NextGraph
/
lmdb
mirror of https://github.com/nextgraph-org/lmdb.git
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

RPAGE_CACHE is now dynamically selectable

Browse Source 
Behavior is controlled by MDB_REMAP_CHUNKS envflag
Remapping is always enabled in MDB_VL32
mdb.master3
Howard Chu 8 years ago
parent ac047d1eff
commit 53799e51da
2 changed files with 81 additions and 54 deletions
Show all changes Ignore whitespace when comparing lines Ignore changes in amount of whitespace Ignore changes in whitespace at EOL
Show Stats Download Patch File Download Diff File
  1. 3
      libraries/liblmdb/lmdb.h
  2. 102
      libraries/liblmdb/mdb.c

3
libraries/liblmdb/lmdb.h
Unescape View File

@ -188,6 +188,7 @@ typedef mode_t mdb_mode_t;
#endif #endif
#if !defined(MDB_RPAGE_CACHE) || (defined(MDB_VL32) && !(MDB_RPAGE_CACHE)) #if !defined(MDB_RPAGE_CACHE) || (defined(MDB_VL32) && !(MDB_RPAGE_CACHE))
/** Support #MDB_REMAP_CHUNKS. Implied by MDB_VL32. Define as 0 to disable. */
#define MDB_RPAGE_CACHE 1 #define MDB_RPAGE_CACHE 1
#endif #endif
@ -336,6 +337,8 @@ typedef void (MDB_rel_func)(MDB_val *item, void *oldptr, void *newptr, void *rel
#define MDB_NORDAHEAD 0x800000 #define MDB_NORDAHEAD 0x800000
/** don't initialize malloc'd memory before writing to datafile */ /** don't initialize malloc'd memory before writing to datafile */
#define MDB_NOMEMINIT 0x1000000 #define MDB_NOMEMINIT 0x1000000
/** don't use a single mmap, remap individual chunks (needs MDB_RPAGE_CACHE) */
#define MDB_REMAP_CHUNKS 0x2000000
/** @} */ /** @} */
/** @defgroup mdb_dbi_open Database Flags /** @defgroup mdb_dbi_open Database Flags

102
libraries/liblmdb/mdb.c
Unescape View File

@ -2051,6 +2051,12 @@ mdb_dlist_free(MDB_txn *txn)
} }
#if MDB_RPAGE_CACHE #if MDB_RPAGE_CACHE
#if defined(MDB_VL32) || ((MDB_RPAGE_CACHE) & 2) /* Always remap */
#define MDB_REMAPPING(flags) 1
#else
#define MDB_REMAPPING(flags) ((flags) & MDB_REMAP_CHUNKS)
#endif
static void static void
mdb_page_unref(MDB_txn *txn, MDB_page *mp) mdb_page_unref(MDB_txn *txn, MDB_page *mp)
{ {
@ -2067,7 +2073,8 @@ mdb_page_unref(MDB_txn *txn, MDB_page *mp)
if (tl[x].mref) if (tl[x].mref)
tl[x].mref--; tl[x].mref--;
} }
#define MDB_PAGE_UNREF(txn, mp) mdb_page_unref(txn, mp) #define MDB_PAGE_UNREF(txn, mp) \
(MDB_REMAPPING(txn->mt_env->me_flags) ? mdb_page_unref(txn, mp) : (void)0)
static void static void
mdb_cursor_unref(MDB_cursor *mc) mdb_cursor_unref(MDB_cursor *mc)
@ -2088,11 +2095,13 @@ mdb_cursor_unref(MDB_cursor *mc)
mc->mc_flags &= ~C_INITIALIZED; mc->mc_flags &= ~C_INITIALIZED;
} }
#define MDB_CURSOR_UNREF(mc, force) \ #define MDB_CURSOR_UNREF(mc, force) \
(((force) || ((mc)->mc_flags & C_INITIALIZED)) \ ((MDB_REMAPPING((mc)->mc_txn->mt_env->me_flags) && \
((force) || ((mc)->mc_flags & C_INITIALIZED))) \
? mdb_cursor_unref(mc) \ ? mdb_cursor_unref(mc) \
: (void)0) : (void)0)
#else #else
#define MDB_REMAPPING(flags) 0
