// this is only used when this node is the first node in the cluster, and we need to create some new empty buckets.
void buckets_init(hash_t mask, struct event_base *evbase)
{
int i;
assert(_evbase == NULL);
assert(evbase);
_evbase = evbase;
assert(_mask == 0);
assert(mask >= _mask);
_mask = mask;
assert(_mask > 0);
_buckets = calloc(_mask+1, sizeof(bucket_t *));
assert(_buckets);
assert(_primary_buckets == 0);
assert(_secondary_buckets == 0);
// for starters we will need to create a bucket for each hash.
for (i=0; i<=_mask; i++) {
_buckets[i] = bucket_new(i);
_primary_buckets ++;
_buckets[i]->level = 0;
// send out a message to all connected clients, to let them know that the buckets have changed.
update_hashmasks(_buckets[i]); // all_hashmask(i, 0);
}
// indicate that we have buckets that do not have backup copies on other nodes.
_nobackup_buckets = _mask + 1;
}
/*
* heap_buckets_init -- (internal) initializes bucket instances
*/
int
heap_buckets_init(struct palloc_heap *heap)
{
struct heap_rt *h = heap->rt;
for (uint8_t i = 0; i < MAX_ALLOCATION_CLASSES; ++i) {
struct alloc_class *c = alloc_class_by_id(h->alloc_classes, i);
if (c != NULL) {
if (heap_create_alloc_class_buckets(heap, c) != 0)
goto error_bucket_create;
}
}
h->default_bucket = bucket_new(container_new_ctree(heap),
alloc_class_by_id(h->alloc_classes, DEFAULT_ALLOC_CLASS_ID));
if (h->default_bucket == NULL)
goto error_bucket_create;
heap_populate_buckets(heap);
return 0;
error_bucket_create:
for (unsigned i = 0; i < h->ncaches; ++i)
bucket_group_destroy(h->caches[i].buckets);
return -1;
}
int htable_add(htable* ht, void* key, size_t keylen, void* data, size_t dlen) { // {{{
if (htable_load_factor(ht) > LOAD_FACTOR)
htable_grow(ht);
int hval;
bucket** b = htable_find_pbucket(ht, key, keylen, &hval);
if (*b)
return HTABLE_NOT_FOUND;
*b = bucket_new(key, keylen, data, dlen, hval);
ht->entries++;
return HTABLE_FOUND;
} // }}}
void htable_set(htable* ht, void* key, size_t keylen, void* data, size_t dlen) { // {{{
if (htable_load_factor(ht) > LOAD_FACTOR)
htable_grow(ht);
int hval;
bucket** b = htable_find_pbucket(ht, key, keylen, &hval);
if (*b) {
free(b[0]->data);
b[0]->data = xmemdup(data, dlen);
b[0]->dlen = dlen;
} else {
*b = bucket_new(key, keylen, data, dlen, hval);
ht->entries++;
}
} // }}}
/*
* heap_create_alloc_class_buckets -- (internal) allocates all cache bucket
* instances of the specified type
*/
static int
heap_create_alloc_class_buckets(struct palloc_heap *heap, struct alloc_class *c)
{
struct heap_rt *h = heap->rt;
int i;
for (i = 0; i < (int)h->ncaches; ++i) {
h->caches[i].buckets[c->id] = bucket_new(
container_new_seglists(heap), c);
if (h->caches[i].buckets[c->id] == NULL)
goto error_cache_bucket_new;
}
return 0;
error_cache_bucket_new:
for (i -= 1; i >= 0; --i) {
bucket_delete(h->caches[i].buckets[c->id]);
}
return -1;
}
void dict_set(dict* d, char* key, void* item) {
int index = hash(key, d->size);
struct bucket* b = d->buckets[index];
struct bucket** p = &d->buckets[index];
while(true) {
if (b == NULL) {
*p = bucket_new(key, item);
return;
}
if( strcmp(b->key, key) == 0) {
b->item = item;
return;
}
p = &b->next;
b = b->next;
}
}
int test_file(FILE *fp, int argc, char **argv) {
char buf[65535 + 1];
char *pos;
unsigned int strategy = 0; /* what bucketing strategy we're
using */
void *ptr = NULL;
unsigned int bucketsize = 0;
