static inline int
si_qgetindex(siquery *q, sinode *node)
{
svindex *second;
svindex *first = si_nodeindex_priority(node, &second);
ssiter i;
ss_iterinit(sv_indexiter, &i);
int rc;
if (first->count > 0) {
rc = ss_iteropen(sv_indexiter, &i, q->r, first,
SS_GTE, q->key, q->keysize);
if (rc) {
goto result;
}
}
if (sslikely(second == NULL || !second->count))
return 0;
rc = ss_iteropen(sv_indexiter, &i, q->r, second,
SS_GTE, q->key, q->keysize);
if (! rc) {
return 0;
}
result:;
sv *v = ss_iterof(sv_indexiter, &i);
assert(v != NULL);
svv *visible = v->v;
if (sslikely(! q->has)) {
visible = sv_visible((svv*)v->v, q->vlsn);
if (visible == NULL)
return 0;
}
sv vret;
sv_init(&vret, &sv_vif, visible, NULL);
return si_qgetresult(q, &vret, 0);
}
sxstate sx_set_autocommit(sxmanager *m, sxindex *index, sx *x, svv *v)
{
if (sslikely(m->count_rw == 0)) {
sx_init(m, x);
svlogv lv;
lv.id = index->dsn;
lv.next = UINT32_MAX;
sv_init(&lv.v, &sv_vif, v, NULL);
sv_logadd(&x->log, m->r->a, &lv, index->ptr);
sr_seq(m->r->seq, SR_TSNNEXT);
return SXCOMMIT;
}
sx_begin(m, x, SXRW, 0);
int rc = sx_set(x, index, v);
if (ssunlikely(rc == -1)) {
sx_rollback(x);
return SXROLLBACK;
}
sxstate s = sx_prepare(x, NULL, NULL);
if (sslikely(s == SXPREPARE))
sx_commit(x);
else
if (s == SXLOCK)
sx_rollback(x);
return s;
}
static inline int
se_document_opt(const char *path)
{
if (sslikely((intptr_t)path <= (intptr_t)SE_DOCUMENT_FIELD_9))
return (int)(intptr_t)path;
switch (path[0]) {
case 'o':
if (sslikely(strcmp(path, "order") == 0))
return SE_DOCUMENT_ORDER;
break;
case 'l':
if (sslikely(strcmp(path, "log") == 0))
return SE_DOCUMENT_LOG;
break;
case 'p':
if (sslikely(strcmp(path, "prefix") == 0))
return SE_DOCUMENT_PREFIX;
break;
case 'r':
if (sslikely(strcmp(path, "raw") == 0))
return SE_DOCUMENT_RAW;
break;
}
return SE_DOCUMENT_FIELD;
}
int se_reqwrite(sereq *r)
{
sereqarg *arg = &r->arg;
svlog *log = r->arg.log;
se *e = se_of(r->object);
/* set lsn */
sl_prepare(&e->lp, log, arg->lsn);
/* log write */
if (! arg->recover) {
sltx tl;
sl_begin(&e->lp, &tl);
int rc = sl_write(&tl, log);
if (ssunlikely(rc == -1)) {
sl_rollback(&tl);
r->rc = -1;
return -1;
}
sl_commit(&tl);
}
/* commit */
if (sslikely(arg->vlsn_generate))
arg->vlsn = sx_vlsn(&e->xm);
uint64_t now = ss_utime();
svlogindex *i = (svlogindex*)log->index.s;
svlogindex *end = (svlogindex*)log->index.p;
while (i < end) {
sedb *db = i->ptr;
sitx ti;
si_begin(&ti, &db->index, arg->vlsn, now, log, i);
si_write(&ti, arg->recover);
si_commit(&ti);
i++;
}
return 0;
}
sxstate sx_begin(sxmanager *m, sx *x, sxtype type, uint64_t vlsn)
{
sx_promote(x, SXREADY);
x->type = type;
x->log_read = -1;
sr_seqlock(m->r->seq);
x->csn = m->csn;
x->id = sr_seqdo(m->r->seq, SR_TSNNEXT);
if (sslikely(vlsn == 0))
x->vlsn = sr_seqdo(m->r->seq, SR_LSN);
else
x->vlsn = vlsn;
sr_sequnlock(m->r->seq);
sx_init(m, x);
ss_spinlock(&m->lock);
ssrbnode *n = NULL;
int rc = sx_matchtx(&m->i, NULL, (char*)&x->id, sizeof(x->id), &n);
if (rc == 0 && n) {
assert(0);
} else {
ss_rbset(&m->i, n, rc, &x->node);
}
if (type == SXRO)
m->count_rd++;
else
m->count_rw++;
ss_spinunlock(&m->lock);
return SXREADY;
}
int si_recover(si *i)
{
sr *r = i->r;
int exist = ss_vfsexists(r->vfs, i->scheme->path);
if (exist == 0)
goto deploy;
if (i->scheme->path_fail_on_exists) {
sr_error(r->e, "directory '%s' already exists", i->scheme->path);
return -1;
}
int rc = si_recoverdrop(i, r);
switch (rc) {
case -1: return -1;
case 1: goto deploy;
}
rc = si_schemerecover(i->scheme, r);
if (ssunlikely(rc == -1))
return -1;
r->scheme = &i->scheme->scheme;
r->fmt = i->scheme->fmt;
r->fmt_storage = i->scheme->fmt_storage;
sdsnapshot snapshot;
sd_snapshot_init(&snapshot);
rc = si_recoversnapshot(i, r, &snapshot);
if (ssunlikely(rc == -1)) {
sd_snapshot_free(&snapshot, r);
return -1;
}
rc = si_recoverindex(i, r, &snapshot);