#define MDB_PAGE_UNREF(txn, mp) #define MDB_PAGE_UNREF(txn, mp)
#define MDB_CURSOR_UNREF(mc, force) ((void)0) #define MDB_CURSOR_UNREF(mc, force) ((void)0)
#endif /* MDB_RPAGE_CACHE */ #endif /* MDB_RPAGE_CACHE */
@ -2554,8 +2563,8 @@ mdb_page_alloc(MDB_cursor *mc, int num, MDB_page **mp)
rc = MDB_MAP_FULL; rc = MDB_MAP_FULL;
goto fail; goto fail;
} }
#if defined(_WIN32) && !MDB_RPAGE_CACHE #if defined(_WIN32)
if (!(env->me_flags & MDB_RDONLY)) { if (!MDB_REMAPPING(env->me_flags) && !(env->me_flags & MDB_RDONLY)) {
void *p; void *p;
p = (MDB_page *)(env->me_map + env->me_psize * pgno); p = (MDB_page *)(env->me_map + env->me_psize * pgno);
p = VirtualAlloc(p, env->me_psize * num, MEM_COMMIT, p = VirtualAlloc(p, env->me_psize * num, MEM_COMMIT,
@ -3068,6 +3077,7 @@ mdb_txn_renew0(MDB_txn *txn)
/* Moved to here to avoid a data race in read TXNs */ /* Moved to here to avoid a data race in read TXNs */
txn->mt_next_pgno = meta->mm_last_pg+1; txn->mt_next_pgno = meta->mm_last_pg+1;
#if MDB_RPAGE_CACHE #if MDB_RPAGE_CACHE
if (MDB_REMAPPING(env->me_flags))
txn->mt_last_pgno = txn->mt_next_pgno - 1; txn->mt_last_pgno = txn->mt_next_pgno - 1;
#endif #endif
@ -3149,7 +3159,7 @@ mdb_txn_begin(MDB_env *env, MDB_txn *parent, unsigned int flags, MDB_txn **ret)
return ENOMEM; return ENOMEM;
} }
#if MDB_RPAGE_CACHE #if MDB_RPAGE_CACHE
if (!parent) { if (MDB_REMAPPING(env->me_flags) && !parent) {
txn->mt_rpages = malloc(MDB_TRPAGE_SIZE * sizeof(MDB_ID3)); txn->mt_rpages = malloc(MDB_TRPAGE_SIZE * sizeof(MDB_ID3));
if (!txn->mt_rpages) { if (!txn->mt_rpages) {
free(txn); free(txn);
@ -3187,6 +3197,7 @@ mdb_txn_begin(MDB_env *env, MDB_txn *parent, unsigned int flags, MDB_txn **ret)
txn->mt_parent = parent; txn->mt_parent = parent;
txn->mt_numdbs = parent->mt_numdbs; txn->mt_numdbs = parent->mt_numdbs;
#if MDB_RPAGE_CACHE #if MDB_RPAGE_CACHE
if (MDB_REMAPPING(env->me_flags))
txn->mt_rpages = parent->mt_rpages; txn->mt_rpages = parent->mt_rpages;
#endif #endif
memcpy(txn->mt_dbs, parent->mt_dbs, txn->mt_numdbs * sizeof(MDB_db)); memcpy(txn->mt_dbs, parent->mt_dbs, txn->mt_numdbs * sizeof(MDB_db));
@ -3216,6 +3227,7 @@ renew:
if (rc) { if (rc) {
if (txn != env->me_txn0) { if (txn != env->me_txn0) {
#if MDB_RPAGE_CACHE #if MDB_RPAGE_CACHE
if (MDB_REMAPPING(env->me_flags))
free(txn->mt_rpages); free(txn->mt_rpages);
#endif #endif
free(txn); free(txn);
@ -3345,7 +3357,7 @@ mdb_txn_end(MDB_txn *txn, unsigned mode)
mdb_midl_free(pghead); mdb_midl_free(pghead);
} }
#if MDB_RPAGE_CACHE #if MDB_RPAGE_CACHE
if (!txn->mt_parent) { if (MDB_REMAPPING(env->me_flags) && !txn->mt_parent) {
MDB_ID3L el = env->me_rpages, tl = txn->mt_rpages; MDB_ID3L el = env->me_rpages, tl = txn->mt_rpages;
unsigned i, x, n = tl[0].mid; unsigned i, x, n = tl[0].mid;
pthread_mutex_lock(&env->me_rpmutex); pthread_mutex_lock(&env->me_rpmutex);
@ -3722,7 +3734,7 @@ retry_seek:
#endif /* _WIN32 */ #endif /* _WIN32 */
} }
#if MDB_RPAGE_CACHE #if MDB_RPAGE_CACHE