struct params params = {0};
struct chash *hash = NULL;
char name[256];
if (!parse_params(argc, argv, ¶ms)) {
fprintf(stderr, "failed to parse params\n");
return 0;
}
while (fgets((char *) buf, 65535, fp)) {
str_rtrim(buf);
pos = (char *) str_ltrim(buf);
if (!str_casecmp(pos, "new")) {
/* creating a new bucket */
unsigned int size = -1;
if (ptr) {
chash_delete(hash);
free(ptr);
}
/* read parameters */
if ((fscanf(fp, "%255s %u %u", name, &strategy, &size) == 3)
&& (size <= 65535)
&& (bucketsize = size)
&& (ptr = malloc(size))
&& (hash = chash_ptr_new(1, 2.0,
/* some fn pointer casting dodginess */
(unsigned int (*)(const void *)) str_len,
(int (*)(const void *, const void *)) str_cmp))
&& (bucket_new(ptr, bucketsize, strategy))) {
/* succeeded, do nothing */
if (params.verbose) {
printf("%s: new bucket with size %u strategy %u\n", name,
size, strategy);
}
} else {
fprintf(stderr, "%s: failed to create bucket\n", name);
return 0;
}
} else if (!str_casecmp(pos, "add")) {
/* adding a term to the bucket */
void *ret;
unsigned int veclen,
succeed,
len;
int toobig;
if (!ptr) { return 0; }
/* read parameters */
if ((fscanf(fp, "%65535s %u %u", buf, &veclen, &succeed) == 3)
&& (veclen <= 65535)) {
len = str_len(buf);
if ((((ret = bucket_alloc(ptr, bucketsize, strategy, buf, len,
veclen, &toobig, NULL))
&& succeed)
|| (!ret && !succeed))) {
/* do nothing */
if (params.verbose) {
printf("%s: added term '%s'\n", name, buf);
}
} else if (succeed) {
fprintf(stderr, "%s: failed to add '%s' to bucket\n",
name, buf);
return 0;
} else if (!succeed) {
fprintf(stderr, "%s: add '%s' succeeded but shouldn't "
"have\n", name, buf);
return 0;
}
} else {
fprintf(stderr, "%s: failed to add\n", name);
return 0;
}
} else if (!str_casecmp(pos, "ls")) {
/* matching stuff in the bucket */
unsigned int numterms,
i,
len,
veclen,
veclen2,
state;
void *addr;
struct chash *tmphash;
const char *term;
void **tmpptr,
*tmp;
if (!ptr) { return 0; }
if (!(tmphash = chash_ptr_new(1, 2.0,
/* some fn pointer casting dodginess */
(unsigned int (*)(const void *)) str_len,
(int (*)(const void *, const void *)) str_cmp))) {
fprintf(stderr, "%s: failed to init hashtable\n", name);
return 0;
}
/* first, fill hashtable with all terms from bucket */
state = 0;
while ((term
= bucket_next_term(ptr, bucketsize, strategy,
&state, &len, &addr, &veclen))) {
if (!((term = str_ndup(term, len))
&& (chash_ptr_ptr_insert(tmphash, term, (void*) term)
== CHASH_OK))) {
fprintf(stderr, "%s: failed to init hashtable\n", name);
return 0;
}
}
/* now, take terms from file, comparing them with hashtable
* entries */
if (fscanf(fp, "%u", &numterms)) {
for (i = 0; i < numterms; i++) {
if (fscanf(fp, "%65535s %u ", buf, &veclen)) {
if (params.verbose) {
printf("%s: ls checking %s\n", name, buf);
}
if ((addr = bucket_find(ptr, bucketsize, strategy,
buf, str_len(buf), &veclen2, NULL))
/* remove it from hashtable */
&& chash_ptr_ptr_find(tmphash, buf, &tmpptr)
== CHASH_OK
&& chash_ptr_ptr_remove(tmphash, *tmpptr, &tmp)
== CHASH_OK
&& (free(tmp), 1)
&& (veclen <= 65535)
&& (veclen2 == veclen)
&& fread(buf, veclen, 1, fp)
&& ((buf[veclen] = '\0'), 1)
&& (!params.verbose
|| printf("%s: ls check read '%s'\n", name, buf))
&& !memcmp(buf, addr, veclen)) {
/* do nothing */
} else {
unsigned int j;
fprintf(stderr, "%s: ls failed cmp '%s' with '",
name, buf);
