static void
freelog(svlog *log, sr *c)
{
ssiter i;
ss_iterinit(ss_bufiter, &i);
ss_iteropen(ss_bufiter, &i, &log->buf, sizeof(svlogv));
for (; ss_iteratorhas(&i); ss_iteratornext(&i)) {
svlogv *v = ss_iteratorof(&i);
ss_free(c->a, v->v.v);
}
sv_logfree(log, c->a);
}
static int
si_querycommited_branch(sr *r, sibranch *b, sv *v)
{
ssiter i;
ss_iterinit(sd_indexiter, &i);
ss_iteropen(sd_indexiter, &i, r, &b->index, SS_GTE,
sv_pointer(v), sv_size(v));
sdindexpage *page = ss_iterof(sd_indexiter, &i);
if (page == NULL)
return 0;
return page->lsnmax >= sv_lsn(v);
}
static void
sv_indexiter_iterate0(void)
{
svindex i;
t( sv_indexinit(&i) == 0 );
int keyb = 3;
int keya = 7;
int keyc = 15;
svv *h = st_svv(&st_r.g, NULL, 0, 0, keyb);
t( sv_indexset(&i, &st_r.r, h) == 0 );
svv *p = st_svv(&st_r.g, NULL, 2, 0, keyc);
t( sv_indexset(&i, &st_r.r, p) == 0 );
svv *va = st_svv(&st_r.g, NULL, 1, 0, keya);
t( sv_indexset(&i, &st_r.r, va) == 0 );
svv *vb = st_svv(&st_r.g, NULL, 2, 0, keya);
t( sv_indexset(&i, &st_r.r, vb) == 0 );
svv *vc = st_svv(&st_r.g, NULL, 3, 0, keya);
t( sv_indexset(&i, &st_r.r, vc) == 0 );
ssiter it;
ss_iterinit(sv_indexiter, &it);
ss_iteropen(sv_indexiter, &it, &st_r.r, &i, SS_GTE, NULL, 0);
t( ss_iteratorhas(&it) != 0 );
sv *v = ss_iteratorof(&it);
t( v->v == h );
ss_iteratornext(&it);
v = ss_iteratorof(&it);
t( v->v == vc );
ss_iteratornext(&it);
v = ss_iteratorof(&it);
t( v->v == vb );
ss_iteratornext(&it);
v = ss_iteratorof(&it);
t( v->v == va );
ss_iteratornext(&it);
v = ss_iteratorof(&it);
t( v->v == p );
ss_iteratornext(&it);
v = ss_iteratorof(&it);
t( v == NULL );
sv_indexfree(&i, &st_r.r);
}
static void
sd_v_test(void)
{
sdbuild b;
sd_buildinit(&b);
t( sd_buildbegin(&b, &st_r.r, 1, 0, 0) == 0);
int i = 7;
int j = 8;
addv(&b, &st_r.r, 3, 0, &i);
addv(&b, &st_r.r, 4, 0, &j);
sd_buildend(&b, &st_r.r);
ssbuf buf;
ss_bufinit(&buf);
ssbuf xfbuf;
ss_bufinit(&xfbuf);
t( ss_bufensure(&xfbuf, &st_r.a, 1024) == 0 );
t( sd_commitpage(&b, &st_r.r, &buf) == 0 );
sdpageheader *h = (sdpageheader*)buf.s;
sdpage page;
sd_pageinit(&page, h);
ssiter it;
ss_iterinit(sd_pageiter, &it);
ss_iteropen(sd_pageiter, &it, &st_r.r, &xfbuf, &page, SS_GTE, NULL, 0);
t( ss_iteratorhas(&it) != 0 );
sv *v = ss_iteratorof(&it);
t( v != NULL );
t( *(int*)sv_key(v, &st_r.r, 0) == i );
t( sv_lsn(v) == 3 );
t( sv_flags(v) == 0 );
ss_iteratornext(&it);
t( ss_iteratorhas(&it) != 0 );
v = ss_iteratorof(&it);
t( v != NULL );
t( *(int*)sv_key(v, &st_r.r, 0) == j );
t( sv_lsn(v) == 4 );
t( sv_flags(v) == 0 );
ss_iteratornext(&it);
v = ss_iteratorof(&it);
t( v == NULL );
sd_buildfree(&b, &st_r.r);
ss_buffree(&buf, &st_r.a);
ss_buffree(&xfbuf, &st_r.a);
}
static void
sv_indexiter_lte_empty(void)
{
svindex i;
t( sv_indexinit(&i) == 0 );
ssiter it;
ss_iterinit(sv_indexiter, &it);
ss_iteropen(sv_indexiter, &it, &st_r.r, &i, SS_LTE, NULL, 0);
t( ss_iteratorhas(&it) == 0 );
sv *v = ss_iteratorof(&it);
t( v == NULL );
sv_indexfree(&i, &st_r.r);
}
static int
si_splitfree(ssbuf *result, sr *r)
{
ssiter i;
ss_iterinit(ss_bufiterref, &i);
ss_iteropen(ss_bufiterref, &i, result, sizeof(sinode*));
while (ss_iterhas(ss_bufiterref, &i))
{
sinode *p = ss_iterof(ss_bufiterref, &i);
si_nodefree(p, r, 0);
ss_iternext(ss_bufiterref, &i);
}
return 0;
}
static void
sv_indexiter_lt_eq(void)
{
svindex i;
t( sv_indexinit(&i) == 0 );
int keya = 7;
int keyb = 5;
int keyc = 2;
svv *va = st_svv(&st_r.g, NULL, 0, 0, keya);
t( sv_indexset(&i, &st_r.r, va) == 0 );
svv *vb = st_svv(&st_r.g, NULL, 0, 0, keyb);
t( sv_indexset(&i, &st_r.r, vb) == 0 );
svv *vc = st_svv(&st_r.g, NULL, 0, 0, keyc);
t( sv_indexset(&i, &st_r.r, vc) == 0 );
ssiter it;
ss_iterinit(sv_indexiter, &it);
ss_iteropen(sv_indexiter, &it, &st_r.r, &i, SS_LT, sv_vpointer(va), va->size);
t( ss_iteratorhas(&it) != 0 );
sv *v = ss_iteratorof(&it);
t( v->v == vb );
ss_iterinit(sv_indexiter, &it);
ss_iteropen(sv_indexiter, &it, &st_r.r, &i, SS_LT, sv_vpointer(vb), vb->size);
t( ss_iteratorhas(&it) != 0 );
v = ss_iteratorof(&it);
t( v->v == vc );
ss_iterinit(sv_indexiter, &it);
ss_iteropen(sv_indexiter, &it, &st_r.r, &i, SS_LT, sv_vpointer(vc), vc->size);
t( ss_iteratorhas(&it) == 0 );
v = ss_iteratorof(&it);
t( v == NULL );
sv_indexfree(&i, &st_r.r);
}
static inline void
si_redistribute_set(si *index, sr *r, svv *v)
{
/* match node */
ssiter i;
ss_iterinit(si_iter, &i);
ss_iteropen(si_iter, &i, r, index, SS_GTE, sv_vpointer(v));
sinode *node = ss_iterof(si_iter, &i);
assert(node != NULL);
/* update node */
svindex *vindex = si_nodeindex(node);
sv_indexset(vindex, r, v);
node->used += sv_vsize(v, &index->r);
/* schedule node */
si_plannerupdate(&index->p, node);
}
static void
sv_indexiter_gte_empty(void)
{
svindex i;
t( sv_indexinit(&i) == 0 );
svv *key = st_svv(&st_r.g, &st_r.gc, 0, 0, 7);
ssiter it;
ss_iterinit(sv_indexiter, &it);
ss_iteropen(sv_indexiter, &it, &st_r.r, &i, SS_GTE, sv_vpointer(key), key->size);
t( ss_iteratorhas(&it) == 0 );
sv *v = ss_iteratorof(&it);
t( v == NULL );
sv_indexfree(&i, &st_r.r);
}
sxstate sx_commit(sx *x)
{
if (x->state == SX_COMMIT)
return SX_COMMIT;
assert(x->state == SX_PREPARE);
sxmanager *m = x->manager;
ssiter i;
ss_iterinit(ss_bufiter, &i);
ss_iteropen(ss_bufiter, &i, &x->log->buf, sizeof(svlogv));
uint64_t csn = ++m->csn;
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;
/* abort conflict reader */
if (v->prev && !sx_vcommitted(v->prev)) {
sxindex *i = v->prev->index;
assert(sv_vflags(v->prev->v, i->r) & SVGET);
sx_vabort(v->prev);
}
/* abort waiters */
sx_vabort_all(v->next);
/* mark stmt as commited */
sx_vcommit(v, csn);
lv->ptr = NULL;
/* schedule read stmt for gc */
sxindex *i = v->index;
if (sv_vflags(v->v, i->r) & SVGET) {
sv_vref(v->v);
v->gc = m->gc;
m->gc = v;
m->count_gc++;
} else {
sx_untrack(v);
sx_vpool_push(&m->pool, v);
}
}
/* rollback latest reads */
sx_rollback_svp(x, &i, 0);
sx_promote(x, SX_COMMIT);
sx_end(x);
return SX_COMMIT;
}
sxstate sx_commit(sx *x)
{
assert(x->state == SXPREPARE);
sxmanager *m = x->manager;
ssiter i;
ss_iterinit(ss_bufiter, &i);
ss_iteropen(ss_bufiter, &i, &x->log.buf, sizeof(svlogv));
uint64_t csn = ++m->csn;
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;
/* abort conflict reader */
if (v->prev && !sx_vcommitted(v->prev)) {
assert(v->prev->v->flags & SVGET);
sx_vabort(v->prev);
}
/* abort waiters */
sx_vabort_all(v->next);
/* mark stmt as commited */
sx_vcommit(v, csn);
/* translate log version from sxv to svv */
sv_init(&lv->v, &sv_vif, v->v, NULL);
/* schedule read stmt for gc */
if (v->v->flags & SVGET) {
sv_vref(v->v);
