void bc_optimize(Bitcask *bc, int limit)
{
int i;
for (i=0; i < bc->curr; i++) {
char data[20], hint[20], datapath[255], hintpath[255];
sprintf(data, DATA_FILE, i);
sprintf(hint, HINT_FILE, i);
sprintf(datapath, "%s/%s", mgr_alloc(bc->mgr, data), data);
sprintf(hintpath, "%s/%s", mgr_alloc(bc->mgr, hint), hint);
uint32_t recoverd = 0;
HTree *cur_tree = optimizeDataFile(bc->tree, i, datapath, hintpath, limit, &recoverd);
if (NULL == cur_tree) continue;
pthread_mutex_lock(&bc->buffer_lock);
bc->bytes -= recoverd;
pthread_mutex_unlock(&bc->buffer_lock);
pthread_mutex_lock(&bc->write_lock);
ht_visit(cur_tree, update_items, bc->tree);
pthread_mutex_unlock(&bc->write_lock);
build_hint(cur_tree, hintpath);
}
}
void bc_optimize(Bitcask *bc, int limit)
{
int i, total, last = -1;
bc->optimize_flag = 1;
const char *base = mgr_base(bc->mgr);
// remove htree
for (i=0; i < bc->curr; i++)
{
mgr_unlink(gen_path(NULL, base, HTREE_FILE, i));
}
bc->last_snapshot = -1;
time_t limit_time = 0;
if (limit > 3600 * 24 * 365 * 10) // more than 10 years
{
limit_time = limit; // absolute time
}
else
{
limit_time = time(NULL) - limit; // relative time
}
struct stat st;
bool skipped = false;
for (i=0; i < bc->curr && bc->optimize_flag == 1; i++)
{
char datapath[255], hintpath[255];
gen_path(datapath, base, DATA_FILE, i);
gen_path(hintpath, base, HINT_FILE, i);
if (stat(datapath, &st) != 0)
{
continue; // skip empty file
}
// skip recent modified file
if (st.st_mtime > limit_time)
{
skipped = true;
last ++;
if (last != i) // rotate data file
{
char npath[255];
gen_path(npath, base, DATA_FILE, last);
if (symlink(datapath, npath) != 0)
{
fprintf(stderr, "symlink failed: %s -> %s\n", datapath, npath);
last = i;
continue;
}
// update HTree to use new index
if (stat(hintpath, &st) != 0)
{
fprintf(stderr, "no hint file: %s, skip it\n", hintpath);
last = i;
continue;
}
HTree *tree = ht_new(bc->depth, bc->pos);
scanHintFile(tree, i, hintpath, NULL);
struct update_args args;
args.tree = bc->tree;
args.index = last;
ht_visit(tree, update_item_pos, &args);
ht_destroy(tree);
unlink(npath);
mgr_rename(datapath, npath);
mgr_rename(hintpath, gen_path(npath, base, HINT_FILE, last));
}
continue;
}
int deleted = count_deleted_record(bc->tree, i, hintpath, &total);
uint64_t curr_size = data_file_size(bc, i) * (total - deleted/2) / (total+1); // guess
uint64_t last_size = last >= 0 ? data_file_size(bc, last) : -1;
// last data file size
uint32_t recoverd = 0;
if (last == -1 || last_size + curr_size > MAX_BUCKET_SIZE)
{
last ++;
}
while (last < i)
{
char ldpath[255], lhpath[255];
new_path(ldpath, bc->mgr, DATA_FILE, last);
new_path(lhpath, bc->mgr, HINT_FILE, last);
recoverd = optimizeDataFile(bc->tree, i, datapath, hintpath,
skipped, MAX_BUCKET_SIZE, last, ldpath, lhpath);
if (recoverd == 0)
{
last ++;
}
else
{
break;
}
}
if (recoverd == 0)
{
// last == i
recoverd = optimizeDataFile(bc->tree, i, datapath, hintpath,
skipped, MAX_BUCKET_SIZE, last, NULL, NULL);
}
if (recoverd < 0) break; // failed
pthread_mutex_lock(&bc->buffer_lock);
bc->bytes -= recoverd;
pthread_mutex_unlock(&bc->buffer_lock);
}
// update pos of items in curr_tree
pthread_mutex_lock(&bc->write_lock);
pthread_mutex_lock(&bc->flush_lock);
if (i == bc->curr && ++last < bc->curr)
{
char opath[255], npath[255];
gen_path(opath, base, DATA_FILE, bc->curr);
if (file_exists(opath))
{
gen_path(npath, base, DATA_FILE, last);
if (symlink(opath, npath) != 0)
fprintf(stderr, "symlink failed: %s -> %s\n", opath, npath);
struct update_args args;
args.tree = bc->tree;
args.index = last;
ht_visit(bc->curr_tree, update_item_pos, &args);
unlink(npath);
mgr_rename(opath, npath);
}
bc->curr = last;
}
pthread_mutex_unlock(&bc->flush_lock);
pthread_mutex_unlock(&bc->write_lock);
bc->optimize_flag = 0;
}