for (j = 0; j < veclen; j++) {
putc(((char *) addr)[j], stderr);
}
fprintf(stderr, "'\n");
return 0;
}
} else {
fprintf(stderr, "%s: ls failed\n", name);
return 0;
}
}
if (chash_size(tmphash)) {
fprintf(stderr, "%s: ls failed\n", name);
return 0;
}
} else {
fprintf(stderr, "%s: ls failed\n", name);
return 0;
}
chash_delete(tmphash);
if (params.verbose) {
printf("%s: matched all (%u) entries\n", name, numterms);
}
} else if (!str_casecmp(pos, "set")) {
/* setting the vector for a term in the bucket */
unsigned int veclen,
reallen;
void *addr;
if (!ptr) { return 0; }
/* read parameters */
if ((fscanf(fp, "%65535s %u ", buf, &veclen) == 2)
&& (veclen <= 65535)) {
addr = bucket_find(ptr, bucketsize, strategy, buf,
str_len(buf), &reallen, NULL);
if (addr && (reallen == veclen)
&& fread(addr, 1, veclen, fp)) {
/* do nothing */
if (params.verbose) {
unsigned int j;
printf("%s: set term '%s' to '", name, buf);
for (j = 0; j < reallen; j++) {
putc(((char *) addr)[j], stdout);
}
printf("'\n");
}
} else {
fprintf(stderr, "%s: failed to set!\n", name);
return 0;
}
} else {
fprintf(stderr, "%s: failed to set\n", name);
return 0;
}
} else if (!str_casecmp(pos, "realloc")) {
/* reallocating a term in the bucket */
unsigned int veclen,
succeed;
int toobig;
if (!ptr) { return 0; }
/* read parameters */
if ((fscanf(fp, "%65535s %u %u", buf, &veclen, &succeed) == 3)
&& (veclen <= 65535)) {
if (!bucket_realloc(ptr, bucketsize, strategy, buf,
str_len(buf), veclen, &toobig)) {
fprintf(stderr, "%s: failed to realloc!\n", name);
return 0;
}
} else {
fprintf(stderr, "%s: failed to realloc\n", name);
return 0;
}
if (params.verbose) {
printf("%s: realloc'd term '%s'\n", name, buf);
}
} else if (!str_casecmp(pos, "rm")) {
/* removing something from the bucket */
unsigned int succeed;
if (!ptr) { return 0; }
if (fscanf(fp, "%65535s %u", buf, &succeed) == 2) {
if (succeed) {
if (!(bucket_remove(ptr, bucketsize, strategy, buf,
str_len(buf)))) {
fprintf(stderr, "%s: failed to rm '%s'\n", name,
buf);
return 0;
} else if (params.verbose) {
printf("%s: rm term '%s'\n", name, buf);
}
} else if (succeed) {
fprintf(stderr, "%s: failed to rm\n", name);
return 0;
}
} else {
fprintf(stderr, "%s: failed to rm\n", name);
return 0;
}
} else if (!str_casecmp(pos, "print")) {
/* printing out the bucket contents */
unsigned int state = 0,
len,
veclen;
const char *term;
char format[100];
void *addr;
if (!ptr) {
printf("can't print, no bucket\n");
} else {
do {
term
= bucket_next_term(ptr, bucketsize, strategy, &state,
&len, &addr, &veclen);
} while (term
&& memcpy(buf, term, len)
&& ((buf[len] = '\0') || 1)
&& snprintf(format, 100, "%%.%us (%%u): '%%.%us' (%%u) "
"(off %%u)\n", len, veclen)
&& printf(format, term, len, (char*) addr, veclen,
((char *) addr) - (char *) ptr));
if (!state) {
printf("(empty)\n");
}
printf("%u entries, %u data, %u string, %u overhead, %u free\n",
bucket_entries(ptr, bucketsize, strategy),
bucket_utilised(ptr, bucketsize, strategy),
bucket_string(ptr, bucketsize, strategy),
bucket_overhead(ptr, bucketsize, strategy),
bucket_unused(ptr, bucketsize, strategy));
}
} else if (!str_casecmp(pos, "match")) {
unsigned int veclen,
veclen2;
void *addr;
if (fscanf(fp, "%65535s %u ", buf, &veclen)) {
if ((addr = bucket_find(ptr, bucketsize, strategy,
buf, str_len(buf), &veclen2, NULL))
&& (veclen <= 65535)
&& (veclen2 >= veclen)
&& (!params.verbose
