Schedule automatic scrubs correctly (not just after previous scrub)

test-double-alloc
Vitaliy Filippov 2023-04-22 11:17:38 +03:00
parent 8d40ad99a6
commit fa90b5a4e7
4 changed files with 33 additions and 5 deletions

View File

@ -210,6 +210,7 @@ void osd_t::parse_config(bool init)
inode_vanish_time = config["inode_vanish_time"].uint64_value();
if (!inode_vanish_time)
inode_vanish_time = 60;
auto old_auto_scrub = auto_scrub;
auto_scrub = json_is_true(config["auto_scrub"]);
global_scrub_interval = parse_time(config["scrub_interval"].string_value());
if (!global_scrub_interval)
@ -224,6 +225,14 @@ void osd_t::parse_config(bool init)
scrub_list_limit = config["scrub_list_limit"].uint64_value();
if (!scrub_list_limit)
scrub_list_limit = 1000;
if (!old_auto_scrub && auto_scrub)
{
// Schedule scrubbing
for (auto & pgp: pgs)
{
plan_scrub(pgp.second);
}
}
if (old_no_scrub && !no_scrub)
{
// Wakeup scrubbing

View File

@ -243,6 +243,7 @@ class osd_t
int pick_next_scrub(object_id & next_oid);
void submit_scrub_op(object_id oid);
bool continue_scrub();
void plan_scrub(pg_t & pg, bool report_state = true);
void schedule_scrub(pg_t & pg);
// op execution

View File

@ -24,8 +24,8 @@ void osd_t::handle_peers()
if (!p.second.peering_state->list_ops.size())
{
p.second.calc_object_states(log_level);
schedule_scrub(p.second);
report_pg_state(p.second);
schedule_scrub(p.second);
incomplete_objects += p.second.incomplete_objects.size();
misplaced_objects += p.second.misplaced_objects.size();
// FIXME: degraded objects may currently include misplaced, too! Report them separately?
@ -494,6 +494,7 @@ void osd_t::report_pg_state(pg_t & pg)
pg.all_peers = pg.target_set;
std::sort(pg.all_peers.begin(), pg.all_peers.end());
pg.cur_peers = pg.target_set;
plan_scrub(pg, false);
// Change pg_config at the same time, otherwise our PG reconciling loop may try to apply the old metadata
auto & pg_cfg = st_cli.pool_config[pg.pool_id].pg_config[pg.pg_num];
pg_cfg.target_history = pg.target_history;
@ -537,6 +538,7 @@ void osd_t::report_pg_state(pg_t & pg)
pg.cur_peers.push_back(pg_osd);
}
}
plan_scrub(pg, false);
auto & pg_cfg = st_cli.pool_config[pg.pool_id].pg_config[pg.pg_num];
pg_cfg.target_history = pg.target_history;
pg_cfg.all_peers = pg.all_peers;

View File

@ -192,13 +192,10 @@ int osd_t::pick_next_scrub(object_id & next_oid)
if (pg_it->second.state & PG_SCRUBBING)
{
scrub_last_pg = {};
auto & pool_cfg = st_cli.pool_config.at(pg_it->first.pool_id);
auto interval = pool_cfg.scrub_interval ? pool_cfg.scrub_interval : global_scrub_interval;
pg_it->second.next_scrub = auto_scrub ? tv_now.tv_sec + interval : 0;
pg_it->second.state = pg_it->second.state & ~PG_SCRUBBING;
pg_it->second.next_scrub = 0;
pg_it->second.history_changed = true;
report_pg_state(pg_it->second);
schedule_scrub(pg_it->second);
}
// The list is definitely not needed anymore
if (scrub_cur_list.buf)
@ -318,6 +315,25 @@ bool osd_t::continue_scrub()
return true;
}
void osd_t::plan_scrub(pg_t & pg, bool report_state)
{
if ((pg.state & PG_ACTIVE) && !pg.next_scrub && auto_scrub)
{
timespec tv_now;
clock_gettime(CLOCK_REALTIME, &tv_now);
auto & pool_cfg = st_cli.pool_config.at(pg.pool_id);
auto interval = pool_cfg.scrub_interval ? pool_cfg.scrub_interval : global_scrub_interval;
if (pg.next_scrub != tv_now.tv_sec + interval)
{
pool_cfg.pg_config[pg.pg_num].next_scrub = pg.next_scrub = tv_now.tv_sec + interval;
pg.history_changed = true;
if (report_state)
report_pg_state(pg);
}
schedule_scrub(pg);
}
}
void osd_t::schedule_scrub(pg_t & pg)
{
if (!no_scrub && pg.next_scrub && (!scrub_nearest_ts || scrub_nearest_ts > pg.next_scrub))