Skip to content

Commit

Permalink
Try to load a minimum amount of free-space, fix timing measurements o…
Browse files Browse the repository at this point in the history 
…f sync, return errors on bad usage of xcursor
  • Loading branch information
@kriszyp
kriszyp committed Mar 7, 2024
1 parent 23cdfee commit 78a00ef
Showing 1 changed file with 139 additions and 87 deletions.
226 changes: 139 additions & 87 deletions dependencies/lmdb/libraries/liblmdb/mdb.c
View file
Original file line number Diff line number Diff line change
Expand Up @@ -2757,100 +2757,103 @@ mdb_page_alloc(MDB_cursor *mc, int num, MDB_page **mp)
unsigned block_size = 0;
ssize_t entry;
empty_entries = 0;
//mdb_midl_print(stderr, mop);
// TODO: Skip this on the first iteration, since we already checked the cache
int start = env->me_freelist_position != 0 ? env->me_freelist_position : 1;
if (start < 0) start = -start;
if (start > mop_len) {
start = 1;
}
if (mop_len && (ssize_t) mop[start - 1] < 0) start++; // don't start in the middle of a length block
unsigned end = start + mop_len;
unsigned check_point = start + MAX_SCAN_SEGMENT;
if (check_point > mop_len) check_point = mop_len;
unsigned wrapped = 0;
// TODO: Don't scan if the list is too small
//fprintf(stderr, "loop from %u to %u over %u\n", start, end, mop_len);
for(i = start; 1; i++) {
//fprintf(stderr, "l %u ", i);
if (i > check_point) {
unsigned counting_i = i + (wrapped ? mop_len : 0);
if (best_fit_start > 0 && best_fit_size - num < (counting_i - start) >> 5) {
goto continue_best_fit;
if (env->me_pgoldest > 0 || mop_len > 100) { // don't scan too small of freelists (unless already started)
int start = env->me_freelist_position != 0 ? env->me_freelist_position : 1;
if (start < 0) start = -start;
if (start > mop_len) {
start = 1;
}
if (mop_len && (ssize_t) mop[start - 1] < 0) start++; // don't start in the middle of a length block
unsigned end = start + mop_len;
unsigned check_point = start + MAX_SCAN_SEGMENT;
if (check_point > mop_len) check_point = mop_len;
unsigned wrapped = 0;
// TODO: Don't scan if the list is too small
//fprintf(stderr, "loop from %u to %u over %u\n", start, end, mop_len);
for (i = start; 1; i++) {
//fprintf(stderr, "l %u ", i);
if (i > check_point) {
unsigned counting_i = i + (wrapped ? mop_len : 0);
if (best_fit_start > 0 && best_fit_size - num < (counting_i - start) >> 5) {
goto continue_best_fit;
}
if (counting_i - start >= mop_len) {
break; // searched everything in current list
}
if (i > mop_len) {
// loop back and continue
i = 1;
wrapped = 1;
check_point = start - 1;
if (check_point > MAX_SCAN_SEGMENT) check_point = MAX_SCAN_SEGMENT;
} else {
check_point = i + MAX_SCAN_SEGMENT;
if (wrapped) {
if (check_point > start - 1) check_point = start - 1;
} else if (check_point > mop_len) check_point = mop_len;
}
}
if (counting_i - start >= mop_len) {
break; // searched everything in current list

entry = mop[i];
//fprintf(stderr, "pgno %u next would be %u\n", entry, block_start + block_size);
if (entry == 0) {
empty_entries++;
continue;
}
if (i > mop_len) {
// loop back and continue
i = 1;
wrapped = 1;
check_point = start - 1;
if (check_point > MAX_SCAN_SEGMENT) check_point = MAX_SCAN_SEGMENT;
if (entry > 0) {
pgno = entry;
block_size = 1;
} else {
check_point = i + MAX_SCAN_SEGMENT;
if (wrapped) {
if (check_point > start - 1) check_point = start -1;
} else if (check_point > mop_len) check_point = mop_len;
block_size = -entry;
pgno = mop[++i];
}
}