|| printf("%s: match on '%s' ", name, buf))
&& fread(buf, veclen, 1, fp)
&& !memcmp(buf, addr, veclen)) {
if (params.verbose) {
printf("content succeeded\n");
}
} else {
fprintf(stderr, "%s: match failed (%s vs %s)\n", name, buf,
(char *) addr);
return 0;
}
} else {
fprintf(stderr, "%s: match failed\n", name);
return 0;
}
} else if ((*pos != '#') && str_len(pos)) {
fprintf(stderr, "%s: unknown command '%s'\n", name, pos);
return 0;
}
}
if (ptr) {
chash_delete(hash);
free(ptr);
}
return 1;
}
int buckets_accept_bucket(client_t *client, hash_t mask, hash_t hashmask)
{
int accepted = -1;
assert(client);
assert(mask > 0);
if (_bucket_transfer) {
// we are currently transferring another bucket, therefore we cannot accept another one.
logger(LOG_WARN, "cant accept bucket, already transferring one.");
accepted = 0;
}
else {
assert(_bucket_transfer == NULL);
if (mask != _mask) {
// the masks are different, we cannot accept a bucket unless our masks match... which should
// balance out once hashmasks have proceeded through all the nodes.
logger(LOG_WARN, "cant accept bucket, masks are not compatible.");
accepted = 0;
}
else {
// now we need to check that this bucket isn't already handled by this server.
assert(hashmask <= mask);
assert(_mask == mask);
// if we dont currently have a buckets list, we need to create one.
if (_buckets == NULL) {
_buckets = calloc(_mask + 1, sizeof(bucket_t *));
assert(_buckets);
assert(_buckets[0] == NULL);
assert(_primary_buckets == 0);
assert(_secondary_buckets == 0);
}
assert(_buckets);
if (_buckets[hashmask]) {
logger(LOG_ERROR, "cant accept bucket, already have that bucket.");
accepted = 0;
}
else {
logger(LOG_INFO, "accepting bucket. (%#llx/%#llx)", mask, hashmask);
// need to create the bucket, and mark it as in-transit.
bucket_t *bucket = bucket_new(hashmask);
assert(bucket);
_buckets[hashmask] = bucket;
assert(data_in_transit() == 0);
assert(bucket->transfer_client == NULL);
assert(client->node);
logger(LOG_DEBUG, "Setting transfer client ('%s') to bucket %#llx.", node_name(client->node), bucket->hashmask);
bucket->transfer_client = client;
assert(bucket->level < 0);
accepted = 1;
assert(_bucket_transfer == NULL);
_bucket_transfer = bucket;
}
}
}
assert(accepted == 0 || accepted == 1);
return(accepted);
}
// this function will take the current array, and put it aside, creating a new array based on the
// new mask supplied (we can only make the mask bigger, and cannot shrink it).
// We create a new array, and for each entry, we compare it against the old mask, and use
// the data for that hash from the old list.
// NOTE: We may be starting off with an empty lists. If this is the first time we've
// received some hashmasks. To implement this easily, if we dont already have hashmasks, we
// may need to create one that has a dummy entry in it.
void buckets_split_mask(hash_t current_mask, hash_t new_mask)
{
bucket_t **newbuckets = NULL;
bucket_t **oldbuckets = NULL;
int i;
int index;
assert(new_mask > current_mask);
logger(LOG_INFO, "Splitting Mask: Old Mask: %#llx, New Mask; %#llx", current_mask, new_mask);
// grab a copy of the existing buckets as the 'oldbuckets';
oldbuckets = _buckets;
_buckets = NULL;
// make an appropriate sized new buckets list;
newbuckets = malloc(sizeof(bucket_t *) * (new_mask+1));
assert(newbuckets);
// go through every hash for this mask.
