFFFFFFF;
+
+ id += 2;
+ if (id > oidMax || id < oidMin)
+ {
+ id = (oidMin | 1);
+ }
+ return id;
+}
+
+int processConfigFile()
+{
+ FILE *configFile;
+ const int maxLineLength = 200;
+ char lineBuffer[maxLineLength];
+ char *token;
+ const char *delimiters = " \t\n";
+ char *commentBegin;
+ in_addr_t tmpAddr;
+
+ configFile = fopen(CONFIG_FILENAME, "r");
+ if (configFile == NULL)
+ {
+ printf("error opening %s:\n", CONFIG_FILENAME);
+ perror("");
+ return -1;
+ }
+
+ numHostsInSystem = 0;
+ sizeOfHostArray = 8;
+ hostIpAddrs = calloc(sizeOfHostArray, sizeof(unsigned int));
+
+ while(fgets(lineBuffer, maxLineLength, configFile) != NULL)
+ {
+ commentBegin = strchr(lineBuffer, '#');
+ if (commentBegin != NULL)
+ *commentBegin = '\0';
+ token = strtok(lineBuffer, delimiters);
+ while (token != NULL)
+ {
+ tmpAddr = inet_addr(token);
+ if ((int)tmpAddr == -1)
+ {
+ printf("error in %s: bad token:%s\n", CONFIG_FILENAME, token);
+ fclose(configFile);
+ return -1;
+ }
+ else
+ addHost(htonl(tmpAddr));
+ token = strtok(NULL, delimiters);
+ }
+ }
+
+ fclose(configFile);
+
+ if (numHostsInSystem < 1)
+ {
+ printf("error in %s: no IP Adresses found\n", CONFIG_FILENAME);
+ return -1;
+ }
+#ifdef MAC
+ myIpAddr = getMyIpAddr("en1");
+#else
+ myIpAddr = getMyIpAddr("eth0");
#endif
- //Insert into cache's lookup table
- chashInsert(record->lookupTable, oid, objcopy);
- return objcopy;
+ myIndexInHostArray = findHost(myIpAddr);
+ if (myIndexInHostArray == -1)
+ {
+ printf("error in %s: IP Address of eth0 not found\n", CONFIG_FILENAME);
+ return -1;
+ }
+ oidsPerBlock = (0xFFFFFFFF / numHostsInSystem) + 1;
+ oidMin = oidsPerBlock * myIndexInHostArray;
+ if (myIndexInHostArray == numHostsInSystem - 1)
+ oidMax = 0xFFFFFFFF;
+ else
+ oidMax = oidsPerBlock * (myIndexInHostArray + 1) - 1;
+
+ return 0;
+}
+
+void addHost(unsigned int hostIp)
+{
+ unsigned int *tmpArray;
+
+ if (findHost(hostIp) != -1)
+ return;
+
+ if (numHostsInSystem == sizeOfHostArray)
+ {
+ tmpArray = calloc(sizeOfHostArray * 2, sizeof(unsigned int));
+ memcpy(tmpArray, hostIpAddrs, sizeof(unsigned int) * numHostsInSystem);
+ free(hostIpAddrs);
+ hostIpAddrs = tmpArray;
+ }
+
+ hostIpAddrs[numHostsInSystem++] = hostIp;
+
+ return;
+}
+
+int findHost(unsigned int hostIp)
+{
+ int i;
+ for (i = 0; i < numHostsInSystem; i++)
+ if (hostIpAddrs[i] == hostIp)
+ return i;
+
+ //not found
+ return -1;
+}
+
+/* This function sends notification request per thread waiting on object(s) whose version
+ * changes */
+int reqNotify(unsigned int *oidarry, unsigned short *versionarry, unsigned int numoid) {
+ int sock,i;
+ objheader_t *objheader;
+ struct sockaddr_in remoteAddr;
+ char msg[1 + numoid * (sizeof(short) + sizeof(unsigned int)) + 3 * sizeof(unsigned int)];
+ char *ptr;
+ int bytesSent;
+ int status, size;
+ unsigned short version;
+ unsigned int oid, threadid, mid;
+ pthread_mutex_t threadnotify; //Lock and condition var for threadjoin and notification
+ pthread_cond_t threadcond;
+ notifydata_t *ndata;
+
+ //FIXME currently all oids belong to one machine
+ oid = oidarry[0];
+ mid = lhashSearch(oid);
+
+ if ((sock = socket(AF_INET, SOCK_STREAM, 0)) < 0){
+ perror("reqNotify():socket()");
+ return -1;
+ }
+
+ bzero(&remoteAddr, sizeof(remoteAddr));
+ remoteAddr.sin_family = AF_INET;
+ remoteAddr.sin_port = htons(LISTEN_PORT);
+ remoteAddr.sin_addr.s_addr = htonl(mid);
+
+ /* Generate unique threadid */
+ threadid = (unsigned int) pthread_self();
+
+ /* Save threadid, numoid, oidarray, versionarray, pthread_cond_variable for later processing */
+ if((ndata = calloc(1, sizeof(notifydata_t))) == NULL) {
+ printf("Calloc Error %s, %d\n", __FILE__, __LINE__);
+ return -1;
+ }
+ ndata->numoid = numoid;
+ ndata->threadid = threadid;
+ ndata->oidarry = oidarry;
+ ndata->versionarry = versionarry;
+ ndata->threadcond = threadcond;
+ if((status = notifyhashInsert(threadid, ndata)) != 0) {
+ printf("reqNotify(): Insert into notify hash table not successful %s, %d\n", __FILE__, __LINE__);
+ free(ndata);
+ return -1;
+ }
+
+ /* Send number of oids, oidarry, version array, machine id and threadid */