entry = mop[i];
//fprintf(stderr, "pgno %u next would be %u\n", entry, block_start + block_size);
if (entry == 0) {
empty_entries++;
continue;
}
if (entry > 0) {
pgno = entry;
block_size = 1;
} else {
block_size = -entry;
pgno = mop[++i];
}

if (block_size >= num) {
if (block_size == num) {
// we found a block of the right size
env->me_freelist_position = i + 1;
mop[i] = 0;
if (env->me_freelist_written_end < i) env->me_freelist_written_end = i;
if (entry < 1)
mop[--i] = 0;
if (env->me_freelist_written_start > i || !env->me_freelist_written_start) env->me_freelist_written_start = i;
goto search_done;
} else if (block_size < best_fit_size || best_fit_size == 0) {
best_fit_start = i - 1;
best_fit_size = block_size;
if (i == 1 - env->me_freelist_position) {
// if we just wrote to this block and we are continuing on this block,
// skip ahead to using this block
goto continue_best_fit;
if (block_size >= num) {
if (block_size == num) {
// we found a block of the right size
env->me_freelist_position = i + 1;
mop[i] = 0;
if (env->me_freelist_written_end < i) env->me_freelist_written_end = i;
if (entry < 1)
mop[--i] = 0;
if (env->me_freelist_written_start > i || !env->me_freelist_written_start)
env->me_freelist_written_start = i;
goto search_done;
} else if (block_size < best_fit_size || best_fit_size == 0) {
best_fit_start = i - 1;
best_fit_size = block_size;
if (i == 1 - env->me_freelist_position) {
// if we just wrote to this block and we are continuing on this block,
// skip ahead to using this block
goto continue_best_fit;
}
}
}
}
// see if there is too many empty entries; after a respread it should be between one third and one quarter full of empty entries
if (empty_entries > ((i + (wrapped ? mop_len : 0) - start) / 3) + 10) {
unsigned old_length = env->me_pghead[0];
mdb_midl_respread(&env->me_pghead);
mop = env->me_pghead;
mop_len = mop[0];
// consider the whole free-list to be updated now
env->me_freelist_written_end = 0x7fffffff;
env->me_freelist_written_start = 1;
//fprintf(stderr, "resized from %u to %u\n", old_length, mop_len);
goto restart_search;
}
env->me_freelist_position = i;
} else {
fprintf(stderr, "Free list too small, trying to load more\n");
}
// see if there is too many empty entries; after a respread it should be between one third and one quarter full of empty entries
if (empty_entries > ((i + (wrapped ? mop_len : 0) - start) / 3) + 10) {
unsigned old_length = env->me_pghead[0];
mdb_midl_respread(&env->me_pghead);
mop = env->me_pghead;
mop_len = mop[0];
// consider the whole free-list to be updated now
env->me_freelist_written_end = 0x7fffffff;
env->me_freelist_written_start = 1;
//fprintf(stderr, "resized from %u to %u\n", old_length, mop_len);
goto restart_search;
}
env->me_freelist_position = i;
i = 0;

if (mop_len > env->me_maxfreepgs_to_load) {
//fprintf(stderr, "Too many entries %u, looking for %u, best fit %u, not loading anymore\n", mop_len, num, best_fit_size);
//mdb_midl_print(stderr, mop);
goto continue_best_fit;
}

load_more:
if (op == MDB_SET_RANGE) { /* 1st iteration */
/* Prepare to fetch more and coalesce */
oldest = env->me_pgoldest;
Expand Down Expand Up @@ -3278,10 +3281,6 @@ mdb_env_sync0(MDB_env *env, int force, pgno_t numpgs)
int rc = 0;
if (env->me_flags & MDB_RDONLY)
return EACCES;
uint64_t start = 0;
if (env->me_flags & MDB_TRACK_METRICS) {
start = get_time64();
}

if (force || !(env->me_flags & MDB_NOSYNC)
#ifdef _WIN32 /* Sync is normally achieved in Windows by doing WRITE_THROUGH writes */
Expand All @@ -3308,9 +3307,6 @@ mdb_env_sync0(MDB_env *env, int force, pgno_t numpgs)
rc = ErrCode();
}
}
if (env->me_flags & MDB_TRACK_METRICS) {
env->me_metrics.time_sync += get_time64() - start;
}
return rc;
}