sd_snapshot_free(&snapshot, r);
if (sslikely(rc <= 0))
return rc;
deploy:
return si_deploy(i, r, !exist);
}
static inline int
si_plannerpeek_gc(siplanner *p, siplan *plan)
{
/* try to peek a node with a biggest number
* of branches which is ready for gc */
int rc_inprogress = 0;
sinode *n;
ssrqnode *pn = NULL;
while ((pn = ss_rqprev(&p->compact, pn))) {
n = sscast(pn, sinode, nodecompact);
sdindexheader *h = n->self.index.h;
if (sslikely(h->dupkeys == 0) || (h->dupmin >= plan->a))
continue;
uint32_t used = (h->dupkeys * 100) / h->keys;
if (used >= plan->b) {
if (n->flags & SI_LOCK) {
rc_inprogress = 2;
continue;
}
goto match;
}
}
if (rc_inprogress)
plan->explain = SI_ERETRY;
return rc_inprogress;
match:
si_nodelock(n);
plan->explain = SI_ENONE;
plan->node = n;
return 1;
}
void sx_gc(sx *t, sr *r)
{
sxmanager *m = t->manager;
ssiter i;
ss_iterinit(ss_bufiter, &i);
ss_iteropen(ss_bufiter, &i, &t->log.buf, sizeof(svlogv));
if (sslikely(t->s == SXCOMMIT)) {
for (; ss_iterhas(ss_bufiter, &i); ss_iternext(ss_bufiter, &i))
{
svlogv *lv = ss_iterof(ss_bufiter, &i);
sxv *v = lv->vgc;
ss_free(m->asxv, v);
}
} else
if (t->s == SXROLLBACK) {
int gc = 0;
for (; ss_iterhas(ss_bufiter, &i); ss_iternext(ss_bufiter, &i))
{
svlogv *lv = ss_iterof(ss_bufiter, &i);
sxv *v = lv->v.v;
gc += sv_vsize((svv*)v->v);
sx_vfree(m->a, m->asxv, v);
}
ss_quota(r->quota, SS_QREMOVE, gc);
}
sv_logfree(&t->log, m->a);
t->s = SXUNDEF;
}
sxstate sx_set_autocommit(sxmanager *m, sxindex *index, sx *x, svlog *log, svv *v)
{
if (sslikely(m->count_rw == 0)) {
sx_init(m, x, log);
svlogv lv;
lv.index_id = index->dsn;
lv.next = UINT32_MAX;
lv.v = v;
lv.ptr = NULL;
sv_logadd(x->log, index->r, &lv);
sr_seq(index->r->seq, SR_TSNNEXT);
sx_promote(x, SX_COMMIT);
return SX_COMMIT;
}
sx_begin(m, x, SX_RW, log, 0);
int rc = sx_set(x, index, v);
if (ssunlikely(rc == -1)) {
sx_rollback(x);
return SX_ROLLBACK;
}
sxstate s = sx_prepare(x, NULL, NULL);
switch (s) {
case SX_PREPARE:
s = sx_commit(x);
break;
case SX_LOCK:
s = sx_rollback(x);
break;
case SX_ROLLBACK:
break;
default:
assert(0);
}
return s;
}
int se_reqread(sereq *r)
{
sereqarg *arg = &r->arg;
sedb *db = (sedb*)r->db;
uint32_t keysize;
void *key;
if (sslikely(arg->v.v)) {
keysize = sv_size(&arg->v);
key = sv_pointer(&arg->v);
} else {
keysize = 0;
key = NULL;
}
char *prefix;
uint32_t prefixsize;
if (arg->vprefix.v) {
void *vptr = sv_vpointer(arg->vprefix.v);
prefix = sf_key(vptr, 0);
prefixsize = sf_keysize(vptr, 0);
} else {
prefix = NULL;
prefixsize = 0;
}
if (sslikely(arg->vlsn_generate))
arg->vlsn = sr_seq(db->r.seq, SR_LSN);
sitx x;
si_begin(&x, &db->index, 1);
siread q;
si_readopen(&q, &x, arg->cache,
arg->order,
arg->vlsn,
prefix,
prefixsize, key, keysize);
if (arg->update)
si_readupdate(&q, &arg->vup, arg->update_eq);
if (arg->cache_only)
si_readcache_only(&q);
if (arg->has)
si_readhas(&q);
r->rc = si_read(&q);
r->read_disk = q.read_disk;
r->read_cache = q.read_cache;
r->v = q.result.v;
si_readclose(&q);
si_commit(&x);
return r->rc;
}
static int
so_snapshotfree(sosnapshot *s)
{
so *e = so_of(&s->o);
sx_rollback(&s->t);
if (sslikely(s->name)) {
ss_free(&e->a, s->name);
s->name = NULL;
}
ss_free(&e->a_snapshot, s);
return 0;
}
static inline int
sx_preparecb(sx *x, svlogv *v, uint64_t lsn, sxpreparef prepare, void *arg)
{
if (sslikely(lsn == x->vlsn))
return 0;
if (prepare == NULL)
return 0;
sxindex *i = ((sxv*)v->ptr)->index;
if (prepare(x, v->v, i->object, arg))
return 1;
return 0;
}
void ss_rbset(ssrb *t, ssrbnode *p, int prel, ssrbnode *n)
{
n->color = SS_RBRED;
n->p = p;
n->l = NULL;
n->r = NULL;
if (sslikely(p)) {
assert(prel != 0);
if (prel > 0)
p->l = n;
else
p->r = n;
} else {
t->root = n;
}
ss_rbset_fixup(t, n);
}
static inline int
si_recoverdrop(si *i, sr *r)
{
char path[1024];
snprintf(path, sizeof(path), "%s/drop", i->scheme->path);