v->gc = m->gc;
m->gc = v;
m->count_gc++;
} else {
sx_untrack(v);
ss_free(m->asxv, v);
}
}
/* rollback latest reads */
sx_rollback_svp(x, &i, 0);
sx_promote(x, SXCOMMIT);
sx_end(x);
return SXCOMMIT;
}
static inline void
si_redistribute_set(si *index, sr *r, uint64_t now, svv *v)
{
index->update_time = now;
/* match node */
ssiter i;
ss_iterinit(si_iter, &i);
ss_iteropen(si_iter, &i, r, index, SS_ROUTE, sv_vpointer(v), v->size);
sinode *node = ss_iterof(si_iter, &i);
assert(node != NULL);
/* update node */
svindex *vindex = si_nodeindex(node);
sv_indexset(vindex, r, v);
node->update_time = index->update_time;
node->used += sv_vsize(v);
/* schedule node */
si_plannerupdate(&index->p, SI_BRANCH, node);
}
static inline int
si_tracksnapshot(sitrack *track, sr *r, si *i, sdsnapshot *s)
{
/* read snapshot */
ssiter iter;
ss_iterinit(sd_snapshotiter, &iter);
int rc;
rc = ss_iteropen(sd_snapshotiter, &iter, r, s);
if (ssunlikely(rc == -1))
return -1;
for (; ss_iterhas(sd_snapshotiter, &iter);
ss_iternext(sd_snapshotiter, &iter))
{
sdsnapshotnode *n = ss_iterof(sd_snapshotiter, &iter);
/* skip updated nodes */
sspath path;
ss_path(&path, i->scheme->path, n->id, ".db");
rc = ss_vfsexists(r->vfs, path.path);
if (! rc)
continue;
uint64_t size = ss_vfssize(r->vfs, path.path);
if (size != n->size_file)
continue;
/* recover node */
sinode *node = si_nodenew(r);
if (ssunlikely(node == NULL))
return -1;
node->recover = SI_RDB;
rc = si_nodeopen(node, r, i->scheme, &path, n);
if (ssunlikely(rc == -1)) {
si_nodefree(node, r, 0);
return -1;
}
si_trackmetrics(track, node);
si_trackset(track, node);
}
/* recover index temperature (read stats) */
sdsnapshotheader *h = sd_snapshot_header(s);
i->read_cache = h->read_cache;
i->read_disk = h->read_disk;
i->lru_v = h->lru_v;
i->lru_steps = h->lru_steps;
return 0;
}
static inline int
si_noderecover(sinode *n, sr *r)
{
/* recover branches */
ssiter i;
ss_iterinit(sd_recover, &i);
ss_iteropen(sd_recover, &i, r, &n->file);
int first = 1;
int rc;
while (ss_iteratorhas(&i))
{
sdindexheader *h = ss_iteratorof(&i);
sibranch *b;
if (first) {
b = &n->self;
} else {
b = si_branchnew(r);
if (ssunlikely(b == NULL))
goto error;
}
sdindex index;
sd_indexinit(&index);
rc = sd_indexcopy(&index, r, h);
if (ssunlikely(rc == -1))
goto error;
si_branchset(b, &index);
b->next = n->branch;
n->branch = b;
n->branch_count++;
first = 0;
ss_iteratornext(&i);
}
rc = sd_recover_complete(&i);
if (ssunlikely(rc == -1))
goto error;
ss_iteratorclose(&i);
return 0;
error:
ss_iteratorclose(&i);
return -1;
}
int si_readcommited(si *index, sr *r, sv *v)
{
ssiter i;
ss_iterinit(si_iter, &i);
ss_iteropen(si_iter, &i, r, index, SS_GTE,
sv_pointer(v), sv_size(v));
sinode *node;
node = ss_iterof(si_iter, &i);
assert(node != NULL);
sibranch *b = node->branch;
int rc;
while (b) {
rc = si_readcommited_branch(r, b, v);
if (rc)
return 1;
b = b->next;
}
rc = si_readcommited_branch(r, &node->self, v);
return rc;
}
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);
}
sxstate sx_rollback(sx *x)
{
sxmanager *m = x->manager;
ssiter i;
ss_iterinit(ss_bufiter, &i);
ss_iteropen(ss_bufiter, &i, &x->log.buf, sizeof(svlogv));
/* support half-commit mode */
if (x->state == SXCOMMIT) {
for (; ss_iterhas(ss_bufiter, &i); ss_iternext(ss_bufiter, &i))
{
svlogv *lv = ss_iterof(ss_bufiter, &i);
svv *v = lv->v.v;
sv_vfree(m->r, v);
}
sx_promote(x, SXROLLBACK);
return SXROLLBACK;
}
sx_rollback_svp(x, &i, 1);
sx_promote(x, SXROLLBACK);
sx_end(x);
return SXROLLBACK;
}
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 int
si_getindex(siread *q, sinode *n)
{
svindex *second;
svindex *first = si_nodeindex_priority(n, &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:;
si_readstat(q, 1, n, 1);
sv *v = ss_iterof(sv_indexiter, &i);
assert(v != NULL);
svv *visible = v->v;
if (sslikely(! q->has)) {
visible = sv_visible(visible, q->vlsn);
if (visible == NULL)
return 0;
}
sv vret;
sv_init(&vret, &sv_vif, visible, NULL);
return si_getresult(q, &vret, 0);
}
static inline int
si_getbranch(siread *q, sinode *n, sibranch *b)
{
sicachebranch *c = si_cachefollow(q->cache);
assert(c->branch == b);
/* 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 = 0,
.use_compression = compression,
.compression_if = compression_if,
.has = q->has,
.has_vlsn = q->vlsn,
.o = SS_GTE,
.mmap = &n->map,
.memory = &b->copy,
.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 <= 0))
return rc;
/* prepare sources */
sv_mergereset(&q->merge);
sv_mergeadd(&q->merge, &c->i);
ssiter i;
ss_iterinit(sv_mergeiter, &i);
ss_iteropen(sv_mergeiter, &i, q->r, &q->merge, SS_GTE);
uint64_t vlsn = q->vlsn;
if (ssunlikely(q->has))
vlsn = UINT64_MAX;
ssiter j;
ss_iterinit(sv_readiter, &j);
ss_iteropen(sv_readiter, &j, q->r, &i, &q->index->u, vlsn, 1);
sv *v = ss_iterof(sv_readiter, &j);
if (ssunlikely(v == NULL))
return 0;
return si_getresult(q, v, 1);
}
static inline int
si_get(siread *q)
{
ssiter i;
ss_iterinit(si_iter, &i);
ss_iteropen(si_iter, &i, q->r, q->index, SS_GTE, q->key, q->keysize);
sinode *node;
node = ss_iterof(si_iter, &i);
assert(node != NULL);
si_txtrack(q->x, node);
/* search in memory */
int rc;
rc = si_getindex(q, node);
if (rc != 0)
return rc;
if (q->cache_only)
return 2;
/* */
rc = si_cachevalidate(q->cache, node);
if (ssunlikely(rc == -1)) {
sr_oom(q->r->e);
return -1;
}
svmerge *m = &q->merge;
rc = sv_mergeprepare(m, q->r, 1);
assert(rc == 0);
/* search on disk */
sibranch *b = node->branch;
while (b) {
rc = si_getbranch(q, node, b);
if (rc != 0)
return rc;
b = b->next;
}
return 0;
}
static inline int
si_split(si *index, sdc *c, ssbuf *result,
sinode *parent,
ssiter *i,
uint64_t size_node,
uint32_t size_stream,
uint64_t vlsn)
{
sr *r = index->r;
int count = 0;
int rc;
sdmergeconf mergeconf = {
.size_stream = size_stream,
.size_node = size_node,
.size_page = index->scheme->node_page_size,
.checksum = index->scheme->node_page_checksum,
.compression = index->scheme->compression,
.compression_key = index->scheme->compression_key,
.offset = 0,
.vlsn = vlsn,
.save_delete = 0,
.save_update = 0
};
sdmerge merge;
sd_mergeinit(&merge, r, i, &c->build, &c->update, &mergeconf);
while ((rc = sd_merge(&merge)) > 0)
{
sinode *n = si_nodenew(r);
if (ssunlikely(n == NULL))
goto error;
sdid id = {
.parent = parent->self.id.id,
.flags = 0,
.id = sr_seq(index->r->seq, SR_NSNNEXT)
};
rc = sd_mergecommit(&merge, &id);
if (ssunlikely(rc == -1))
goto error;
rc = si_nodecreate(n, r, index->scheme, &id, &merge.index, &c->build);
if (ssunlikely(rc == -1))
goto error;
rc = ss_bufadd(result, index->r->a, &n, sizeof(sinode*));
