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;
}
static inline void
sx_end(sx *t)
{
sxmanager *m = t->manager;
ss_spinlock(&m->lock);
ss_rbremove(&m->i, &t->node);
m->count--;
ss_spinunlock(&m->lock);
}
static inline int
ss_testvfs_call(ssvfs *f)
{
sstestvfs *o = (sstestvfs*)f->priv;
ss_spinlock(&o->lock);
int generate_fail = o->n >= o->fail_from;
o->n++;
ss_spinunlock(&o->lock);
return generate_fail;
}
uint64_t sx_max(sxmanager *m)
{
ss_spinlock(&m->lock);
uint64_t id = 0;
if (sx_count(m) > 0) {
ssrbnode *node = ss_rbmax(&m->i);
sx *max = sscast(node, sx, node);
id = max->id;
}
ss_spinunlock(&m->lock);
return id;
}
static inline void
sx_end(sx *x)
{
sxmanager *m = x->manager;
ss_spinlock(&m->lock);
ss_rbremove(&m->i, &x->node);
if (x->type == SXRO)
m->count_rd--;
else
m->count_rw--;
ss_spinunlock(&m->lock);
}
uint32_t sx_min(sxmanager *m)
{
ss_spinlock(&m->lock);
uint32_t id = 0;
if (sx_count(m) > 0) {
ssrbnode *node = ss_rbmin(&m->i);
sx *min = sscast(node, sx, node);
id = min->id;
}
ss_spinunlock(&m->lock);
return id;
}
static inline sshot void
ss_slabafree(ssa *a, void *ptr)
{
ssslaba *s = (ssslaba*)a->priv;
assert(ptr != NULL);
ss_spinlock(&s->lock);
*(char**)ptr = s->pool_free;
s->pool_free = ptr;
s->pool_free_count++;
ss_spinunlock(&s->lock);
}
uint64_t sx_vlsn(sxmanager *m)
{
ss_spinlock(&m->lock);
uint64_t vlsn;
if (sx_count(m) > 0) {
ssrbnode *node = ss_rbmin(&m->i);
sx *min = sscast(node, sx, node);
vlsn = min->vlsn;
} else {
vlsn = sr_seq(m->r->seq, SR_LSN);
}
ss_spinunlock(&m->lock);
return vlsn;
}
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 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 int
se_schedule(sescheduler *s, setask *task, seworker *w)
{
ss_trace(&w->trace, "%s", "schedule");
si_planinit(&task->plan);
uint64_t now = ss_utime();
se *e = (se*)s->env;
sedb *db;
srzone *zone = se_zoneof(e);
assert(zone != NULL);
task->checkpoint_complete = 0;
task->backup_complete = 0;
task->rotate = 0;
task->req = 0;
task->gc = 0;
task->db = NULL;
ss_mutexlock(&s->lock);
/* asynchronous reqs dispatcher */
if (s->req == 0) {
switch (zone->async) {
case 2:
if (se_reqqueue(e) == 0)
break;
case 1:
s->req = 1;
task->req = zone->async;
ss_mutexunlock(&s->lock);
return 0;
}
}
/* log gc and rotation */
if (s->rotate == 0)
{
task->rotate = 1;
s->rotate = 1;
}
/* checkpoint */
int in_progress = 0;
int rc;
checkpoint:
if (s->checkpoint) {
task->plan.plan = SI_CHECKPOINT;
task->plan.a = s->checkpoint_lsn;
rc = se_schedule_plan(s, &task->plan, &db);
switch (rc) {
case 1:
s->workers_branch++;
se_dbref(db, 1);
task->db = db;
task->gc = 1;
ss_mutexunlock(&s->lock);
return 1;
case 2: /* work in progress */
in_progress = 1;
break;
case 0: /* complete checkpoint */
s->checkpoint = 0;
s->checkpoint_lsn_last = s->checkpoint_lsn;
s->checkpoint_lsn = 0;
task->checkpoint_complete = 1;
break;
}
}
/* apply zone policy */
switch (zone->mode) {
case 0: /* compact_index */
case 1: /* compact_index + branch_count prio */
assert(0);
break;
case 2: /* checkpoint */
{
if (in_progress) {
ss_mutexunlock(&s->lock);
return 0;
}
uint64_t lsn = sr_seq(&e->seq, SR_LSN);
s->checkpoint_lsn = lsn;
s->checkpoint = 1;
goto checkpoint;
}
default: /* branch + compact */
assert(zone->mode == 3);
}
/* database shutdown-drop */
if (s->workers_gc_db < zone->gc_db_prio) {
ss_spinlock(&e->dblock);
db = NULL;
if (ssunlikely(e->db_shutdown.n > 0)) {
db = (sedb*)so_listfirst(&e->db_shutdown);
if (se_dbgarbage(db)) {
so_listdel(&e->db_shutdown, &db->o);
} else {
db = NULL;
}
}
ss_spinunlock(&e->dblock);
if (ssunlikely(db)) {
if (db->dropped)
task->plan.plan = SI_DROP;
else
task->plan.plan = SI_SHUTDOWN;
s->workers_gc_db++;
se_dbref(db, 1);
task->db = db;
ss_mutexunlock(&s->lock);
return 1;
}
}
/* backup */
if (s->backup && (s->workers_backup < zone->backup_prio))
{
/* backup procedure.