if (pgno > txn->mt_last_pgno) if (MDB_REMAPPING(env->me_flags) && pgno > txn->mt_last_pgno)
txn->mt_last_pgno = pgno; txn->mt_last_pgno = pgno;
#endif #endif
@ -4327,20 +4339,20 @@ mdb_env_map(MDB_env *env, void *addr)
if (rc) if (rc)
return mdb_nt2win32(rc); return mdb_nt2win32(rc);
map = addr; map = addr;
#if MDB_RPAGE_CACHE if (MDB_REMAPPING(env->me_flags))
msize = NUM_METAS * env->me_psize; msize = NUM_METAS * env->me_psize;
#endif
rc = NtMapViewOfSection(mh, GetCurrentProcess(), &map, 0, 0, NULL, &msize, ViewUnmap, alloctype, pageprot); rc = NtMapViewOfSection(mh, GetCurrentProcess(), &map, 0, 0, NULL, &msize, ViewUnmap, alloctype, pageprot);
#if MDB_RPAGE_CACHE #if MDB_RPAGE_CACHE
if (MDB_REMAPPING(env->me_flags))
env->me_fmh = mh; env->me_fmh = mh;
#else else
NtClose(mh);
#endif #endif
NtClose(mh);
if (rc) if (rc)
return mdb_nt2win32(rc); return mdb_nt2win32(rc);
env->me_map = map; env->me_map = map;
#else #else /* !_WIN32 */
#if MDB_RPAGE_CACHE if (MDB_REMAPPING(env->me_flags)) {
(void) flags; (void) flags;
env->me_map = mmap(addr, NUM_METAS * env->me_psize, PROT_READ, MAP_SHARED, env->me_map = mmap(addr, NUM_METAS * env->me_psize, PROT_READ, MAP_SHARED,
env->me_fd, 0); env->me_fd, 0);
@ -4348,7 +4360,8 @@ mdb_env_map(MDB_env *env, void *addr)
env->me_map = NULL; env->me_map = NULL;
return ErrCode(); return ErrCode();
} }
#else } else
{
int prot = PROT_READ; int prot = PROT_READ;
if (flags & MDB_WRITEMAP) { if (flags & MDB_WRITEMAP) {
prot |= PROT_WRITE; prot |= PROT_WRITE;
@ -4372,6 +4385,7 @@ mdb_env_map(MDB_env *env, void *addr)
#endif /* POSIX_MADV_RANDOM */ #endif /* POSIX_MADV_RANDOM */
#endif /* MADV_RANDOM */ #endif /* MADV_RANDOM */
} }
}
#endif /* _WIN32 */ #endif /* _WIN32 */
/* Can happen because the address argument to mmap() is just a /* Can happen because the address argument to mmap() is just a
@ -4381,7 +4395,6 @@ mdb_env_map(MDB_env *env, void *addr)
*/ */
if (addr && env->me_map != addr) if (addr && env->me_map != addr)
return EBUSY; /* TODO: Make a new MDB_* error code? */ return EBUSY; /* TODO: Make a new MDB_* error code? */
#endif
p = (MDB_page *)env->me_map; p = (MDB_page *)env->me_map;
env->me_metas[0] = METADATA(p); env->me_metas[0] = METADATA(p);
@ -4398,10 +4411,9 @@ mdb_env_set_mapsize(MDB_env *env, mdb_size_t size)
*/ */
if (env->me_map) { if (env->me_map) {
MDB_meta *meta; MDB_meta *meta;
#if !MDB_RPAGE_CACHE
void *old; void *old;
int rc; int rc;
#endif
if (env->me_txn) if (env->me_txn)
return EINVAL; return EINVAL;
meta = mdb_env_pick_meta(env); meta = mdb_env_pick_meta(env);
@ -4413,8 +4425,9 @@ mdb_env_set_mapsize(MDB_env *env, mdb_size_t size)
if (size < minsize) if (size < minsize)
size = minsize; size = minsize;
} }
#if !MDB_RPAGE_CACHE if (!(MDB_REMAPPING(env->me_flags)))
/* For MDB_RPAGE_CACHE this bit is a noop since we dynamically remap {
/* For MDB_REMAP_CHUNKS this bit is a noop since we dynamically remap
* chunks of the DB anyway. * chunks of the DB anyway.