if (ssunlikely(rc == -1)) {
sr_oom_malfunction(index->r->e);
si_nodefree(n, r, 1);
goto error;
}
sd_buildreset(&c->build);
count++;
}
if (ssunlikely(rc == -1))
goto error;
return 0;
error:
si_splitfree(result, r);
sd_mergefree(&merge);
return -1;
}
int si_compaction(si *index, sdc *c, uint64_t vlsn,
sinode *node,
ssiter *stream, uint32_t size_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->node_size,
size_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->self.id.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, SI_COMPACT|SI_BRANCH, node);
switch (count) {
case 0: /* delete */
si_remove(index, node);
si_redistribute_index(index, r, c, node);
uint32_t used = sv_indexused(j);
if (used) {
ss_quota(r->quota, SS_QREMOVE, used);
}
break;
case 1: /* self update */
n = *(sinode**)result->s;
n->i0 = *j;
n->used = sv_indexused(j);
si_nodelock(n);
si_replace(index, node, n);
si_plannerupdate(&index->p, SI_COMPACT|SI_BRANCH, 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 = sv_indexused(&n->i0);
si_nodelock(n);
si_replace(index, node, n);
si_plannerupdate(&index->p, SI_COMPACT|SI_BRANCH, 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 = sv_indexused(&n->i0);
si_nodelock(n);
si_insert(index, n);
si_plannerupdate(&index->p, SI_COMPACT|SI_BRANCH, 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 = si_nodesync(n, r);
if (ssunlikely(rc == -1))
return -1;
}
rc = si_nodeseal(n, r, index->scheme);
if (ssunlikely(rc == -1))
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 old node */
rc = si_nodefree(node, r, 1);
if (ssunlikely(rc == -1))
return -1;
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_nodecomplete(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;
}
static void
sd_read_gt0(void)
{
sdbuild b;
sd_buildinit(&b);
t( sd_buildbegin(&b, &st_r.r, 1, 0, 0, 0, NULL) == 0);
int key = 7;
addv(&b, &st_r.r, 3, 0, &key);
key = 8;
addv(&b, &st_r.r, 4, 0, &key);
key = 9;
addv(&b, &st_r.r, 5, 0, &key);
sd_buildend(&b, &st_r.r);
sdindex index;
sd_indexinit(&index);
t( sd_indexbegin(&index, &st_r.r) == 0 );
int rc;
rc = sd_indexadd(&index, &st_r.r, &b, sizeof(sdseal));
t( rc == 0 );
sdid id;
memset(&id, 0, sizeof(id));
ssfile f;
ss_fileinit(&f, &st_r.vfs);
t( ss_filenew(&f, "./0000.db") == 0 );
t( sd_writeseal(&st_r.r, &f, NULL) == 0 );
t( sd_writepage(&st_r.r, &f, NULL, &b) == 0 );
t( sd_indexcommit(&index, &st_r.r, &id, NULL, f.size) == 0 );
t( sd_writeindex(&st_r.r, &f, NULL, &index) == 0 );
t( sd_seal(&st_r.r, &f, NULL, &index, 0) == 0 );
ssmmap map;
t( ss_vfsmmap(&st_r.vfs, &map, f.fd, f.size, 1) == 0 );
ssbuf buf;
ss_bufinit(&buf);
ssbuf xfbuf;
ss_bufinit(&xfbuf);
t( ss_bufensure(&xfbuf, &st_r.a, 1024) == 0 );
ssiter index_iter;
ssiter page_iter;
sdreadarg arg = {
.index = &index,
.buf = &buf,
.buf_xf = &xfbuf,
.buf_read = NULL,
.index_iter = &index_iter,
.page_iter = &page_iter,
.mmap = &map,
.memory = NULL,
.file = NULL,
.o = SS_GT,
.use_memory = 0,
.use_mmap = 1,
.use_mmap_copy = 0,
.use_compression = 0,
.compression_if = NULL,
.has = 0,
.has_vlsn = 0,
.r = &st_r.r
};
ssiter it;
ss_iterinit(sd_read, &it);
ss_iteropen(sd_read, &it, &arg, NULL, 0);
t( ss_iteratorhas(&it) == 1 );
sv *v = ss_iteratorof(&it);
t( *(int*)sv_field(v, &st_r.r, 0, NULL) == 7);
ss_iteratornext(&it);
v = ss_iteratorof(&it);
t( *(int*)sv_field(v, &st_r.r, 0, NULL) == 8);
ss_iteratornext(&it);
v = ss_iteratorof(&it);
t( *(int*)sv_field(v, &st_r.r, 0, NULL) == 9);
ss_iteratornext(&it);
t( ss_iteratorhas(&it) == 0 );
ss_iteratorclose(&it);
ss_fileclose(&f);
t( ss_vfsmunmap(&st_r.vfs, &map) == 0 );
t( ss_vfsunlink(&st_r.vfs, "./0000.db") == 0 );
sd_indexfree(&index, &st_r.r);
sd_buildfree(&b, &st_r.r);
ss_buffree(&xfbuf, &st_r.a);
ss_buffree(&buf, &st_r.a);
}
static void
sd_read_gt1(void)
{
ssfile f;
ss_fileinit(&f, &st_r.vfs);
t( ss_filenew(&f, "./0000.db") == 0 );
t( sd_writeseal(&st_r.r, &f, NULL) == 0 );
sdbuild b;
sd_buildinit(&b);
t( sd_buildbegin(&b, &st_r.r, 1, 0, 0, 0, NULL) == 0);
int key = 7;
addv(&b, &st_r.r, 3, 0, &key);
key = 8;
addv(&b, &st_r.r, 4, 0, &key);
key = 9;
addv(&b, &st_r.r, 5, 0, &key);
sd_buildend(&b, &st_r.r);
uint64_t poff = f.size;
t( sd_writepage(&st_r.r, &f, NULL, &b) == 0 );
sdindex index;
sd_indexinit(&index);
t( sd_indexbegin(&index, &st_r.r) == 0 );
int rc;
rc = sd_indexadd(&index, &st_r.r, &b, poff);
t( rc == 0 );
t( sd_buildcommit(&b, &st_r.r) == 0 );
t( sd_buildbegin(&b, &st_r.r, 1, 0, 0, 0, NULL) == 0);
key = 10;
addv(&b, &st_r.r, 6, 0, &key);
key = 11;
addv(&b, &st_r.r, 7, 0, &key);
key = 13;
addv(&b, &st_r.r, 8, 0, &key);
sd_buildend(&b, &st_r.r);
poff = f.size;
t( sd_writepage(&st_r.r, &f, NULL, &b) == 0 );
rc = sd_indexadd(&index, &st_r.r, &b, poff);
t( rc == 0 );
t( sd_buildcommit(&b, &st_r.r) == 0 );
t( sd_buildbegin(&b, &st_r.r, 1, 0, 0, 0, NULL) == 0);
key = 15;
addv(&b, &st_r.r, 9, 0, &key);
key = 18;
addv(&b, &st_r.r, 10, 0, &key);
key = 20;
addv(&b, &st_r.r, 11, 0, &key);
sd_buildend(&b, &st_r.r);
poff = f.size;
t( sd_writepage(&st_r.r, &f, NULL, &b) == 0 );
rc = sd_indexadd(&index, &st_r.r, &b, poff);
t( rc == 0 );
t( sd_buildcommit(&b, &st_r.r) == 0 );
sdid id;
memset(&id, 0, sizeof(id));
t( sd_indexcommit(&index, &st_r.r, &id, NULL, f.size) == 0 );
t( sd_writeindex(&st_r.r, &f, NULL, &index) == 0 );
t( sd_seal(&st_r.r, &f, NULL, &index, 0) == 0 );
ssmmap map;
t( ss_vfsmmap(&st_r.vfs, &map, f.fd, f.size, 1) == 0 );
ssbuf buf;
ss_bufinit(&buf);
ssbuf xfbuf;
ss_bufinit(&xfbuf);
t( ss_bufensure(&xfbuf, &st_r.a, 1024) == 0 );
ssiter index_iter;
ssiter page_iter;
sdreadarg arg = {
.index = &index,
.buf = &buf,
.buf_xf = &xfbuf,
.buf_read = NULL,
.index_iter = &index_iter,
.page_iter = &page_iter,
.mmap = &map,
.memory = NULL,
.file = NULL,
.o = SS_GT,
.use_memory = 0,
.use_mmap = 1,
.use_mmap_copy = 0,
.use_compression = 0,
.compression_if = NULL,
.has = 0,
.has_vlsn = 0,
.r = &st_r.r
};
ssiter it;
ss_iterinit(sd_read, &it);
ss_iteropen(sd_read, &it, &arg, NULL, 0);
t( ss_iteratorhas(&it) == 1 );
/* page 0 */
t( ss_iteratorhas(&it) != 0 );
sv *v = ss_iteratorof(&it);
t( *(int*)sv_field(v, &st_r.r, 0, NULL) == 7);
ss_iteratornext(&it);
v = ss_iteratorof(&it);
t( *(int*)sv_field(v, &st_r.r, 0, NULL) == 8);
ss_iteratornext(&it);
v = ss_iteratorof(&it);
t( *(int*)sv_field(v, &st_r.r, 0, NULL) == 9);
ss_iteratornext(&it);
/* page 1 */
v = ss_iteratorof(&it);
t( *(int*)sv_field(v, &st_r.r, 0, NULL) == 10);
ss_iteratornext(&it);
v = ss_iteratorof(&it);
t( *(int*)sv_field(v, &st_r.r, 0, NULL) == 11);
ss_iteratornext(&it);
v = ss_iteratorof(&it);