int rc = ss_vfsexists(r->vfs, path);
if (sslikely(! rc))
return 0;
if (i->scheme->path_fail_on_drop) {
sr_malfunction(r->e, "attempt to recover a dropped database: %s:",
i->scheme->path);
return -1;
}
rc = si_droprepository(i->scheme, r, 0);
if (ssunlikely(rc == -1))
return -1;
return 1;
}
static inline int
sl_iterprepare(sliter *i)
{
srversion *ver = (srversion*)i->map.p;
if (! sr_versioncheck(ver))
return sr_malfunction(i->r->e, "bad log file '%s' version",
i->log->file);
if (ssunlikely(i->log->size < (sizeof(srversion))))
return sr_malfunction(i->r->e, "corrupted log file '%s': bad size",
i->log->file);
slv *next = (slv*)((char*)i->map.p + sizeof(srversion));
int rc = sl_iternext_of(i, next, 1);
if (ssunlikely(rc == -1))
return -1;
if (sslikely(i->next))
return sl_itercontinue_of(i);
return 0;
}
sxstate sx_prepare(sx *x, sxpreparef prepare, void *arg)
{
uint64_t lsn = sr_seq(x->manager->seq, SR_LSN);
/* proceed read-only transactions */
if (x->type == SX_RO || sv_logcount_write(x->log) == 0)
return sx_promote(x, SX_PREPARE);
ssiter i;
ss_iterinit(ss_bufiter, &i);
ss_iteropen(ss_bufiter, &i, &x->log->buf, sizeof(svlogv));
sxstate rc;
for (; ss_iterhas(ss_bufiter, &i); ss_iternext(ss_bufiter, &i))
{
svlogv *lv = ss_iterof(ss_bufiter, &i);
sxv *v = lv->ptr;
if ((int)v->lo == x->log_read)
break;
if (sx_vaborted(v))
return sx_promote(x, SX_ROLLBACK);
if (sslikely(v->prev == NULL)) {
rc = sx_preparecb(x, lv, lsn, prepare, arg);
if (ssunlikely(rc != 0))
return sx_promote(x, SX_ROLLBACK);
continue;
}
if (sx_vcommitted(v->prev)) {
if (v->prev->csn > x->csn)
return sx_promote(x, SX_ROLLBACK);
continue;
}
/* force commit for read-only conflicts */
sxindex *i = v->prev->index;
if (sv_vflags(v->prev->v, i->r) & SVGET) {
rc = sx_preparecb(x, lv, lsn, prepare, arg);
if (ssunlikely(rc != 0))
return sx_promote(x, SX_ROLLBACK);
continue;
}
return sx_promote(x, SX_LOCK);
}
return sx_promote(x, SX_PREPARE);
}
static inline void
ss_rbrotate_right(ssrb *t, ssrbnode *n)
{
ssrbnode *p = n;
ssrbnode *q = n->l;
ssrbnode *parent = n->p;
if (sslikely(p->p != NULL)) {
if (parent->l == p)
parent->l = q;
else
parent->r = q;
} else {
t->root = q;
}
q->p = parent;
p->p = q;
p->l = q->r;
if (p->l)
p->l->p = p;
q->r = p;
}
static inline void
sx_rollback_svp(sx *x, ssiter *i, int free)
{
sxmanager *m = x->manager;
int gc = 0;
for (; ss_iterhas(ss_bufiter, i); ss_iternext(ss_bufiter, i))
{
svlogv *lv = ss_iterof(ss_bufiter, i);
sxv *v = lv->v.v;
/* remove from index and replace head with
* a first waiter */
sx_untrack(v);
/* translate log version from sxv to svv */
sv_init(&lv->v, &sv_vif, v->v, NULL);
if (free) {
if (sslikely(! (v->v->flags & SVGET)))
gc += sv_vsize((svv*)v->v);
sx_vfree(m->r, m->asxv, v);
}
}
ss_quota(m->r->quota, SS_QREMOVE, gc);
}
int sd_buildend(sdbuild *b, sr *r)
{
/* update sizes */
sdbuildref *ref = sd_buildref(b);
ref->msize = ss_bufused(&b->m) - ref->m;
ref->vsize = ss_bufused(&b->v) - ref->v;
ref->ksize = ss_bufused(&b->k) - ref->k;
ref->csize = 0;
/* calculate data crc (non-compressed) */
sdpageheader *h = sd_buildheader(b);
uint32_t crc = 0;
if (sslikely(b->crc)) {
crc = ss_crcp(r->crc, b->m.s + ref->m, ref->msize, 0);
crc = ss_crcp(r->crc, b->v.s + ref->v, ref->vsize, crc);
crc = ss_crcp(r->crc, b->k.s + ref->k, ref->ksize, crc);
}
h->crcdata = crc;
/* compression */
if (b->compress) {
int rc = sd_buildcompress(b, r);
if (ssunlikely(rc == -1))
return -1;
ref->csize = ss_bufused(&b->c) - ref->c;
}
/* update page header */
int total = ref->msize + ref->vsize + ref->ksize;
h->sizekeys = ref->ksize;
h->sizeorigin = total - sizeof(sdpageheader);
h->size = h->sizeorigin;
if (b->compress)
h->size = ref->csize - sizeof(sdpageheader);
else
h->size = h->sizeorigin;
h->crc = ss_crcs(r->crc, h, sizeof(sdpageheader), 0);
if (b->compress)
memcpy(b->c.s + ref->c, h, sizeof(sdpageheader));
return 0;
}