*/ */
munmap(env->me_map, env->me_mapsize); munmap(env->me_map, env->me_mapsize);
@ -4423,7 +4436,7 @@ mdb_env_set_mapsize(MDB_env *env, mdb_size_t size)
rc = mdb_env_map(env, old); rc = mdb_env_map(env, old);
if (rc) if (rc)
return rc; return rc;
#endif /* !MDB_RPAGE_CACHE */ }
} }
env->me_mapsize = size; env->me_mapsize = size;
if (env->me_psize) if (env->me_psize)
@ -5332,7 +5345,7 @@ fail:
*/ */
#define CHANGEABLE (MDB_NOSYNC|MDB_NOMETASYNC|MDB_MAPASYNC|MDB_NOMEMINIT) #define CHANGEABLE (MDB_NOSYNC|MDB_NOMETASYNC|MDB_MAPASYNC|MDB_NOMEMINIT)
#define CHANGELESS (MDB_FIXEDMAP|MDB_NOSUBDIR|MDB_RDONLY| \ #define CHANGELESS (MDB_FIXEDMAP|MDB_NOSUBDIR|MDB_RDONLY| \
MDB_WRITEMAP|MDB_NOTLS|MDB_NOLOCK|MDB_NORDAHEAD) MDB_WRITEMAP|MDB_NOTLS|MDB_NOLOCK|MDB_NORDAHEAD|MDB_REMAP_CHUNKS)
#if VALID_FLAGS & PERSISTENT_FLAGS & (CHANGEABLE|CHANGELESS) #if VALID_FLAGS & PERSISTENT_FLAGS & (CHANGEABLE|CHANGELESS)
# error "Persistent DB flags & env flags overlap, but both go in mm_flags" # error "Persistent DB flags & env flags overlap, but both go in mm_flags"
@ -5347,23 +5360,23 @@ mdb_env_open(MDB_env *env, const char *path, unsigned int flags, mdb_mode_t mode
if (env->me_fd!=INVALID_HANDLE_VALUE || (flags & ~(CHANGEABLE|CHANGELESS))) if (env->me_fd!=INVALID_HANDLE_VALUE || (flags & ~(CHANGEABLE|CHANGELESS)))
return EINVAL; return EINVAL;
#if MDB_RPAGE_CACHE flags |= env->me_flags;
if (flags & MDB_WRITEMAP) { if (MDB_REMAPPING(0)) /* if we always remap chunks */
/* silently ignore WRITEMAP with RPAGE_CACHE */ flags |= MDB_REMAP_CHUNKS;
flags ^= MDB_WRITEMAP; if (MDB_REMAPPING(flags)) {
} /* silently ignore WRITEMAP with REMAP_CHUNKS */
if (flags & MDB_FIXEDMAP) { flags &= ~MDB_WRITEMAP;
/* cannot support FIXEDMAP */ /* cannot support FIXEDMAP */
if (flags & MDB_FIXEDMAP)
return EINVAL; return EINVAL;
} }
#endif
flags |= env->me_flags;
rc = mdb_fname_init(path, flags, &fname); rc = mdb_fname_init(path, flags, &fname);
if (rc) if (rc)
return rc; return rc;
#if MDB_RPAGE_CACHE #if MDB_RPAGE_CACHE
if (MDB_REMAPPING(flags)) {
#ifdef _WIN32 #ifdef _WIN32
env->me_rpmutex = CreateMutex(NULL, FALSE, NULL); env->me_rpmutex = CreateMutex(NULL, FALSE, NULL);
if (!env->me_rpmutex) { if (!env->me_rpmutex) {
@ -5375,6 +5388,7 @@ mdb_env_open(MDB_env *env, const char *path, unsigned int flags, mdb_mode_t mode
if (rc) if (rc)
goto leave; goto leave;
#endif #endif
}
#endif #endif
flags |= MDB_ENV_ACTIVE; /* tell mdb_env_close0() to clean up */ flags |= MDB_ENV_ACTIVE; /* tell mdb_env_close0() to clean up */
@ -5392,6 +5406,7 @@ mdb_env_open(MDB_env *env, const char *path, unsigned int flags, mdb_mode_t mode