t( *(int*)sv_field(v, &st_r.r, 0, NULL) == 13);
ss_iteratornext(&it);
/* page 2 */
v = ss_iteratorof(&it);
t( *(int*)sv_field(v, &st_r.r, 0, NULL) == 15);
ss_iteratornext(&it);
v = ss_iteratorof(&it);
t( *(int*)sv_field(v, &st_r.r, 0, NULL) == 18);
ss_iteratornext(&it);
v = ss_iteratorof(&it);
t( *(int*)sv_field(v, &st_r.r, 0, NULL) == 20);
ss_iteratornext(&it);
t( ss_iteratorhas(&it) == 0 );
ss_iteratorclose(&it);
ss_fileclose(&f);
t( ss_vfsmunmap(&st_r.vfs, &map) == 0 );
t( ss_vfsunlink(&st_r.vfs, "./0000.db") == 0 );
sd_indexfree(&index, &st_r.r);
sd_buildfree(&b, &st_r.r);
ss_buffree(&xfbuf, &st_r.a);
ss_buffree(&buf, &st_r.a);
}
static void
sd_read_gt0_compression_zstd(void)
{
ssa a;
ss_aopen(&a, &ss_stda);
ssa aref;
ss_aopen(&aref, &ss_stda);
ssvfs vfs;
ss_vfsinit(&vfs, &ss_stdvfs);
sfscheme cmp;
sf_schemeinit(&cmp);
sffield *field = sf_fieldnew(&a, "key");
t( sf_fieldoptions(field, &a, "u32,key(0)") == 0 );
t( sf_schemeadd(&cmp, &a, field) == 0 );
field = sf_fieldnew(&a, "value");
t( sf_fieldoptions(field, &a, "string") == 0 );
t( sf_schemeadd(&cmp, &a, field) == 0 );
t( sf_schemevalidate(&cmp, &a) == 0 );
ssinjection ij;
memset(&ij, 0, sizeof(ij));
srstat stat;
memset(&stat, 0, sizeof(stat));
srerror error;
sr_errorinit(&error);
srseq seq;
sr_seqinit(&seq);
sscrcf crc = ss_crc32c_function();
sr r;
sr_init(&r, NULL, &error, &a, &aref, &vfs, NULL, NULL, &seq, SF_RAW,
NULL, &cmp, &ij, &stat, crc);
sdbuild b;
sd_buildinit(&b);
t( sd_buildbegin(&b, &r, 1, 0, 0, 1, &ss_zstdfilter) == 0);
int key = 7;
addv(&b, &r, 3, 0, &key);
key = 8;
addv(&b, &r, 4, 0, &key);
key = 9;
addv(&b, &r, 5, 0, &key);
t( sd_buildend(&b, &r) == 0 );
sdindex index;
sd_indexinit(&index);
t( sd_indexbegin(&index, &r) == 0 );
int rc;
rc = sd_indexadd(&index, &r, &b, sizeof(sdseal));
t( rc == 0 );
sdid id;
memset(&id, 0, sizeof(id));
ssfile f;
ss_fileinit(&f, &vfs);
t( ss_filenew(&f, "./0000.db") == 0 );
t( sd_writeseal(&r, &f, NULL) == 0 );
t( sd_writepage(&r, &f, NULL, &b) == 0 );
t( sd_indexcommit(&index, &r, &id, NULL, f.size) == 0 );
t( sd_writeindex(&r, &f, NULL, &index) == 0 );
t( sd_seal(&r, &f, NULL, &index, 0) == 0 );
t( sd_buildcommit(&b, &r) == 0 );
ssmmap map;
t( ss_vfsmmap(&st_r.vfs, &map, f.fd, f.size, 1) == 0 );
ssbuf buf;
ss_bufinit(&buf);
ssbuf xfbuf;
ss_bufinit(&xfbuf);
t( ss_bufensure(&xfbuf, &a, 1024) == 0 );
ssiter index_iter;
ssiter page_iter;
sdreadarg arg = {
.index = &index,
.buf = &buf,
.buf_xf = &xfbuf,
.buf_read = NULL,
.index_iter = &index_iter,
.page_iter = &page_iter,
.mmap = &map,
.memory = NULL,
.file = NULL,
.o = SS_GT,
.use_memory = 0,
.use_mmap = 1,
.use_mmap_copy = 0,
.use_compression = 1,
.compression_if = &ss_zstdfilter,
.has = 0,
.has_vlsn = 0,
.r = &r
};
ssiter it;
ss_iterinit(sd_read, &it);
ss_iteropen(sd_read, &it, &arg, NULL, 0);
t( ss_iteratorhas(&it) == 1 );
sv *v = ss_iteratorof(&it);
t( *(int*)sv_field(v, &r, 0, NULL) == 7);
ss_iteratornext(&it);
v = ss_iteratorof(&it);
t( *(int*)sv_field(v, &r, 0, NULL) == 8);
ss_iteratornext(&it);
v = ss_iteratorof(&it);
t( *(int*)sv_field(v, &r, 0, NULL) == 9);
ss_iteratornext(&it);
t( ss_iteratorhas(&it) == 0 );
ss_iteratorclose(&it);
ss_fileclose(&f);
t( ss_vfsmunmap(&st_r.vfs, &map) == 0 );
t( ss_vfsunlink(&vfs, "./0000.db") == 0 );
sd_indexfree(&index, &r);
sd_buildfree(&b, &r);
ss_buffree(&xfbuf, &a);
ss_buffree(&buf, &a);
sf_schemefree(&cmp, &a);
}
static void
sd_read_gt0_compression_lz4(void)
{
ssa a;
ss_aopen(&a, &ss_stda);
ssa aref;
ss_aopen(&aref, &ss_stda);
ssvfs vfs;
ss_vfsinit(&vfs, &ss_stdvfs);
sfscheme cmp;
sf_schemeinit(&cmp);
sffield *field = sf_fieldnew(&a, "key");
t( sf_fieldoptions(field, &a, "u32,key(0)") == 0 );
t( sf_schemeadd(&cmp, &a, field) == 0 );
field = sf_fieldnew(&a, "value");
t( sf_fieldoptions(field, &a, "string") == 0 );
t( sf_schemeadd(&cmp, &a, field) == 0 );
t( sf_schemevalidate(&cmp, &a) == 0 );
ssinjection ij;
memset(&ij, 0, sizeof(ij));
srstat stat;
memset(&stat, 0, sizeof(stat));
srerror error;
sr_errorinit(&error);
srseq seq;
sr_seqinit(&seq);
sscrcf crc = ss_crc32c_function();
sr r;
sr_init(&r, NULL, &error, &a, &aref, &vfs, NULL, NULL, &seq, SF_RAW,
NULL, &cmp, &ij, &stat, crc);
sdbuild b;
sd_buildinit(&b);
t( sd_buildbegin(&b, &r, 1, 0, 0, 1, &ss_lz4filter) == 0);
int key = 7;
addv(&b, &r, 3, 0, &key);
key = 8;
addv(&b, &r, 4, 0, &key);
key = 9;
addv(&b, &r, 5, 0, &key);
t( sd_buildend(&b, &r) == 0 );
sdindex index;
sd_indexinit(&index);
t( sd_indexbegin(&index, &r) == 0 );
int rc;
rc = sd_indexadd(&index, &r, &b, sizeof(sdseal));
t( rc == 0 );
sdid id;
memset(&id, 0, sizeof(id));
t( sd_indexcommit(&index, &r, &id, NULL, 0) == 0 );
ssfile f;
ss_fileinit(&f, &vfs);
t( ss_filenew(&f, "./0000.db") == 0 );
t( sd_writeseal(&r, &f, NULL) == 0 );
t( sd_writepage(&r, &f, NULL, &b) == 0 );
t( sd_indexcommit(&index, &r, &id, NULL, f.size) == 0 );
t( sd_writeindex(&r, &f, NULL, &index) == 0 );
t( sd_seal(&r, &f, NULL, &index, 0) == 0 );
ssmmap map;
t( ss_vfsmmap(&st_r.vfs, &map, f.fd, f.size, 1) == 0 );
t( sd_buildcommit(&b, &r) == 0 );
ssbuf buf;
ss_bufinit(&buf);
ssbuf xfbuf;
ss_bufinit(&xfbuf);
t( ss_bufensure(&xfbuf, &a, 1024) == 0 );
ssiter index_iter;
ssiter page_iter;
sdreadarg arg = {
.index = &index,
.buf = &buf,
.buf_xf = &xfbuf,
.buf_read = NULL,
.index_iter = &index_iter,
.page_iter = &page_iter,
.mmap = &map,
.memory = NULL,
.file = NULL,
.o = SS_GT,
.use_memory = 0,
.use_mmap = 1,
.use_mmap_copy = 0,
.use_compression = 1,
.compression_if = &ss_lz4filter,
.has = 0,
.has_vlsn = 0,
.r = &r
};
ssiter it;
ss_iterinit(sd_read, &it);
ss_iteropen(sd_read, &it, &arg, NULL, 0);
t( ss_iteratorhas(&it) == 1 );
sv *v = ss_iteratorof(&it);
t( *(int*)sv_field(v, &r, 0, NULL) == 7);
ss_iteratornext(&it);
v = ss_iteratorof(&it);
t( *(int*)sv_field(v, &r, 0, NULL) == 8);
ss_iteratornext(&it);
v = ss_iteratorof(&it);
t( *(int*)sv_field(v, &r, 0, NULL) == 9);
ss_iteratornext(&it);
t( ss_iteratorhas(&it) == 0 );
ss_iteratorclose(&it);
ss_fileclose(&f);
t( ss_vfsmunmap(&st_r.vfs, &map) == 0 );
t( ss_vfsunlink(&vfs, "./0000.db") == 0 );
sd_indexfree(&index, &r);
sd_buildfree(&b, &r);
ss_buffree(&xfbuf, &a);
ss_buffree(&buf, &a);
sf_schemefree(&cmp, &a);
}
static void
sd_read_gt1_compression_zstd(void)
{
ssa a;
ss_aopen(&a, &ss_stda);
ssa aref;
ss_aopen(&aref, &ss_stda);
ssvfs vfs;
ss_vfsinit(&vfs, &ss_stdvfs);
sfscheme cmp;
sf_schemeinit(&cmp);
sffield *field = sf_fieldnew(&a, "key");
t( sf_fieldoptions(field, &a, "u32,key(0)") == 0 );
t( sf_schemeadd(&cmp, &a, field) == 0 );
field = sf_fieldnew(&a, "value");
t( sf_fieldoptions(field, &a, "string") == 0 );