static inline int
se_confsophia_error(srconf *c, srconfstmt *s)
{
se *e = s->ptr;
char *errorp;
char error[128];
error[0] = 0;
int len = sr_errorcopy(&e->error, error, sizeof(error));
if (sslikely(len == 0))
errorp = NULL;
else
errorp = error;
srconf conf = {
.key = c->key,
.flags = c->flags,
.type = c->type,
.function = NULL,
.value = errorp,
.ptr = NULL,
.next = NULL
};
return se_confv(&conf, s);
}
static inline srconf*
se_confsophia(se *e, seconfrt *rt, srconf **pc)
{
srconf *sophia = *pc;
srconf *p = NULL;
sr_C(&p, pc, se_confv, "version", SS_STRING, rt->version, SR_RO, NULL);
sr_C(&p, pc, se_confv, "version_storage", SS_STRING, rt->version_storage, SR_RO, NULL);
sr_C(&p, pc, se_confv, "build", SS_STRING, rt->build, SR_RO, NULL);
sr_C(&p, pc, se_confsophia_error, "error", SS_STRING, NULL, SR_RO, NULL);
sr_c(&p, pc, se_confv_offline, "path", SS_STRINGPTR, &e->conf.path);
sr_c(&p, pc, se_confv_offline, "path_create", SS_U32, &e->conf.path_create);
sr_c(&p, pc, se_confv_offline, "recover", SS_U32, &e->conf.recover);
return sr_C(NULL, pc, NULL, "sophia", SS_UNDEF, sophia, SR_NS, NULL);
}
sxstate sx_prepare(sx *x, sxpreparef prepare, void *arg)
{
uint64_t lsn = sr_seq(x->manager->r->seq, SR_LSN);
/* proceed read-only transactions */
if (x->type == SXRO || sv_logcount_write(&x->log) == 0)
return sx_promote(x, SXPREPARE);
ssiter i;
ss_iterinit(ss_bufiter, &i);
ss_iteropen(ss_bufiter, &i, &x->log.buf, sizeof(svlogv));
for (; ss_iterhas(ss_bufiter, &i); ss_iternext(ss_bufiter, &i))
{
svlogv *lv = ss_iterof(ss_bufiter, &i);
sxv *v = lv->v.v;
if ((int)v->lo == x->log_read)
break;
if (sx_vaborted(v))
return sx_promote(x, SXROLLBACK);
if (sslikely(v->prev == NULL)) {
if (prepare && lsn != x->vlsn) {
sxindex *i = v->index;
if (prepare(x, &lv->v, arg, i->ptr))
return sx_promote(x, SXROLLBACK);
}
continue;
}
if (sx_vcommitted(v->prev)) {
if (v->prev->csn > x->csn)
return sx_promote(x, SXROLLBACK);
continue;
}
/* force commit for read-only conflicts */
if (v->prev->v->flags & SVGET)
continue;
return sx_promote(x, SXLOCK);
}
return sx_promote(x, SXPREPARE);
}
static inline svv*
sv_vset(svv *head, svv *v)
{
/* default */
if (sslikely(head->lsn < v->lsn)) {
v->next = head;
head->flags |= SVDUP;
return v;
}
/* redistribution (starting from highest lsn) */
svv *prev = head;
svv *c = head->next;
while (c) {
assert(c->lsn != v->lsn);
if (c->lsn < v->lsn)
break;
prev = c;
c = c->next;
}
prev->next = v;
v->next = c;
v->flags |= SVDUP;
return head;
}
static inline sshot void*
ss_slabamalloc(ssa *a, int size ssunused)
{
ssslaba *s = (ssslaba*)a->priv;
assert(size == (int)s->slab_size);
ss_spinlock(&s->lock);
char *slab;
if (sslikely(s->pool_free_count)) {
slab = s->pool_free;
s->pool_free = *(char**)slab;
s->pool_free_count--;
if (ssunlikely(s->pool_free_count == 0))
s->pool_free = NULL;
} else
if (ssunlikely(s->pool_next == s->pool_end)) {
slab = NULL;
} else {
slab = s->pool_next;
s->pool_next += s->slab_size;
}
ss_spinunlock(&s->lock);
return slab;
}
static inline int
se_metasophia_error(srmeta *c, srmetastmt *s)
{
se *e = s->ptr;
char *errorp;
char error[128];
error[0] = 0;
int len = sr_errorcopy(&e->error, error, sizeof(error));
if (sslikely(len == 0))
errorp = NULL;
else
errorp = error;
srmeta meta = {
.key = c->key,
.flags = c->flags,
.type = c->type,
.function = NULL,
.value = errorp,
.ptr = NULL,
.next = NULL
};
return se_metav(&meta, s);
}
static inline srmeta*
se_metasophia(se *e, semetart *rt, srmeta **pc)
{
srmeta *sophia = *pc;
srmeta *p = NULL;
sr_M(&p, pc, se_metav, "version", SS_STRING, rt->version, SR_RO, NULL);
sr_M(&p, pc, se_metav, "build", SS_STRING, rt->build, SR_RO, NULL);
sr_M(&p, pc, se_metasophia_error, "error", SS_STRING, NULL, SR_RO, NULL);
sr_m(&p, pc, se_metav_offline, "path", SS_STRINGPTR, &e->meta.path);
sr_m(&p, pc, se_metav_offline, "path_create", SS_U32, &e->meta.path_create);