Expand All @@ -3330,6 +3326,10 @@ mdb_env_sync(MDB_env *env, int force)
UNLOCK_MUTEX(env->me_sync_mutex);
return 0;
}
uint64_t start = 0;
if (env->me_flags & MDB_TRACK_METRICS) {
start = get_time64();
}
MDB_txn sync_txn;
MDB_db dbs[2];
do {
Expand All @@ -3344,6 +3344,10 @@ mdb_env_sync(MDB_env *env, int force)
sync_txn.mt_next_pgno = m->mm_last_pg + 1;
} while(ti->mti_txnid != last_txn_id); // avoid race condition in copying data by verifying that this is updated
rc = mdb_env_sync0(env, force, sync_txn.mt_next_pgno);
if (env->me_flags & MDB_TRACK_METRICS) {
env->me_metrics.time_sync += get_time64() - start;
}

if (rc) {
UNLOCK_MUTEX(env->me_sync_mutex);
return rc;
Expand All @@ -3355,7 +3359,15 @@ mdb_env_sync(MDB_env *env, int force)
UNLOCK_MUTEX(env->me_sync_mutex);
return rc;
} else {
return mdb_env_sync0(env, force, m->mm_last_pg+1);
uint64_t start = 0;
if (env->me_flags & MDB_TRACK_METRICS) {
start = get_time64();
}
int rc = mdb_env_sync0(env, force, m->mm_last_pg+1);
if (env->me_flags & MDB_TRACK_METRICS) {
env->me_metrics.time_sync += get_time64() - start;
}
return rc;
}
// </lmdb-js>
}
Expand Down Expand Up @@ -4961,6 +4973,10 @@ mdb_txn_commit(MDB_txn *txn)
}
if (rc)
return rc;
uint64_t start = 0;
if (env->me_flags & MDB_TRACK_METRICS) {
start = get_time64();
}
rc = mdb_env_sync0(env, 0, sync_txn.mt_next_pgno);
if (rc) {
UNLOCK_MUTEX(env->me_sync_mutex);
Expand All @@ -4969,6 +4985,9 @@ mdb_txn_commit(MDB_txn *txn)
rc = mdb_env_write_meta(&sync_txn);
if (rc == 0)
env->me_synced_txn_id = sync_txn.mt_txnid;
if (env->me_flags & MDB_TRACK_METRICS) {
env->me_metrics.time_sync += get_time64() - start;
}
UNLOCK_MUTEX(env->me_sync_mutex);
} else
mdb_txn_end(txn, end_mode);
Expand Down Expand Up @@ -5256,7 +5275,7 @@ mdb_env_write_meta(MDB_txn *txn)
off = offsetof(MDB_meta, mm_mapsize);
ptr = (char *)&meta + off;
len = sizeof(MDB_meta) - off;
if (flags & 2) {
if (flags & 2) { // write to the sync meta, half page
off += PAGEHDRSZ + (env->me_psize >> 1);
} else
off += (char *)mp - env->me_map;
Expand Down Expand Up @@ -8385,6 +8404,10 @@ mdb_cursor_next(MDB_cursor *mc, MDB_val *key, MDB_val *data, MDB_cursor_op op)
leaf = NODEPTR(mp, mc->mc_ki[mc->mc_top]);

if (F_ISSET(leaf->mn_flags, F_DUPDATA)) {
if (!mc->mc_xcursor) {
last_error = "Invalid dupdata flag with no mc_xcursor";
return MDB_BAD_TXN;
}
mdb_xcursor_init1(mc, leaf);
rc = mdb_cursor_first(&mc->mc_xcursor->mx_cursor, data, NULL);
if (rc != MDB_SUCCESS)
Expand Down Expand Up @@ -8470,6 +8493,10 @@ mdb_cursor_prev(MDB_cursor *mc, MDB_val *key, MDB_val *data, MDB_cursor_op op)
leaf = NODEPTR(mp, mc->mc_ki[mc->mc_top]);