t( sf_schemeadd(&cmp, &a, field) == 0 );
t( sf_schemevalidate(&cmp, &a) == 0 );
ssinjection ij;
memset(&ij, 0, sizeof(ij));
srstat stat;
memset(&stat, 0, sizeof(stat));
srerror error;
sr_errorinit(&error);
srseq seq;
sr_seqinit(&seq);
sscrcf crc = ss_crc32c_function();
sr r;
sr_init(&r, NULL, &error, &a, &aref, &vfs, NULL, NULL, &seq, SF_RAW,
NULL, &cmp, &ij, &stat, crc);
ssfile f;
ss_fileinit(&f, &vfs);
t( ss_filenew(&f, "./0000.db") == 0 );
t( sd_writeseal(&r, &f, NULL) == 0 );
sdbuild b;
sd_buildinit(&b);
t( sd_buildbegin(&b, &r, 1, 0, 0, 1, &ss_zstdfilter) == 0);
int key = 7;
addv(&b, &r, 3, 0, &key);
key = 8;
addv(&b, &r, 4, 0, &key);
key = 9;
addv(&b, &r, 5, 0, &key);
sd_buildend(&b, &r);
uint64_t poff = f.size;
t( sd_writepage(&r, &f, NULL, &b) == 0 );
sdindex index;
sd_indexinit(&index);
t( sd_indexbegin(&index, &r) == 0 );
int rc;
rc = sd_indexadd(&index, &r, &b, poff);
t( rc == 0 );
t( sd_buildcommit(&b, &r) == 0 );
sd_buildreset(&b, &r);
t( sd_buildbegin(&b, &r, 1, 0, 0, 1, &ss_zstdfilter) == 0);
key = 10;
addv(&b, &r, 6, 0, &key);
key = 11;
addv(&b, &r, 7, 0, &key);
key = 13;
addv(&b, &r, 8, 0, &key);
sd_buildend(&b, &r);
poff = f.size;
t( sd_writepage(&r, &f, NULL, &b) == 0 );
rc = sd_indexadd(&index, &r, &b, poff);
t( rc == 0 );
t( sd_buildcommit(&b, &r) == 0 );
sd_buildreset(&b, &r);
t( sd_buildbegin(&b, &r, 1, 0, 0, 1, &ss_zstdfilter) == 0);
key = 15;
addv(&b, &r, 9, 0, &key);
key = 18;
addv(&b, &r, 10, 0, &key);
key = 20;
addv(&b, &r, 11, 0, &key);
sd_buildend(&b, &r);
poff = f.size;
t( sd_writepage(&r, &f, NULL, &b) == 0 );
rc = sd_indexadd(&index, &r, &b, poff);
t( rc == 0 );
t( sd_buildcommit(&b, &r) == 0 );
sdid id;
memset(&id, 0, sizeof(id));
t( sd_indexcommit(&index, &r, &id, NULL, f.size) == 0 );
t( sd_writeindex(&r, &f, NULL, &index) == 0 );
t( sd_seal(&r, &f, NULL, &index, 0) == 0 );
ssmmap map;
t( ss_vfsmmap(&st_r.vfs, &map, f.fd, f.size, 1) == 0 );
ssbuf buf;
ss_bufinit(&buf);
ssbuf xfbuf;
ss_bufinit(&xfbuf);
t( ss_bufensure(&xfbuf, &a, 1024) == 0 );
ssiter index_iter;
ssiter page_iter;
sdreadarg arg = {
.index = &index,
.buf = &buf,
.buf_xf = &xfbuf,
.buf_read = NULL,
.index_iter = &index_iter,
.page_iter = &page_iter,
.mmap = &map,
.memory = NULL,
.file = NULL,
.o = SS_GT,
.use_memory = 0,
.use_mmap = 1,
.use_mmap_copy = 0,
.use_compression = 1,
.compression_if = &ss_zstdfilter,
.has = 0,
.has_vlsn = 0,
.r = &r
};
ssiter it;
ss_iterinit(sd_read, &it);
ss_iteropen(sd_read, &it, &arg, NULL, 0);
t( ss_iteratorhas(&it) == 1 );
/* page 0 */
t( ss_iteratorhas(&it) != 0 );
sv *v = ss_iteratorof(&it);
t( *(int*)sv_field(v, &r, 0, NULL) == 7);
ss_iteratornext(&it);
v = ss_iteratorof(&it);
t( *(int*)sv_field(v, &r, 0, NULL) == 8);
ss_iteratornext(&it);
v = ss_iteratorof(&it);
t( *(int*)sv_field(v, &r, 0, NULL) == 9);
ss_iteratornext(&it);
/* page 1 */
v = ss_iteratorof(&it);
t( *(int*)sv_field(v, &r, 0, NULL) == 10);
ss_iteratornext(&it);
v = ss_iteratorof(&it);
t( *(int*)sv_field(v, &r, 0, NULL) == 11);
ss_iteratornext(&it);
v = ss_iteratorof(&it);
t( *(int*)sv_field(v, &r, 0, NULL) == 13);
ss_iteratornext(&it);
/* page 2 */
v = ss_iteratorof(&it);
t( *(int*)sv_field(v, &r, 0, NULL) == 15);
ss_iteratornext(&it);
v = ss_iteratorof(&it);
t( *(int*)sv_field(v, &r, 0, NULL) == 18);
ss_iteratornext(&it);
v = ss_iteratorof(&it);
t( *(int*)sv_field(v, &r, 0, NULL) == 20);
ss_iteratornext(&it);
t( ss_iteratorhas(&it) == 0 );
ss_iteratorclose(&it);
ss_fileclose(&f);
t( ss_vfsmunmap(&st_r.vfs, &map) == 0 );
t( ss_vfsunlink(&vfs, "./0000.db") == 0 );
sd_indexfree(&index, &r);
sd_buildfree(&b, &r);
ss_buffree(&buf, &a);
ss_buffree(&xfbuf, &a);
sf_schemefree(&cmp, &a);
}
static void
sd_read_gt1_compression_lz4(void)
{
ssa a;
ss_aopen(&a, &ss_stda);
ssa aref;
ss_aopen(&aref, &ss_stda);
ssvfs vfs;
ss_vfsinit(&vfs, &ss_stdvfs);
sfscheme cmp;
sf_schemeinit(&cmp);
sffield *field = sf_fieldnew(&a, "key");
t( sf_fieldoptions(field, &a, "u32,key(0)") == 0 );
t( sf_schemeadd(&cmp, &a, field) == 0 );
field = sf_fieldnew(&a, "value");
t( sf_fieldoptions(field, &a, "string") == 0 );
t( sf_schemeadd(&cmp, &a, field) == 0 );
t( sf_schemevalidate(&cmp, &a) == 0 );
ssinjection ij;
memset(&ij, 0, sizeof(ij));
srstat stat;
memset(&stat, 0, sizeof(stat));
srerror error;
sr_errorinit(&error);
srseq seq;
sr_seqinit(&seq);
sscrcf crc = ss_crc32c_function();
sr r;
sr_init(&r, NULL, &error, &a, &aref, &vfs, NULL, NULL, &seq, SF_RAW,
NULL, &cmp, &ij, &stat, crc);
ssfile f;
ss_fileinit(&f, &vfs);
t( ss_filenew(&f, "./0000.db") == 0 );
t( sd_writeseal(&r, &f, NULL) == 0 );
sdbuild b;
sd_buildinit(&b);
t( sd_buildbegin(&b, &r, 1, 0, 0, 1, &ss_lz4filter) == 0);
int key = 7;
addv(&b, &r, 3, 0, &key);
key = 8;
addv(&b, &r, 4, 0, &key);
key = 9;
addv(&b, &r, 5, 0, &key);
sd_buildend(&b, &r);
uint64_t poff = f.size;
t( sd_writepage(&r, &f, NULL, &b) == 0 );
sdindex index;
sd_indexinit(&index);
t( sd_indexbegin(&index, &r) == 0 );
int rc;
rc = sd_indexadd(&index, &r, &b, poff);
t( rc == 0 );
t( sd_buildcommit(&b, &r) == 0 );
sd_buildreset(&b, &r);
t( sd_buildbegin(&b, &r, 1, 0, 0, 1, &ss_lz4filter) == 0);
key = 10;
addv(&b, &r, 6, 0, &key);
key = 11;
addv(&b, &r, 7, 0, &key);
key = 13;
addv(&b, &r, 8, 0, &key);
sd_buildend(&b, &r);
poff = f.size;
t( sd_writepage(&r, &f, NULL, &b) == 0 );
rc = sd_indexadd(&index, &r, &b, poff);
t( rc == 0 );
t( sd_buildcommit(&b, &r) == 0 );
sd_buildreset(&b, &r);
t( sd_buildbegin(&b, &r, 1, 0, 0, 1, &ss_lz4filter) == 0);
key = 15;
addv(&b, &r, 9, 0, &key);
key = 18;
addv(&b, &r, 10, 0, &key);
key = 20;
addv(&b, &r, 11, 0, &key);
sd_buildend(&b, &r);
poff = f.size;
t( sd_writepage(&r, &f, NULL, &b) == 0 );
rc = sd_indexadd(&index, &r, &b, poff);
t( rc == 0 );
t( sd_buildcommit(&b, &r) == 0 );
sdid id;
memset(&id, 0, sizeof(id));
t( sd_indexcommit(&index, &r, &id, NULL, f.size) == 0 );
t( sd_writeindex(&r, &f, NULL, &index) == 0 );
t( sd_seal(&r, &f, NULL, &index, 0) == 0 );
ssmmap map;
t( ss_vfsmmap(&st_r.vfs, &map, f.fd, f.size, 1) == 0 );
ssbuf buf;
ss_bufinit(&buf);
ssbuf xfbuf;
ss_bufinit(&xfbuf);
t( ss_bufensure(&xfbuf, &a, 1024) == 0 );
ssiter index_iter;
ssiter page_iter;
sdreadarg arg = {
.index = &index,
.buf = &buf,
.buf_xf = &xfbuf,
.buf_read = NULL,
.index_iter = &index_iter,
.page_iter = &page_iter,
.mmap = &map,
.memory = NULL,
.file = NULL,
.o = SS_GT,
.use_memory = 0,
.use_mmap = 1,
.use_mmap_copy = 0,
.use_compression = 1,
.compression_if = &ss_lz4filter,
.has = 0,
.has_vlsn = 0,
.r = &r
};
ssiter it;
ss_iterinit(sd_read, &it);
ss_iteropen(sd_read, &it, &arg, NULL, 0);
t( ss_iteratorhas(&it) == 1 );