return sr_M(NULL, pc, NULL, "sophia", SS_UNDEF, sophia, SR_NS, NULL);
}
sxstate sx_begin(sxmanager *m, sx *t, uint64_t vlsn)
{
t->s = SXREADY;
t->complete = 0;
sr_seqlock(m->seq);
t->id = sr_seqdo(m->seq, SR_TSNNEXT);
if (sslikely(vlsn == 0))
t->vlsn = sr_seqdo(m->seq, SR_LSN);
else
t->vlsn = vlsn;
sr_sequnlock(m->seq);
sx_init(m, t);
ss_spinlock(&m->lock);
ssrbnode *n = NULL;
int rc = sx_matchtx(&m->i, NULL, (char*)&t->id, sizeof(t->id), &n);
if (rc == 0 && n) {
assert(0);
} else {
ss_rbset(&m->i, n, rc, &t->node);
}
m->count++;
ss_spinunlock(&m->lock);
return SXREADY;
}
static inline int
si_rangebranch(siread *q, sinode *n, sibranch *b, svmerge *m)
{
sicachebranch *c = si_cachefollow(q->cache);
assert(c->branch == b);
/* iterate cache */
if (ss_iterhas(sd_read, &c->i)) {
svmergesrc *s = sv_mergeadd(m, &c->i);
si_readstat(q, 1, n, 1);
s->ptr = c;
return 1;
}
if (c->open) {
return 1;
}
if (q->cache_only) {
return 2;
}
c->open = 1;
/* choose compression type */
int compression;
ssfilterif *compression_if;
if (! si_branchis_root(b)) {
compression = q->index->scheme->compression_branch;
compression_if = q->index->scheme->compression_branch_if;
} else {
compression = q->index->scheme->compression;
compression_if = q->index->scheme->compression_if;
}
sdreadarg arg = {
.index = &b->index,
.buf = &c->buf_a,
.buf_xf = &c->buf_b,
.buf_read = &q->index->readbuf,
.index_iter = &c->index_iter,
.page_iter = &c->page_iter,
.use_memory = n->in_memory,
.use_mmap = q->index->scheme->mmap,
.use_mmap_copy = 1,
.use_compression = compression,
.compression_if = compression_if,
.has = 0,
.has_vlsn = 0,
.o = q->order,
.memory = &b->copy,
.mmap = &n->map,
.file = &n->file,
.r = q->r
};
ss_iterinit(sd_read, &c->i);
int rc = ss_iteropen(sd_read, &c->i, &arg, q->key, q->keysize);
int reads = sd_read_stat(&c->i);
si_readstat(q, 0, n, reads);
if (ssunlikely(rc == -1))
return -1;
if (ssunlikely(! ss_iterhas(sd_read, &c->i)))
return 0;
svmergesrc *s = sv_mergeadd(m, &c->i);
s->ptr = c;
return 1;
}
static inline int
si_range(siread *q)
{
assert(q->has == 0);
ssiter i;
ss_iterinit(si_iter, &i);
ss_iteropen(si_iter, &i, q->r, q->index, q->order, q->key, q->keysize);
sinode *node;
next_node:
node = ss_iterof(si_iter, &i);
if (ssunlikely(node == NULL))
return 0;
si_txtrack(q->x, node);
/* prepare sources */
svmerge *m = &q->merge;
int count = node->branch_count + 2 + 1;
int rc = sv_mergeprepare(m, q->r, count);
if (ssunlikely(rc == -1)) {
sr_errorreset(q->r->e);
return -1;
}
/* external source (upsert) */
svmergesrc *s;
sv upbuf_reserve;
ssbuf upbuf;
if (ssunlikely(q->upsert_v && q->upsert_v->v)) {
ss_bufinit_reserve(&upbuf, &upbuf_reserve, sizeof(upbuf_reserve));
ss_bufadd(&upbuf, NULL, (void*)&q->upsert_v, sizeof(sv*));
s = sv_mergeadd(m, NULL);
ss_iterinit(ss_bufiterref, &s->src);
ss_iteropen(ss_bufiterref, &s->src, &upbuf, sizeof(sv*));
}
/* in-memory indexes */
svindex *second;
svindex *first = si_nodeindex_priority(node, &second);
if (first->count) {
s = sv_mergeadd(m, NULL);
ss_iterinit(sv_indexiter, &s->src);
ss_iteropen(sv_indexiter, &s->src, q->r, first, q->order,
q->key, q->keysize);
}
if (ssunlikely(second && second->count)) {
s = sv_mergeadd(m, NULL);
ss_iterinit(sv_indexiter, &s->src);
ss_iteropen(sv_indexiter, &s->src, q->r, second, q->order,
q->key, q->keysize);
}
/* cache and branches */
rc = si_cachevalidate(q->cache, node);
if (ssunlikely(rc == -1)) {
sr_oom(q->r->e);
return -1;
}
sibranch *b = node->branch;
while (b) {
rc = si_rangebranch(q, node, b, m);
if (ssunlikely(rc == -1 || rc == 2))
return rc;
b = b->next;
}
/* merge and filter data stream */
ssiter j;
ss_iterinit(sv_mergeiter, &j);
ss_iteropen(sv_mergeiter, &j, q->r, m, q->order);
ssiter k;
ss_iterinit(sv_readiter, &k);
ss_iteropen(sv_readiter, &k, q->r, &j, &q->index->u, q->vlsn, 0);
sv *v = ss_iterof(sv_readiter, &k);