goto leave; goto leave;
#if MDB_RPAGE_CACHE #if MDB_RPAGE_CACHE
if (MDB_REMAPPING(flags))
{ {
env->me_rpages = malloc(MDB_ERPAGE_SIZE * sizeof(MDB_ID3)); env->me_rpages = malloc(MDB_ERPAGE_SIZE * sizeof(MDB_ID3));
if (!env->me_rpages) { if (!env->me_rpages) {
@ -5460,6 +5475,7 @@ mdb_env_open(MDB_env *env, const char *path, unsigned int flags, mdb_mode_t mode
txn->mt_dbflags = (unsigned char *)(txn->mt_dbiseqs + env->me_maxdbs); txn->mt_dbflags = (unsigned char *)(txn->mt_dbiseqs + env->me_maxdbs);
txn->mt_env = env; txn->mt_env = env;
#if MDB_RPAGE_CACHE #if MDB_RPAGE_CACHE
if (MDB_REMAPPING(env->me_flags)) {
txn->mt_rpages = malloc(MDB_TRPAGE_SIZE * sizeof(MDB_ID3)); txn->mt_rpages = malloc(MDB_TRPAGE_SIZE * sizeof(MDB_ID3));
if (!txn->mt_rpages) { if (!txn->mt_rpages) {
free(txn); free(txn);
@ -5468,6 +5484,7 @@ mdb_env_open(MDB_env *env, const char *path, unsigned int flags, mdb_mode_t mode
} }
txn->mt_rpages[0].mid = 0; txn->mt_rpages[0].mid = 0;
txn->mt_rpcheck = MDB_TRPAGE_SIZE/2; txn->mt_rpcheck = MDB_TRPAGE_SIZE/2;
}
#endif #endif
txn->mt_dbxs = env->me_dbxs; txn->mt_dbxs = env->me_dbxs;
txn->mt_flags = MDB_TXN_FINISHED; txn->mt_flags = MDB_TXN_FINISHED;
@ -5508,6 +5525,7 @@ mdb_env_close0(MDB_env *env, int excl)
free(env->me_path); free(env->me_path);
free(env->me_dirty_list); free(env->me_dirty_list);
#if MDB_RPAGE_CACHE #if MDB_RPAGE_CACHE
if (MDB_REMAPPING(env->me_flags)) {
if (env->me_txn0 && env->me_txn0->mt_rpages) if (env->me_txn0 && env->me_txn0->mt_rpages)
free(env->me_txn0->mt_rpages); free(env->me_txn0->mt_rpages);
if (env->me_rpages) { if (env->me_rpages) {
@ -5517,6 +5535,7 @@ mdb_env_close0(MDB_env *env, int excl)
munmap(el[x].mptr, el[x].mcnt * env->me_psize); munmap(el[x].mptr, el[x].mcnt * env->me_psize);
free(el); free(el);
} }
}
#endif #endif
free(env->me_txn0); free(env->me_txn0);
mdb_midl_free(env->me_free_pgs); mdb_midl_free(env->me_free_pgs);
@ -5535,11 +5554,10 @@ mdb_env_close0(MDB_env *env, int excl)
} }
if (env->me_map) { if (env->me_map) {
#if MDB_RPAGE_CACHE if (MDB_REMAPPING(env->me_flags))
munmap(env->me_map, NUM_METAS*env->me_psize); munmap(env->me_map, NUM_METAS*env->me_psize);
#else else
munmap(env->me_map, env->me_mapsize); munmap(env->me_map, env->me_mapsize);
#endif
} }
if (env->me_mfd != INVALID_HANDLE_VALUE) if (env->me_mfd != INVALID_HANDLE_VALUE)
(void) close(env->me_mfd); (void) close(env->me_mfd);
@ -5605,12 +5623,15 @@ mdb_env_close0(MDB_env *env, int excl)
(void) close(env->me_lfd); (void) close(env->me_lfd);
} }
#if MDB_RPAGE_CACHE #if MDB_RPAGE_CACHE
if (MDB_REMAPPING(env->me_flags))
{
#ifdef _WIN32 #ifdef _WIN32