/* page 0 */
t( ss_iteratorhas(&it) != 0 );
sv *v = ss_iteratorof(&it);
t( *(int*)sv_field(v, &r, 0, NULL) == 7);
ss_iteratornext(&it);
v = ss_iteratorof(&it);
t( *(int*)sv_field(v, &r, 0, NULL) == 8);
ss_iteratornext(&it);
v = ss_iteratorof(&it);
t( *(int*)sv_field(v, &r, 0, NULL) == 9);
ss_iteratornext(&it);
/* page 1 */
v = ss_iteratorof(&it);
t( *(int*)sv_field(v, &r, 0, NULL) == 10);
ss_iteratornext(&it);
v = ss_iteratorof(&it);
t( *(int*)sv_field(v, &r, 0, NULL) == 11);
ss_iteratornext(&it);
v = ss_iteratorof(&it);
t( *(int*)sv_field(v, &r, 0, NULL) == 13);
ss_iteratornext(&it);
/* page 2 */
v = ss_iteratorof(&it);
t( *(int*)sv_field(v, &r, 0, NULL) == 15);
ss_iteratornext(&it);
v = ss_iteratorof(&it);
t( *(int*)sv_field(v, &r, 0, NULL) == 18);
ss_iteratornext(&it);
v = ss_iteratorof(&it);
t( *(int*)sv_field(v, &r, 0, NULL) == 20);
ss_iteratornext(&it);
t( ss_iteratorhas(&it) == 0 );
ss_iteratorclose(&it);
ss_fileclose(&f);
t( ss_vfsmunmap(&st_r.vfs, &map) == 0 );
t( ss_vfsunlink(&vfs, "./0000.db") == 0 );
sd_indexfree(&index, &r);
sd_buildfree(&b, &r);
ss_buffree(&buf, &a);
ss_buffree(&xfbuf, &a);
sf_schemefree(&cmp, &a);
}
stgroup *sd_read_group(void)
{
stgroup *group = st_group("sdread");
st_groupadd(group, st_test("gt0", sd_read_gt0));
st_groupadd(group, st_test("gt1", sd_read_gt1));
st_groupadd(group, st_test("gt0_compression_zstd", sd_read_gt0_compression_zstd));
st_groupadd(group, st_test("gt0_compression_lz4", sd_read_gt0_compression_lz4));
st_groupadd(group, st_test("gt1_compression_zstd", sd_read_gt1_compression_zstd));
st_groupadd(group, st_test("gt1_compression_lz4", sd_read_gt1_compression_lz4));
return group;
}
static inline sibranch*
si_branchcreate(si *index, sdc *c, sinode *parent, svindex *vindex, uint64_t vlsn)
{
sr *r = index->r;
sibranch *branch = NULL;
/* in-memory mode blob */
int rc;
ssblob copy, *blob = NULL;
if (parent->in_memory) {
ss_blobinit(©, r->vfs);
rc = ss_blobensure(©, 10ULL * 1024 * 1024);
if (ssunlikely(rc == -1)) {
sr_oom_malfunction(r->e);
return NULL;
}
blob = ©
}
svmerge vmerge;
sv_mergeinit(&vmerge);
rc = sv_mergeprepare(&vmerge, r, 1);
if (ssunlikely(rc == -1))
return NULL;
svmergesrc *s = sv_mergeadd(&vmerge, NULL);
ss_iterinit(sv_indexiter, &s->src);
ss_iteropen(sv_indexiter, &s->src, r, vindex, SS_GTE, NULL, 0);
ssiter i;
ss_iterinit(sv_mergeiter, &i);
ss_iteropen(sv_mergeiter, &i, r, &vmerge, SS_GTE);
/* merge iter is not used */
sdmergeconf mergeconf = {
.size_stream = UINT32_MAX,
.size_node = UINT64_MAX,
.size_page = index->scheme->node_page_size,
.checksum = index->scheme->node_page_checksum,
.compression_key = index->scheme->compression_key,
.compression = index->scheme->compression_branch,
.compression_if = index->scheme->compression_branch_if,
.vlsn = vlsn,
.vlsn_lru = 0,
.save_delete = 1,
.save_upsert = 1
};
sdmerge merge;
sd_mergeinit(&merge, r, &i, &c->build, &c->upsert, &mergeconf);
while ((rc = sd_merge(&merge)) > 0)
{
assert(branch == NULL);
/* write open seal */
uint64_t seal = parent->file.size;
rc = sd_writeseal(r, &parent->file, blob);
if (ssunlikely(rc == -1))
goto e0;
/* write pages */
uint64_t offset = parent->file.size;
while ((rc = sd_mergepage(&merge, offset)) == 1)
{
rc = sd_writepage(r, &parent->file, blob, merge.build);
if (ssunlikely(rc == -1))
goto e0;
offset = parent->file.size;
}
if (ssunlikely(rc == -1))
goto e0;
sdid id = {
.parent = parent->self.id.id,
.flags = SD_IDBRANCH,
.id = sr_seq(r->seq, SR_NSNNEXT)
};
rc = sd_mergecommit(&merge, &id, parent->file.size);
if (ssunlikely(rc == -1))
goto e0;
/* write index */
rc = sd_writeindex(r, &parent->file, blob, &merge.index);
if (ssunlikely(rc == -1))
goto e0;
if (index->scheme->sync) {
rc = ss_filesync(&parent->file);
if (ssunlikely(rc == -1)) {
sr_malfunction(r->e, "file '%s' sync error: %s",
ss_pathof(&parent->file.path),
strerror(errno));
goto e0;
}
}
SS_INJECTION(r->i, SS_INJECTION_SI_BRANCH_0,
sd_mergefree(&merge);
sr_malfunction(r->e, "%s", "error injection");
return NULL);
/* seal the branch */
rc = sd_seal(r, &parent->file, blob, &merge.index, seal);
if (ssunlikely(rc == -1))
goto e0;
if (index->scheme->sync == 2) {
rc = ss_filesync(&parent->file);
if (ssunlikely(rc == -1)) {
sr_malfunction(r->e, "file '%s' sync error: %s",
ss_pathof(&parent->file.path),
strerror(errno));
goto e0;
}
}
/* create new branch object */
branch = si_branchnew(r);
if (ssunlikely(branch == NULL))
goto e0;
si_branchset(branch, &merge.index);
}
sv_mergefree(&vmerge, r->a);
if (ssunlikely(rc == -1)) {
sr_oom_malfunction(r->e);
goto e0;
}
assert(branch != NULL);
/* in-memory mode support */
if (blob) {
rc = ss_blobfit(blob);
if (ssunlikely(rc == -1)) {
ss_blobfree(blob);
goto e1;
}
branch->copy = copy;
}
/* mmap support */
if (index->scheme->mmap) {
ss_mmapinit(&parent->map_swap);
rc = ss_vfsmmap(r->vfs, &parent->map_swap, parent->file.fd,
parent->file.size, 1);
if (ssunlikely(rc == -1)) {
sr_malfunction(r->e, "db file '%s' mmap error: %s",
ss_pathof(&parent->file.path),
strerror(errno));
goto e1;
}
}
return branch;
e0:
sd_mergefree(&merge);
if (blob)
ss_blobfree(blob);
return NULL;
e1:
si_branchfree(branch, r);
return NULL;
}
int si_branch(si *index, sdc *c, siplan *plan, uint64_t vlsn)
{
sr *r = index->r;
sinode *n = plan->node;
assert(n->flags & SI_LOCK);
si_lock(index);
if (ssunlikely(n->used == 0)) {
si_nodeunlock(n);
si_unlock(index);
return 0;
}
svindex *i;
i = si_noderotate(n);
si_unlock(index);
sibranch *branch = si_branchcreate(index, c, n, i, vlsn);
if (ssunlikely(branch == NULL))
return -1;
/* commit */
si_lock(index);
branch->next = n->branch;
n->branch->link = branch;
n->branch = branch;
n->branch_count++;
uint32_t used = sv_indexused(i);
n->used -= used;
ss_quota(r->quota, SS_QREMOVE, used);
index->size +=
sd_indexsize(branch->index.h) +
sd_indextotal(&branch->index);
svindex swap = *i;
si_nodeunrotate(n);
si_nodeunlock(n);
si_plannerupdate(&index->p, SI_BRANCH|SI_COMPACT, n);
ssmmap swap_map = n->map;
n->map = n->map_swap;
memset(&n->map_swap, 0, sizeof(n->map_swap));
si_unlock(index);
/* gc */
if (index->scheme->mmap) {
int rc = ss_vfsmunmap(r->vfs, &swap_map);
if (ssunlikely(rc == -1)) {
sr_malfunction(r->e, "db file '%s' munmap error: %s",
ss_pathof(&n->file.path),
strerror(errno));
return -1;
}
}
si_nodegc_index(r, &swap);
return 1;
}
int si_compact(si *index, sdc *c, siplan *plan,
uint64_t vlsn,
uint64_t vlsn_lru,
ssiter *vindex,
uint64_t vindex_used)
{
sr *r = index->r;
sinode *node = plan->node;
assert(node->flags & SI_LOCK);
/* prepare for compaction */
int rc;