if (ssunlikely(v == NULL)) {
sv_mergereset(&q->merge);
ss_iternext(si_iter, &i);
goto next_node;
}
rc = 1;
/* convert upsert search to SS_EQ */
if (q->upsert_eq) {
rc = sr_compare(q->r->scheme, sv_pointer(v), sv_size(v),
q->key, q->keysize);
rc = rc == 0;
}
/* do prefix search */
if (q->prefix && rc) {
rc = sr_compareprefix(q->r->scheme, q->prefix, q->prefixsize,
sv_pointer(v),
sv_size(v));
}
if (sslikely(rc == 1)) {
if (ssunlikely(si_readdup(q, v) == -1))
return -1;
}
/* skip a possible duplicates from data sources */
ss_iternext(sv_readiter, &k);
return rc;
}
sinode *si_bootstrap(si *i, uint64_t parent)
{
sr *r = i->r;
/* create node */
sinode *n = si_nodenew(r);
if (ssunlikely(n == NULL))
return NULL;
sdid id = {
.parent = parent,
.flags = 0,
.id = sr_seq(r->seq, SR_NSNNEXT)
};
int rc;
rc = si_nodecreate(n, r, i->scheme, &id);
if (ssunlikely(rc == -1))
goto e0;
n->branch = &n->self;
n->branch_count++;
/* in-memory mode support */
ssblob *blob = NULL;
if (i->scheme->storage == SI_SIN_MEMORY) {
blob = &n->self.copy;
rc = ss_blobensure(blob, 4096);
if (ssunlikely(rc == -1))
goto e0;
n->in_memory = 1;
}
/* create index with one empty page */
sdindex index;
sd_indexinit(&index);
rc = sd_indexbegin(&index, r);
if (ssunlikely(rc == -1))
goto e0;
ssqf f, *qf = NULL;
ss_qfinit(&f);
sdbuild build;
sd_buildinit(&build);
rc = sd_buildbegin(&build, r,
i->scheme->node_page_checksum,
i->scheme->compression_key,
i->scheme->compression,
i->scheme->compression_if);
if (ssunlikely(rc == -1))
goto e1;
sd_buildend(&build, r);
rc = sd_indexadd(&index, r, &build, sizeof(sdseal));
if (ssunlikely(rc == -1))
goto e1;
/* write seal */
uint64_t seal = n->file.size;
rc = sd_writeseal(r, &n->file, blob);
if (ssunlikely(rc == -1))
goto e1;
/* write page */
rc = sd_writepage(r, &n->file, blob, &build);
if (ssunlikely(rc == -1))
goto e1;
/* amqf */
if (i->scheme->amqf) {
rc = ss_qfensure(&f, r->a, 0);
if (ssunlikely(rc == -1))
goto e1;
qf = &f;
}
rc = sd_indexcommit(&index, r, &id, qf, n->file.size);
if (ssunlikely(rc == -1))
goto e1;
ss_qffree(&f, r->a);
/* write index */
rc = sd_writeindex(r, &n->file, blob, &index);
if (ssunlikely(rc == -1))
goto e1;
/* close seal */
rc = sd_seal(r, &n->file, blob, &index, seal);
if (ssunlikely(rc == -1))
goto e1;
if (blob) {
rc = ss_blobfit(blob);
if (ssunlikely(rc == -1))
goto e1;
}
if (i->scheme->mmap) {
rc = si_nodemap(n, r);
if (ssunlikely(rc == -1))
goto e1;
}
si_branchset(&n->self, &index);
sd_buildcommit(&build, r);
sd_buildfree(&build, r);
return n;
e1:
ss_qffree(&f, r->a);
sd_indexfree(&index, r);
sd_buildfree(&build, r);
e0:
si_nodefree(n, r, 0);
return NULL;
}
static inline int
si_deploy(si *i, sr *r, int create_directory)
{
/* create directory */
int rc;
if (sslikely(create_directory)) {
rc = ss_vfsmkdir(r->vfs, i->scheme->path, 0755);
if (ssunlikely(rc == -1)) {
sr_malfunction(r->e, "directory '%s' create error: %s",
i->scheme->path, strerror(errno));
return -1;
}
}
/* create scheme file */
rc = si_schemedeploy(i->scheme, r);
if (ssunlikely(rc == -1)) {
sr_malfunction_set(r->e);
return -1;
}
/* create initial node */
sinode *n = si_bootstrap(i, 0);
if (ssunlikely(n == NULL))
return -1;
SS_INJECTION(r->i, SS_INJECTION_SI_RECOVER_0,
si_nodefree(n, r, 0);
sr_malfunction(r->e, "%s", "error injection");
return -1);
rc = si_nodecomplete(n, r, i->scheme);
if (ssunlikely(rc == -1)) {
si_nodefree(n, r, 1);
return -1;
}
si_insert(i, n);
si_plannerupdate(&i->p, SI_COMPACT|SI_BRANCH|SI_TEMP, n);
i->size = si_nodesize(n);
return 1;
}
static inline int
si_trackdir(sitrack *track, sr *r, si *i)
{
DIR *dir = opendir(i->scheme->path);
if (ssunlikely(dir == NULL)) {
sr_malfunction(r->e, "directory '%s' open error: %s",
i->scheme->path, strerror(errno));
return -1;
}
struct dirent *de;
while ((de = readdir(dir))) {
if (ssunlikely(de->d_name[0] == '.'))