if (F_ISSET(leaf->mn_flags, F_DUPDATA)) {
if (!mc->mc_xcursor) {
last_error = "Invalid dupdata flag with no mc_xcursor";
return MDB_BAD_TXN;
}
mdb_xcursor_init1(mc, leaf);
rc = mdb_cursor_last(&mc->mc_xcursor->mx_cursor, data, NULL);
if (rc != MDB_SUCCESS)
Expand Down Expand Up @@ -8633,6 +8660,10 @@ mdb_cursor_set(MDB_cursor *mc, MDB_val *key, MDB_val *data,
}

if (F_ISSET(leaf->mn_flags, F_DUPDATA)) {
if (!mc->mc_xcursor) {
last_error = "Invalid dupdata flag with no mc_xcursor";
return MDB_BAD_TXN;
}
mdb_xcursor_init1(mc, leaf);
if (op == MDB_SET || op == MDB_SET_KEY || op == MDB_SET_RANGE) {
rc = mdb_cursor_first(&mc->mc_xcursor->mx_cursor, data, NULL);
Expand Down Expand Up @@ -8715,6 +8746,10 @@ mdb_cursor_first(MDB_cursor *mc, MDB_val *key, MDB_val *data)
}

if (F_ISSET(leaf->mn_flags, F_DUPDATA)) {
if (!mc->mc_xcursor) {
last_error = "Invalid dupdata flag with no mc_xcursor";
return MDB_BAD_TXN;
}
mdb_xcursor_init1(mc, leaf);
rc = mdb_cursor_first(&mc->mc_xcursor->mx_cursor, data, NULL);
if (rc)
Expand Down Expand Up @@ -8760,6 +8795,10 @@ mdb_cursor_last(MDB_cursor *mc, MDB_val *key, MDB_val *data)
}

if (F_ISSET(leaf->mn_flags, F_DUPDATA)) {
if (!mc->mc_xcursor) {
last_error = "Invalid dupdata flag with no mc_xcursor";
return MDB_BAD_TXN;
}
mdb_xcursor_init1(mc, leaf);
rc = mdb_cursor_last(&mc->mc_xcursor->mx_cursor, data, NULL);
if (rc)
Expand Down Expand Up @@ -9443,6 +9482,10 @@ mdb_cursor_put(MDB_cursor *mc, MDB_val *key, MDB_val *data,
if ((flags & (MDB_CURRENT|MDB_APPENDDUP)) == MDB_CURRENT) {
xflags = MDB_CURRENT|MDB_NOSPILL;
} else {
if (!mc->mc_xcursor) {
last_error = "Invalid dupdata flag with no mc_xcursor";
return MDB_BAD_TXN;
}
mdb_xcursor_init1(mc, leaf);
xflags = (flags & MDB_NODUPDATA) ?
MDB_NOOVERWRITE|MDB_NOSPILL : MDB_NOSPILL;
Expand Down Expand Up @@ -10994,6 +11037,11 @@ mdb_cursor_del0(MDB_cursor *mc)
* Else (xcursor was initd, not a subDB) needs mc_pg[0] reset.
*/
if (node->mn_flags & F_DUPDATA) {
if (!m3->mc_xcursor) {
last_error = "Invalid dupdata flag with no mc_xcursor";
rc = MDB_BAD_TXN;
goto fail;
}
if (m3->mc_xcursor->mx_cursor.mc_flags & C_INITIALIZED) {
if (!(node->mn_flags & F_SUBDATA))
m3->mc_xcursor->mx_cursor.mc_pg[0] = NODEDATA(node);
Expand Down Expand Up @@ -12542,6 +12590,10 @@ mdb_drop0(MDB_cursor *mc, int subs)
if (!mc->mc_db->md_overflow_pages && !subs)
break;
} else if (subs && (ni->mn_flags & F_SUBDATA)) {
if (!mc->mc_xcursor) {
last_error = "Invalid dupdata flag with no mc_xcursor";
return MDB_BAD_TXN;
}
mdb_xcursor_init1(mc, ni);
rc = mdb_drop0(&mc->mc_xcursor->mx_cursor, 0);
if (rc)
Expand Down

0 comments on commit 78a00ef

Please sign in to comment.