rc = sd_censure(c, r, node->branch_count);
if (ssunlikely(rc == -1))
return sr_oom_malfunction(r->e);
svmerge merge;
sv_mergeinit(&merge);
rc = sv_mergeprepare(&merge, r, node->branch_count + 1);
if (ssunlikely(rc == -1))
return -1;
/* read node file into memory */
int use_mmap = index->scheme->mmap;
ssmmap *map = &node->map;
ssmmap preload;
if (index->scheme->node_compact_load) {
rc = si_noderead(node, r, &c->c);
if (ssunlikely(rc == -1))
return -1;
preload.p = c->c.s;
preload.size = ss_bufused(&c->c);
map = &preload;
use_mmap = 1;
}
/* include vindex into merge process */
svmergesrc *s;
uint64_t size_stream = 0;
if (vindex) {
s = sv_mergeadd(&merge, vindex);
size_stream = vindex_used;
}
sdcbuf *cbuf = c->head;
sibranch *b = node->branch;
while (b) {
s = sv_mergeadd(&merge, NULL);
/* choose compression type */
int compression;
ssfilterif *compression_if;
if (! si_branchis_root(b)) {
compression = index->scheme->compression_branch;
compression_if = index->scheme->compression_branch_if;
} else {
compression = index->scheme->compression;
compression_if = index->scheme->compression_if;
}
sdreadarg arg = {
.index = &b->index,
.buf = &cbuf->a,
.buf_xf = &cbuf->b,
.buf_read = &c->d,
.index_iter = &cbuf->index_iter,
.page_iter = &cbuf->page_iter,
.use_memory = node->in_memory,
.use_mmap = use_mmap,
.use_mmap_copy = 0,
.use_compression = compression,
.compression_if = compression_if,
.has = 0,
.has_vlsn = 0,
.o = SS_GTE,
.memory = &b->copy,
.mmap = map,
.file = &node->file,
.r = r
};
ss_iterinit(sd_read, &s->src);
int rc = ss_iteropen(sd_read, &s->src, &arg, NULL, 0);
if (ssunlikely(rc == -1))
return sr_oom_malfunction(r->e);
size_stream += sd_indextotal(&b->index);
cbuf = cbuf->next;
b = b->next;
}
ssiter i;
ss_iterinit(sv_mergeiter, &i);
ss_iteropen(sv_mergeiter, &i, r, &merge, SS_GTE);
rc = si_merge(index, c, node, vlsn, vlsn_lru, &i, size_stream);
sv_mergefree(&merge, r->a);
return rc;
}
int si_compact_index(si *index, sdc *c, siplan *plan,
uint64_t vlsn,
uint64_t vlsn_lru)
{
sinode *node = plan->node;
si_lock(index);
if (ssunlikely(node->used == 0)) {
si_nodeunlock(node);
si_unlock(index);
return 0;
}
svindex *vindex;
vindex = si_noderotate(node);
si_unlock(index);
uint64_t size_stream = sv_indexused(vindex);
ssiter i;
ss_iterinit(sv_indexiter, &i);
ss_iteropen(sv_indexiter, &i, index->r, vindex, SS_GTE, NULL, 0);
return si_compact(index, c, plan, vlsn, vlsn_lru, &i, size_stream);
}
static inline int
si_qgetbranch(siquery *q, sinode *n, sibranch *b)
{
sicachebranch *cb = si_cachefollow(q->cache);
assert(cb->branch == b);
sireadarg arg = {
.scheme = q->index->scheme,
.index = q->index,
.n = n,
.b = b,
.buf = &cb->buf_a,
.buf_xf = &cb->buf_b,
.buf_read = &q->index->readbuf,
.index_iter = &cb->index_iter,
.page_iter = &cb->page_iter,
.vlsn = q->vlsn,
.has = q->has,
.mmap_copy = 0,
.o = SS_GTE,
.r = q->r
};
ss_iterinit(si_read, &cb->i);
int rc = ss_iteropen(si_read, &cb->i, &arg, q->key, q->keysize);
if (ssunlikely(rc <= 0))
return rc;
uint64_t vlsn = q->vlsn;
if (q->has) {
vlsn = UINT64_MAX;
}
/* prepare sources */
sv_mergereset(&q->merge);
sv_mergeadd(&q->merge, &cb->i);
ssiter i;
ss_iterinit(sv_mergeiter, &i);
ss_iteropen(sv_mergeiter, &i, q->r, &q->merge, SS_GTE);
ssiter j;
ss_iterinit(sv_readiter, &j);
ss_iteropen(sv_readiter, &j, q->r, &i, &q->index->u, vlsn, 1);
sv *v = ss_iterof(sv_readiter, &j);
if (ssunlikely(v == NULL))
return 0;
return si_qgetresult(q, v, 1);
}
static inline int
si_qget(siquery *q)
{
ssiter i;
ss_iterinit(si_iter, &i);
ss_iteropen(si_iter, &i, q->r, q->index, SS_GTE, q->key, q->keysize);
sinode *node;
node = ss_iterof(si_iter, &i);
assert(node != NULL);
/* search in memory */
int rc;
rc = si_qgetindex(q, node);
if (rc != 0)
return rc;
/* */
rc = si_cachevalidate(q->cache, node);
if (ssunlikely(rc == -1)) {
sr_oom(q->r->e);
return -1;
}
svmerge *m = &q->merge;
rc = sv_mergeprepare(m, q->r, 1);
assert(rc == 0);
/* search on disk */
sibranch *b = node->branch;
while (b) {
rc = si_qgetbranch(q, node, b);
if (rc != 0)
return rc;
b = b->next;
}
return 0;
}
static inline void
si_qrangebranch(siquery *q, sinode *n, sibranch *b, svmerge *m)
{
sicachebranch *cb = si_cachefollow(q->cache);
assert(cb->branch == b);
/* iterate cache */
if (ss_iterhas(si_read, &cb->i)) {
svmergesrc *s = sv_mergeadd(m, &cb->i);
q->index->read_cache++;
s->ptr = cb;
return;
}
if (cb->open) {
return;
}
cb->open = 1;
sireadarg arg = {
.scheme = q->index->scheme,
.index = q->index,
.n = n,
.b = b,
.buf = &cb->buf_a,
.buf_xf = &cb->buf_b,
.buf_read = &q->index->readbuf,
.index_iter = &cb->index_iter,
.page_iter = &cb->page_iter,
.vlsn = q->vlsn,
.has = 0,
.mmap_copy = 1,
.o = q->order,
.r = q->r
};
ss_iterinit(si_read, &cb->i);
int rc = ss_iteropen(si_read, &cb->i, &arg, q->key, q->keysize);
if (ssunlikely(rc == -1))
return;
if (ssunlikely(! ss_iterhas(si_read, &cb->i)))
return;
svmergesrc *s = sv_mergeadd(m, &cb->i);
s->ptr = cb;
}
static inline int
si_qrange(siquery *q)
{
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;
/* 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 (update) */
svmergesrc *s;
sv upbuf_reserve;
ssbuf upbuf;
if (ssunlikely(q->update_v && q->update_v->v)) {
ss_bufinit_reserve(&upbuf, &upbuf_reserve, sizeof(upbuf_reserve));
ss_bufadd(&upbuf, NULL, (void*)&q->update_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) {
si_qrangebranch(q, node, b, m);
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 update search to SS_EQ */
if (q->update_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_querydup(q, v) == -1))
return -1;
}
/* skip a possible duplicates from data sources */
ss_iternext(sv_readiter, &k);
return rc;
}
int si_compaction(si *index, sdc *c, siplan *plan, uint64_t vlsn)
{
sr *r = &index->r;
sinode *node = plan->node;
assert(node->flags & SI_LOCK);
si_lock(index);
svindex *vindex;
vindex = si_noderotate(node);
si_unlock(index);
uint64_t size_stream = vindex->used;
ssiter vindex_iter;
ss_iterinit(sv_indexiter, &vindex_iter);
ss_iteropen(sv_indexiter, &vindex_iter, &index->r, vindex, SS_GTE, NULL);
/* prepare direct_io stream */
int rc;
if (index->scheme.direct_io) {
rc = sd_ioprepare(&c->io, r,
index->scheme.direct_io,
index->scheme.direct_io_page_size,
index->scheme.direct_io_buffer_size);
if (ssunlikely(rc == -1))
return sr_oom(r->e);
}
/* prepare for compaction */
svmerge merge;
sv_mergeinit(&merge);
rc = sv_mergeprepare(&merge, r, 1 + 1);
if (ssunlikely(rc == -1))
return -1;
svmergesrc *s;
s = sv_mergeadd(&merge, &vindex_iter);
sdcbuf *cbuf = &c->e;
s = sv_mergeadd(&merge, NULL);
sdreadarg arg = {