*
* state 0 (start)
* -------
*
* a. disable log gc
* b. mark to start backup (state 1)
*
* state 1 (background, delayed start)
* -------
*
* a. create backup_path/<bsn.incomplete> directory
* b. create database directories
* c. create log directory
* d. state 2
*
* state 2 (background, copy)
* -------
*
* a. schedule and execute node backup which bsn < backup_bsn
* b. state 3
*
* state 3 (background, completion)
* -------
*
* a. rotate log file
* b. copy log files
* c. enable log gc, schedule gc
* d. rename <bsn.incomplete> into <bsn>
* e. set last backup, set COMPLETE
*
*/
if (s->backup == 1) {
/* state 1 */
rc = se_backupstart(s);
if (ssunlikely(rc == -1)) {
se_backuperror(s);
goto backup_error;
}
s->backup = 2;
}
/* state 2 */
task->plan.plan = SI_BACKUP;
task->plan.a = s->backup_bsn;
rc = se_schedule_plan(s, &task->plan, &db);
switch (rc) {
case 1:
s->workers_backup++;
se_dbref(db, 1);
task->db = db;
ss_mutexunlock(&s->lock);
return 1;
case 2: /* work in progress */
break;
case 0: /* state 3 */
rc = se_backupcomplete(s, w);
if (ssunlikely(rc == -1)) {
se_backuperror(s);
goto backup_error;
}
s->backup_events++;
task->gc = 1;
task->backup_complete = 1;
break;
}
backup_error:;
}
/* garbage-collection */
if (s->gc) {
if (s->workers_gc < zone->gc_prio) {
task->plan.plan = SI_GC;
task->plan.a = sx_vlsn(&e->xm);
task->plan.b = zone->gc_wm;
rc = se_schedule_plan(s, &task->plan, &db);
switch (rc) {
case 1:
s->workers_gc++;
se_dbref(db, 1);
task->db = db;
ss_mutexunlock(&s->lock);
return 1;
case 2: /* work in progress */
break;
case 0: /* state 3 */
s->gc = 0;
s->gc_last = now;
break;
}
}
} else {
if (zone->gc_prio && zone->gc_period) {
if ( (now - s->gc_last) >= ((uint64_t)zone->gc_period * 1000000) ) {
s->gc = 1;
}
}
}
/* index aging */
if (s->age) {
if (s->workers_branch < zone->branch_prio) {
task->plan.plan = SI_AGE;
task->plan.a = zone->branch_age * 1000000; /* ms */
task->plan.b = zone->branch_age_wm;
rc = se_schedule_plan(s, &task->plan, &db);
switch (rc) {
case 1:
s->workers_branch++;
se_dbref(db, 1);
task->db = db;
ss_mutexunlock(&s->lock);
return 1;
case 0:
s->age = 0;
s->age_last = now;
break;
}
}
} else {
if (zone->branch_prio && zone->branch_age_period) {
if ( (now - s->age_last) >= ((uint64_t)zone->branch_age_period * 1000000) ) {
s->age = 1;
}
}
}
/* branching */
if (s->workers_branch < zone->branch_prio)
{
/* schedule branch task using following
* priority:
*
* a. peek node with the largest in-memory index
* which is equal or greater then branch
* watermark.
* If nothing is found, stick to b.
*
* b. peek node with the largest in-memory index,
* which has oldest update time.
*
* c. if no branch work is needed, schedule a
* compaction job
*
*/
task->plan.plan = SI_BRANCH;
task->plan.a = zone->branch_wm;
rc = se_schedule_plan(s, &task->plan, &db);
if (rc == 1) {
s->workers_branch++;
se_dbref(db, 1);
task->db = db;
task->gc = 1;
ss_mutexunlock(&s->lock);
return 1;
}
}
/* compaction */
task->plan.plan = SI_COMPACT;
task->plan.a = zone->compact_wm;
task->plan.b = zone->compact_mode;
rc = se_schedule_plan(s, &task->plan, &db);
if (rc == 1) {
se_dbref(db, 1);
task->db = db;
ss_mutexunlock(&s->lock);
return 1;
}
ss_mutexunlock(&s->lock);
return 0;
}