+ if (connect(sock, (struct sockaddr *)&remoteAddr, sizeof(remoteAddr)) < 0){
+ printf("reqNotify():error %d connecting to %s:%d\n", errno,
+ inet_ntoa(remoteAddr.sin_addr), LISTEN_PORT);
+ free(ndata);
+ return -1;
+ } else {
+ msg[0] = THREAD_NOTIFY_REQUEST;
+ *((unsigned int *)(&msg[1])) = numoid;
+ /* Send array of oids */
+ size = sizeof(unsigned int);
+ {
+ i = 0;
+ while(i < numoid) {
+ oid = oidarry[i];
+ *((unsigned int *)(&msg[1] + size)) = oid;
+ size += sizeof(unsigned int);
+ i++;
+ }
+ }
+
+ /* Send array of version */
+ {
+ i = 0;
+ while(i < numoid) {
+ version = versionarry[i];
+ *((unsigned short *)(&msg[1] + size)) = version;
+ size += sizeof(unsigned short);
+ i++;
+ }
+ }
+
+ *((unsigned int *)(&msg[1] + size)) = myIpAddr;
+ size += sizeof(unsigned int);
+ *((unsigned int *)(&msg[1] + size)) = threadid;
+
+ pthread_mutex_lock(&threadnotify);
+ bytesSent = send(sock, msg, 1 + numoid * (sizeof(unsigned int) + sizeof(unsigned short)) + 2 * sizeof(unsigned int) , 0);
+ if (bytesSent < 0){
+ perror("reqNotify():send()");
+ status = -1;
+ } else if (bytesSent != 1 + numoid*(sizeof(unsigned int) + sizeof(unsigned short)) + 2 * sizeof(unsigned int)){
+ printf("reNotify(): error, sent %d bytes\n", bytesSent);
+ status = -1;
+ } else {
+ status = 0;
+ }
+ pthread_cond_wait(&threadcond, &threadnotify);
+ pthread_mutex_unlock(&threadnotify);
+ }
+
+ close(sock);
+ return status;
+}
+void threadNotify(unsigned int oid, unsigned short version, unsigned int tid) {
+ notifydata_t *ndata;
+ int i, objIsFound = 0, index;
+ void *ptr;
+
+ //Look up the tid and call the corresponding pthread_cond_signal
+ if((ndata = notifyhashSearch(tid)) == NULL) {
+ printf("threadnotify(): No such threadid is present %s, %d\n", __FILE__, __LINE__);
+ return;
+ } else {
+ for(i = 0; i < ndata->numoid; i++) {
+ if(ndata->oidarry[i] == oid){
+ objIsFound = 1;
+ index = i;
+ }
+ }
+ if(objIsFound == 0){
+ return;
+ } else {
+ if(version <= ndata->versionarry[index]){
+ return;
+ } else {
+ /* Clear from prefetch cache and free thread related data structure */
+ if((ptr = prehashSearch(oid)) == NULL) {
+ //TODO Ask about freeing
+ printf("threadnotify(): No such oid %s, %d\n", __FILE__, __LINE__);
+ pthread_cond_signal(&ndata->threadcond);
+ free(ndata);
+ return;
+ } else {
+ prehashRemove(oid);
+ pthread_cond_signal(&ndata->threadcond);
+ free(ndata);
+ }
+ }
+ }
+ }
+ return;
+}
+
+int notifyAll(threadlist_t **head, unsigned int oid, unsigned int version) {
+ threadlist_t *ptr;
+ unsigned int mid;
+ struct sockaddr_in remoteAddr;
+ char msg[1 + sizeof(unsigned short) + 2*sizeof(unsigned int)];
+ int sock, status, size, bytesSent;
+ while(*head != NULL) {
+ ptr = *head;
+ mid = ptr->mid;
+ //create a socket connection to that machine
+ if ((sock = socket(AF_INET, SOCK_STREAM, 0)) < 0){
+ perror("notifyAll():socket()");
+ return -1;
+ }
+
+ bzero(&remoteAddr, sizeof(remoteAddr));
+ remoteAddr.sin_family = AF_INET;
+ remoteAddr.sin_port = htons(LISTEN_PORT);
+ remoteAddr.sin_addr.s_addr = htonl(mid);
+ //send Thread Notify response and threadid to that machine
+ if (connect(sock, (struct sockaddr *)&remoteAddr, sizeof(remoteAddr)) < 0){
+ printf("notifyAll():error %d connecting to %s:%d\n", errno,
+ inet_ntoa(remoteAddr.sin_addr), LISTEN_PORT);
+ status = -1;
+ } else {
+ msg[0] = THREAD_NOTIFY_RESPONSE;
+ msg[1] = oid;
+ size = sizeof(unsigned int);
+ *((unsigned short *)(&msg[1]+ size)) = version;
+ size+= sizeof(unsigned short);
+ *((unsigned int *)(&msg[1]+ size)) = ptr->threadid;
+
+ bytesSent = send(sock, msg, 1 + 2*sizeof(unsigned int) + sizeof(unsigned short), 0);
+ if (bytesSent < 0){
+ perror("notifyAll():send()");
+ status = -1;
+ } else if (bytesSent != 1 + 2*sizeof(unsigned int) + sizeof(unsigned short)){
+ printf("notifyAll(): error, sent %d bytes\n", bytesSent);
+ status = -1;
+ } else {
+ status = 0;
+ }
+ }
+ //close socket
+ close(sock);
+ // Update head
+ *head = ptr->next;
+ free(ptr);
+ }
+ return status;
+}
+
+void transAbort(transrecord_t *trans) {
+ objstrDelete(trans->cache);
+ chashDelete(trans->lookupTable);
+ free(trans);
}