for (i=0; i<=new_mask; i++) {
// determine what the old index is.
index = i & current_mask;
// create the new bucket ONLY if we already have a bucket object for that index.
if (oldbuckets == NULL || oldbuckets[index] == NULL) {
newbuckets[i] = NULL;
}
else {
// we have a bucket for this old index. So we need to create a new one for this index.
newbuckets[i] = bucket_new(i);
assert(newbuckets[i]->data);
assert(newbuckets[i]->data->next == NULL);
assert(oldbuckets[index]->data);
assert(oldbuckets[index]->data->ref > 0);
newbuckets[i]->data->next = oldbuckets[index]->data;
oldbuckets[index]->data->ref ++;
assert(newbuckets[i]->hashmask == i);
newbuckets[i]->level = oldbuckets[index]->level;
newbuckets[i]->source_node = oldbuckets[index]->source_node;
newbuckets[i]->backup_node = oldbuckets[index]->backup_node;
newbuckets[i]->logging_node = oldbuckets[index]->logging_node;
newbuckets[i]->primary_node = oldbuckets[index]->primary_node;
newbuckets[i]->secondary_node = oldbuckets[index]->secondary_node;
assert(data_in_transit() == 0);
}
}
assert(_mask >= 0);
assert(new_mask > _mask);
_mask = new_mask;
// clean up the old buckets list.
if (oldbuckets) {
for (i=0; i<=_mask; i++) {
if (oldbuckets[i]) {
assert(oldbuckets[i]->data);
assert(oldbuckets[i]->data->ref > 1);
oldbuckets[i]->data->ref --;
assert(oldbuckets[i]->data->ref > 0);
oldbuckets[i]->data = NULL;
bucket_close(oldbuckets[i]);
}
}
}
_buckets = newbuckets;
assert(_buckets);
}
/*
* heap_buckets_init -- (internal) initializes bucket instances
*/
static int
heap_buckets_init(PMEMobjpool *pop)
{
struct pmalloc_heap *h = pop->heap;
int i;
//printf("calling heap_buckets_init \n");
bucket_proto[0].unit_max = RUN_UNIT_MAX;
/*
* To take use of every single bit available in the run the unit size
* would have to be calculated using following expression:
* (RUNSIZE / (MAX_BITMAP_VALUES * BITS_PER_VALUE)), but to preserve
* cacheline alignment a little bit of memory at the end of the run
* is left unused.
*/
bucket_proto[0].unit_size = MIN_RUN_SIZE;
for (i = 1; i < MAX_BUCKETS - 1; ++i) {
bucket_proto[i].unit_max = RUN_UNIT_MAX;
bucket_proto[i].unit_size =
bucket_proto[i - 1].unit_size *
bucket_proto[i - 1].unit_max;
}
bucket_proto[i].unit_max = -1;
bucket_proto[i].unit_size = CHUNKSIZE;
h->last_run_max_size = bucket_proto[i - 1].unit_size *
(bucket_proto[i - 1].unit_max - 1);
h->bucket_map = Malloc(sizeof (*h->bucket_map) * h->last_run_max_size);
if (h->bucket_map == NULL)
goto error_bucket_map_malloc;
for (i = 0; i < MAX_BUCKETS; ++i) {
h->buckets[i] = bucket_new(bucket_proto[i].unit_size,
bucket_proto[i].unit_max);
if (h->buckets[i] == NULL)
goto error_bucket_new;
}
/* XXX better way to fill the bucket map */
for (i = 0; i < h->last_run_max_size; ++i) {
for (int j = 0; j < MAX_BUCKETS - 1; ++j) {
/*
* Skip the last unit, so that the distribution
* of buckets in the map is better.
*/
if ((bucket_proto[j].unit_size *
((bucket_proto[j].unit_max - 1))) >= i) {
h->bucket_map[i] = h->buckets[j];
break;
}
}
}
heap_populate_buckets(pop);
return 0;
error_bucket_new:
Free(h->bucket_map);
for (i = i - 1; i >= 0; --i)
bucket_delete(h->buckets[i]);
error_bucket_map_malloc:
return ENOMEM;
}