continue;
uint64_t id_parent = 0;
uint64_t id = 0;
int rc = si_process(de->d_name, &id, &id_parent);
if (ssunlikely(rc == -1))
continue; /* skip unknown file */
si_tracknsn(track, id_parent);
si_tracknsn(track, id);
sinode *head, *node;
sspath path;
switch (rc) {
case SI_RDB_DBI:
case SI_RDB_DBSEAL: {
/* find parent node and mark it as having
* incomplete compaction process */
head = si_trackget(track, id_parent);
if (sslikely(head == NULL)) {
head = si_nodenew(r);
if (ssunlikely(head == NULL))
goto error;
head->self.id.id = id_parent;
head->recover = SI_RDB_UNDEF;
si_trackset(track, head);
}
head->recover |= rc;
/* remove any incomplete file made during compaction */
if (rc == SI_RDB_DBI) {
ss_pathcompound(&path, i->scheme->path, id_parent, id,
".db.incomplete");
rc = ss_vfsunlink(r->vfs, path.path);
if (ssunlikely(rc == -1)) {
sr_malfunction(r->e, "db file '%s' unlink error: %s",
path.path, strerror(errno));
goto error;
}
continue;
}
assert(rc == SI_RDB_DBSEAL);
/* recover 'sealed' node */
node = si_nodenew(r);
if (ssunlikely(node == NULL))
goto error;
node->recover = SI_RDB_DBSEAL;
ss_pathcompound(&path, i->scheme->path, id_parent, id,
".db.seal");
rc = si_nodeopen(node, r, i->scheme, &path, NULL);
if (ssunlikely(rc == -1)) {
si_nodefree(node, r, 0);
goto error;
}
si_trackset(track, node);
si_trackmetrics(track, node);
continue;
}
}
assert(rc == SI_RDB);
head = si_trackget(track, id);
if (head != NULL && (head->recover & SI_RDB)) {
/* loaded by snapshot */
continue;
}
/* recover node */
node = si_nodenew(r);
if (ssunlikely(node == NULL))
goto error;
node->recover = SI_RDB;
ss_path(&path, i->scheme->path, id, ".db");
rc = si_nodeopen(node, r, i->scheme, &path, NULL);
if (ssunlikely(rc == -1)) {
si_nodefree(node, r, 0);
goto error;
}
si_trackmetrics(track, node);
/* track node */
if (sslikely(head == NULL)) {
si_trackset(track, node);
} else {
/* replace a node previously created by a
* incomplete compaction */
si_trackreplace(track, head, node);
head->recover &= ~SI_RDB_UNDEF;
node->recover |= head->recover;
si_nodefree(head, r, 0);
}
}
closedir(dir);
return 0;
error:
closedir(dir);
return -1;
}
static inline int
se_document_opt(const char *path)
{
switch (path[0]) {
case 'v':
if (sslikely(strcmp(path, "value") == 0))
return SE_DOCUMENT_VALUE;
break;
case 'k':
if (sslikely(strncmp(path, "key", 3) == 0))
return SE_DOCUMENT_KEY;
break;
case 'o':
if (sslikely(strcmp(path, "order") == 0))
return SE_DOCUMENT_ORDER;
break;
case 'l':
if (sslikely(strcmp(path, "lsn") == 0))
return SE_DOCUMENT_LSN;
if (sslikely(strcmp(path, "log") == 0))
return SE_DOCUMENT_LOG;
break;
case 'p':
if (sslikely(strcmp(path, "prefix") == 0))
return SE_DOCUMENT_PREFIX;
break;
case 'a':
if (sslikely(strcmp(path, "arg") == 0))
return SE_DOCUMENT_ARG;
if (strcmp(path, "async") == 0)
return SE_DOCUMENT_ASYNC;
break;
case 'r':
if (sslikely(strcmp(path, "raw") == 0))
return SE_DOCUMENT_RAW;
break;
case 'f':
if (sslikely(strcmp(path, "flags") == 0))
return SE_DOCUMENT_FLAGS;
break;
case 't':
if (sslikely(strcmp(path, "type") == 0))
return SE_DOCUMENT_TYPE;
break;
case 'c':
if (sslikely(strcmp(path, "cache_only") == 0))
return SE_DOCUMENT_CACHE_ONLY;
break;
case 'i':
if (sslikely(strcmp(path, "immutable") == 0))
return SE_DOCUMENT_IMMUTABLE;
break;
case 's':
if (sslikely(strcmp(path, "status") == 0))
return SE_DOCUMENT_STATUS;
if (strcmp(path, "seq") == 0)
return SE_DOCUMENT_SEQ;
break;
}
return SE_DOCUMENT_UNKNOWN;
}
static inline int
si_split(si *index, sdc *c, ssbuf *result,
sinode *parent,
ssiter *i,
uint64_t size_node,
uint64_t size_stream,
uint32_t stream,
uint64_t vlsn)
{
sr *r = &index->r;
uint32_t timestamp = ss_timestamp();
int rc;
sdmergeconf mergeconf = {
.stream = stream,
.size_stream = size_stream,
.size_node = size_node,
.size_page = index->scheme.compaction.node_page_size,
.checksum = index->scheme.compaction.node_page_checksum,
.expire = index->scheme.expire,
.timestamp = timestamp,
.compression = index->scheme.compression,
.compression_if = index->scheme.compression_if,
.direct_io = index->scheme.direct_io,
.direct_io_page_size = index->scheme.direct_io_page_size,
.vlsn = vlsn
};
sinode *n = NULL;
sdmerge merge;
rc = sd_mergeinit(&merge, r, i, &c->build, &c->build_index,
&c->upsert, &mergeconf);
if (ssunlikely(rc == -1))
return -1;
while ((rc = sd_merge(&merge)) > 0)
{
/* create new node */
uint64_t id = sr_seq(index->r.seq, SR_NSNNEXT);
n = si_nodenew(r, id, parent->id);
if (ssunlikely(n == NULL))
goto error;
rc = si_nodecreate(n, r, &index->scheme);
if (ssunlikely(rc == -1))
goto error;
/* write pages */
uint64_t offset;
offset = sd_iosize(&c->io, &n->file);
while ((rc = sd_mergepage(&merge, offset)) == 1) {
rc = sd_writepage(r, &n->file, &c->io, merge.build);
if (ssunlikely(rc == -1))
goto error;
offset = sd_iosize(&c->io, &n->file);
}
if (ssunlikely(rc == -1))
goto error;
offset = sd_iosize(&c->io, &n->file);
rc = sd_mergeend(&merge, offset);
if (ssunlikely(rc == -1))
goto error;
/* write index */
rc = sd_writeindex(r, &n->file, &c->io, &merge.index);
if (ssunlikely(rc == -1))
goto error;
/* mmap mode */
if (index->scheme.mmap) {
rc = si_nodemap(n, r);
if (ssunlikely(rc == -1))
goto error;
}
/* add node to the list */
rc = ss_bufadd(result, index->r.a, &n, sizeof(sinode*));
if (ssunlikely(rc == -1)) {
sr_oom_malfunction(index->r.e);
goto error;
}
n->index = merge.index;
}
if (ssunlikely(rc == -1))
goto error;
return 0;
error:
if (n)
si_nodefree(n, r, 0);
sd_mergefree(&merge);
si_splitfree(result, r);
return -1;
}
static int
si_merge(si *index, sdc *c, sinode *node,
uint64_t vlsn,
ssiter *stream,
uint64_t size_stream,
uint32_t n_stream)
{
sr *r = &index->r;
ssbuf *result = &c->a;
ssiter i;
/* begin compaction.