if (env->me_fmh) CloseHandle(env->me_fmh); if (env->me_fmh) CloseHandle(env->me_fmh);
if (env->me_rpmutex) CloseHandle(env->me_rpmutex); if (env->me_rpmutex) CloseHandle(env->me_rpmutex);
#else #else
pthread_mutex_destroy(&env->me_rpmutex); pthread_mutex_destroy(&env->me_rpmutex);
#endif #endif
}
#endif #endif
env->me_flags &= ~(MDB_ENV_ACTIVE|MDB_ENV_TXKEY); env->me_flags &= ~(MDB_ENV_ACTIVE|MDB_ENV_TXKEY);
@ -6221,17 +6242,18 @@ mdb_page_get(MDB_cursor *mc, pgno_t pgno, MDB_page **ret, int *lvl)
level = 0; level = 0;
mapped: mapped:
{
#if MDB_RPAGE_CACHE #if MDB_RPAGE_CACHE
if (MDB_REMAPPING(txn->mt_env->me_flags)) {
int rc = mdb_rpage_get(txn, pgno, &p); int rc = mdb_rpage_get(txn, pgno, &p);
if (rc) { if (rc) {
txn->mt_flags |= MDB_TXN_ERROR; txn->mt_flags |= MDB_TXN_ERROR;
return rc; return rc;
} }
#else } else
#endif
{
MDB_env *env = txn->mt_env; MDB_env *env = txn->mt_env;
p = (MDB_page *)(env->me_map + env->me_psize * pgno); p = (MDB_page *)(env->me_map + env->me_psize * pgno);
#endif
} }
done: done:
@ -6422,10 +6444,11 @@ mdb_page_search(MDB_cursor *mc, MDB_val *key, int flags)
} }
#if MDB_RPAGE_CACHE #if MDB_RPAGE_CACHE
if (MDB_REMAPPING(mc->mc_txn->mt_env->me_flags))
{ {
int i; int i;
for (i=1; i<mc->mc_snum; i++) for (i=1; i<mc->mc_snum; i++)
MDB_PAGE_UNREF(mc->mc_txn, mc->mc_pg[i]); mdb_page_unref(mc->mc_txn, mc->mc_pg[i]);
} }
#endif #endif
mc->mc_snum = 1; mc->mc_snum = 1;
@ -6577,7 +6600,7 @@ mdb_get(MDB_txn *txn, MDB_dbi dbi,
mdb_cursor_init(&mc, txn, dbi, &mx); mdb_cursor_init(&mc, txn, dbi, &mx);
rc = mdb_cursor_set(&mc, key, data, MDB_SET, &exact); rc = mdb_cursor_set(&mc, key, data, MDB_SET, &exact);
/* unref all the pages when MDB_RPAGE_CACHE - caller must copy the data /* unref all the pages when MDB_REMAP_CHUNKS - caller must copy the data
* before doing anything else * before doing anything else
*/ */
MDB_CURSOR_UNREF(&mc, 1); MDB_CURSOR_UNREF(&mc, 1);
@ -10723,11 +10746,12 @@ mdb_drop0(MDB_cursor *mc, int subs)
mdb_cursor_pop(mc); mdb_cursor_pop(mc);
mdb_cursor_copy(mc, &mx); mdb_cursor_copy(mc, &mx);
#if MDB_RPAGE_CACHE if (MDB_REMAPPING(mc->mc_txn->mt_env->me_flags)) {
/* bump refcount for mx's pages */ /* bump refcount for mx's pages */
for (i=0; i<mc->mc_snum; i++) for (i=0; i<mc->mc_snum; i++)
mdb_page_get(&mx, mc->mc_pg[i]->mp_pgno, &mx.mc_pg[i], NULL); mdb_page_get(&mx, mc->mc_pg[i]->mp_pgno, &mx.mc_pg[i], NULL);
#endif }
while (mc->mc_snum > 0) { while (mc->mc_snum > 0) {
MDB_page *mp = mc->mc_pg[mc->mc_top]; MDB_page *mp = mc->mc_pg[mc->mc_top];
unsigned n = NUMKEYS(mp); unsigned n = NUMKEYS(mp);

Write Preview
Loading…
Cancel
Save