.from_compaction = 1,
.io = &c->io,
.index = &node->index,
.buf = &cbuf->a,
.buf_read = &c->d,
.index_iter = &cbuf->index_iter,
.page_iter = &cbuf->page_iter,
.use_mmap = index->scheme.mmap,
.use_mmap_copy = 0,
.use_compression = index->scheme.compression,
.use_direct_io = index->scheme.direct_io,
.direct_io_page_size = index->scheme.direct_io_page_size,
.compression_if = index->scheme.compression_if,
.has = 0,
.has_vlsn = 0,
.o = SS_GTE,
.mmap = &node->map,
.file = &node->file,
.r = r
};
ss_iterinit(sd_read, &s->src);
rc = ss_iteropen(sd_read, &s->src, &arg, NULL);
if (ssunlikely(rc == -1))
return -1;
size_stream += sd_indextotal(&node->index);
ssiter i;
ss_iterinit(sv_mergeiter, &i);
ss_iteropen(sv_mergeiter, &i, r, &merge, SS_GTE);
rc = si_merge(index, c, node, vlsn, &i, size_stream,
sd_indexkeys(&node->index));
sv_mergefree(&merge, r->a);
return rc;
}
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;
}
static inline sibranch*
si_branchcreate(si *index, sdc *c, sinode *parent, svindex *vindex, uint64_t vlsn)
{
sr *r = index->r;
svmerge vmerge;
sv_mergeinit(&vmerge);
int rc = sv_mergeprepare(&vmerge, r, 1);
if (ssunlikely(rc == -1))
return NULL;
svmergesrc *s = sv_mergeadd(&vmerge, NULL);
ss_iterinit(sv_indexiterraw, &s->src);
ss_iteropen(sv_indexiterraw, &s->src, vindex);
ssiter i;
ss_iterinit(sv_mergeiter, &i);
ss_iteropen(sv_mergeiter, &i, r, &vmerge, SS_GTE, 1);
/* merge iter is not used */
sdmergeconf mergeconf = {
.size_stream = UINT32_MAX,
.size_node = UINT64_MAX,
.size_page = index->scheme->node_page_size,
.checksum = index->scheme->node_page_checksum,
.compression = index->scheme->compression,
.compression_key = index->scheme->compression_key,
.offset = parent->file.size,
.vlsn = vlsn,
.save_delete = 1
};
sdmerge merge;
sd_mergeinit(&merge, r, &i, &c->build, &mergeconf);
rc = sd_merge(&merge);
if (ssunlikely(rc == -1)) {
sv_mergefree(&vmerge, r->a);
sr_oom_malfunction(r->e);
goto error;
}
assert(rc == 1);
sv_mergefree(&vmerge, r->a);
sibranch *branch = si_branchnew(r);
if (ssunlikely(branch == NULL))
goto error;
sdid id = {
.parent = parent->self.id.id,
.flags = SD_IDBRANCH,
.id = sr_seq(r->seq, SR_NSNNEXT)
};
rc = sd_mergecommit(&merge, &id);
if (ssunlikely(rc == -1))
goto error;
si_branchset(branch, &merge.index);
rc = sd_commit(&c->build, r, &branch->index, &parent->file);
if (ssunlikely(rc == -1)) {
si_branchfree(branch, r);
return NULL;
}
SS_INJECTION(r->i, SS_INJECTION_SI_BRANCH_0,
sr_malfunction(r->e, "%s", "error injection");
si_branchfree(branch, r);
return NULL);
if (index->scheme->sync) {
rc = si_nodesync(parent, r);
if (ssunlikely(rc == -1)) {
si_branchfree(branch, r);
return NULL;
}
}
if (index->scheme->mmap) {
ss_mmapinit(&parent->map_swap);
rc = ss_mmap(&parent->map_swap, parent->file.fd,
parent->file.size, 1);
if (ssunlikely(rc == -1)) {
sr_malfunction(r->e, "db file '%s' mmap error: %s",
parent->file.file, strerror(errno));
return NULL;
}
}
return branch;
error:
sd_mergefree(&merge);
return NULL;
}
int si_branch(si *index, sdc *c, siplan *plan, uint64_t vlsn)
{
sr *r = index->r;
sinode *n = plan->node;
assert(n->flags & SI_LOCK);
si_lock(index);
if (ssunlikely(n->used == 0)) {
si_nodeunlock(n);
si_unlock(index);
return 0;
}
svindex *i;
i = si_noderotate(n);
si_unlock(index);
sd_creset(c);
sibranch *branch = si_branchcreate(index, c, n, i, vlsn);
if (ssunlikely(branch == NULL))
return -1;
/* commit */
si_lock(index);
branch->next = n->branch;
n->branch = branch;
n->branch_count++;
uint32_t used = sv_indexused(i);
n->used -= used;
ss_quota(r->quota, SS_QREMOVE, used);
svindex swap = *i;
si_nodeunrotate(n);
si_nodeunlock(n);
si_plannerupdate(&index->p, SI_BRANCH|SI_COMPACT, n);
ssmmap swap_map = n->map;
n->map = n->map_swap;
memset(&n->map_swap, 0, sizeof(n->map_swap));
si_unlock(index);
/* gc */
if (index->scheme->mmap) {
int rc = ss_munmap(&swap_map);
if (ssunlikely(rc == -1)) {
sr_malfunction(r->e, "db file '%s' munmap error: %s",
n->file.file, strerror(errno));
return -1;
}
}
si_nodegc_index(r, &swap);
return 1;
}
static inline char*
si_noderead(si *index, ssbuf *dest, sinode *node)
{
sr *r = index->r;
if (index->scheme->mmap) {
return node->map.p;
}
int rc = ss_bufensure(dest, r->a, node->file.size);
if (ssunlikely(rc == -1)) {
sr_oom_malfunction(r->e);
return NULL;
}
rc = ss_filepread(&node->file, 0, dest->s, node->file.size);
if (ssunlikely(rc == -1)) {
sr_malfunction(r->e, "db file '%s' read error: %s",
node->file.file, strerror(errno));
return NULL;
}
ss_bufadvance(dest, node->file.size);
return dest->s;
}
int si_schemerecover(sischeme *s, sr *r)
{
sdscheme c;
sd_schemeinit(&c);
char path[PATH_MAX];
snprintf(path, sizeof(path), "%s/scheme", s->path);
int version_storage_set = 0;
int rc;
rc = sd_schemerecover(&c, r, path);
if (ssunlikely(rc == -1))
goto error;
ssiter i;
ss_iterinit(sd_schemeiter, &i);
rc = ss_iteropen(sd_schemeiter, &i, r, &c, 1);
if (ssunlikely(rc == -1))
goto error;
while (ss_iterhas(sd_schemeiter, &i))
{
sdschemeopt *opt = ss_iterof(sd_schemeiter, &i);
switch (opt->id) {
case SI_SCHEME_VERSION:
break;
case SI_SCHEME_VERSION_STORAGE: {
if (opt->size != sizeof(srversion))
goto error;
srversion *version = (srversion*)sd_schemesz(opt);
if (! sr_versionstorage_check(version))
goto error_format;
version_storage_set = 1;
break;
}
case SI_SCHEME_SCHEME: {
sf_schemefree(&s->scheme, r->a);
sf_schemeinit(&s->scheme);
ssbuf buf;
ss_bufinit(&buf);
rc = sf_schemeload(&s->scheme, r->a, sd_schemesz(opt), opt->size);
if (ssunlikely(rc == -1))
goto error;
rc = sf_schemevalidate(&s->scheme, r->a);
if (ssunlikely(rc == -1))
goto error;
ss_buffree(&buf, r->a);
break;
}
case SI_SCHEME_NODE_SIZE:
s->compaction.node_size = sd_schemeu64(opt);
break;
case SI_SCHEME_NODE_PAGE_SIZE:
s->compaction.node_page_size = sd_schemeu32(opt);
break;
case SI_SCHEME_COMPRESSION: {
char *name = sd_schemesz(opt);
ssfilterif *cif = ss_filterof(name);
if (ssunlikely(cif == NULL))
goto error;
s->compression_if = cif;
s->compression = s->compression_if != &ss_nonefilter;
ss_free(r->a, s->compression_sz);
s->compression_sz = ss_strdup(r->a, cif->name);
if (ssunlikely(s->compression_sz == NULL))
goto error;
break;
}
case SI_SCHEME_EXPIRE:
s->expire = sd_schemeu32(opt);
break;
default: /* skip unknown */
break;
}
ss_iternext(sd_schemeiter, &i);
}
if (ssunlikely(! version_storage_set))
goto error_format;
sd_schemefree(&c, r);
return 0;
error_format:
sr_error(r->e, "%s", "incompatible storage format version");
error:
sd_schemefree(&c, r);
return -1;
}
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;
}