*
* Split merge stream into a number of
* a new nodes.
*/
int rc;
rc = si_split(index, c, result,
node, stream,
index->scheme.compaction.node_size,
size_stream,
n_stream,
vlsn);
if (ssunlikely(rc == -1))
return -1;
SS_INJECTION(r->i, SS_INJECTION_SI_COMPACTION_0,
si_splitfree(result, r);
sr_malfunction(r->e, "%s", "error injection");
return -1);
/* mask removal of a single node as a
* single node update */
int count = ss_bufused(result) / sizeof(sinode*);
int count_index;
si_lock(index);
count_index = index->n;
si_unlock(index);
sinode *n;
if (ssunlikely(count == 0 && count_index == 1))
{
n = si_bootstrap(index, node->id);
if (ssunlikely(n == NULL))
return -1;
rc = ss_bufadd(result, r->a, &n, sizeof(sinode*));
if (ssunlikely(rc == -1)) {
sr_oom_malfunction(r->e);
si_nodefree(n, r, 1);
return -1;
}
count++;
}
/* commit compaction changes */
si_lock(index);
svindex *j = si_nodeindex(node);
si_plannerremove(&index->p, node);
si_nodesplit(node);
switch (count) {
case 0: /* delete */
si_remove(index, node);
si_redistribute_index(index, r, c, node);
break;
case 1: /* self update */
n = *(sinode**)result->s;
n->i0 = *j;
n->used = j->used;
si_nodelock(n);
si_replace(index, node, n);
si_plannerupdate(&index->p, n);
break;
default: /* split */
rc = si_redistribute(index, r, c, node, result);
if (ssunlikely(rc == -1)) {
si_unlock(index);
si_splitfree(result, r);
return -1;
}
ss_iterinit(ss_bufiterref, &i);
ss_iteropen(ss_bufiterref, &i, result, sizeof(sinode*));
n = ss_iterof(ss_bufiterref, &i);
n->used = n->i0.used;
si_nodelock(n);
si_replace(index, node, n);
si_plannerupdate(&index->p, n);
for (ss_iternext(ss_bufiterref, &i); ss_iterhas(ss_bufiterref, &i);
ss_iternext(ss_bufiterref, &i)) {
n = ss_iterof(ss_bufiterref, &i);
n->used = n->i0.used;
si_nodelock(n);
si_insert(index, n);
si_plannerupdate(&index->p, n);
}
break;
}
sv_indexinit(j);
si_unlock(index);
/* compaction completion */
/* seal nodes */
ss_iterinit(ss_bufiterref, &i);
ss_iteropen(ss_bufiterref, &i, result, sizeof(sinode*));
while (ss_iterhas(ss_bufiterref, &i))
{
n = ss_iterof(ss_bufiterref, &i);
if (index->scheme.sync) {
rc = ss_filesync(&n->file);
if (ssunlikely(rc == -1)) {
sr_malfunction(r->e, "db file '%s' sync error: %s",
ss_pathof(&n->file.path),
strerror(errno));
return -1;
}
}
rc = si_noderename_seal(n, r, &index->scheme);
if (ssunlikely(rc == -1)) {
si_nodefree(node, r, 0);
return -1;
}
SS_INJECTION(r->i, SS_INJECTION_SI_COMPACTION_3,
si_nodefree(node, r, 0);
sr_malfunction(r->e, "%s", "error injection");
return -1);
ss_iternext(ss_bufiterref, &i);
}
SS_INJECTION(r->i, SS_INJECTION_SI_COMPACTION_1,
si_nodefree(node, r, 0);
sr_malfunction(r->e, "%s", "error injection");
return -1);
/* gc node */
uint16_t refs = si_noderefof(node);
if (sslikely(refs == 0)) {
rc = si_nodefree(node, r, 1);
if (ssunlikely(rc == -1))
return -1;
} else {
/* node concurrently being read, schedule for
* delayed removal */
si_nodegc(node, r, &index->scheme);
si_lock(index);
ss_listappend(&index->gc, &node->gc);
index->gc_count++;
si_unlock(index);
}
SS_INJECTION(r->i, SS_INJECTION_SI_COMPACTION_2,
sr_malfunction(r->e, "%s", "error injection");
return -1);
/* complete new nodes */
ss_iterinit(ss_bufiterref, &i);
ss_iteropen(ss_bufiterref, &i, result, sizeof(sinode*));
while (ss_iterhas(ss_bufiterref, &i))
{
n = ss_iterof(ss_bufiterref, &i);
rc = si_noderename_complete(n, r, &index->scheme);
if (ssunlikely(rc == -1))
return -1;
SS_INJECTION(r->i, SS_INJECTION_SI_COMPACTION_4,
sr_malfunction(r->e, "%s", "error injection");
return -1);
ss_iternext(ss_bufiterref, &i);
}
/* unlock */
si_lock(index);
ss_iterinit(ss_bufiterref, &i);
ss_iteropen(ss_bufiterref, &i, result, sizeof(sinode*));
while (ss_iterhas(ss_bufiterref, &i))
{
n = ss_iterof(ss_bufiterref, &i);
si_nodeunlock(n);
ss_iternext(ss_bufiterref, &i);
}
si_unlock(index);
return 0;
}