RepoMirrors/ceph
1
0
Fork 0
mirror of https://github.com/ceph/ceph synced 2025-03-30 23:40:09 +00:00
Code Issues Packages Projects Releases Wiki Activity

osd: Improve pg degraded state setting based on _update_calc_stats() degraded count

Browse Source 
Signed-off-by: David Zafman <dzafman@redhat.com>
This commit is contained in:
David Zafman 2017-10-31 18:15:53 -07:00
parent b769b98686
commit 74f9e70903
3 changed files with 24 additions and 22 deletions

42
src/osd/PG.cc
View File

@ -2016,8 +2016,11 @@ void PG::all_activated_and_committed()
// Degraded?
_update_calc_stats();
if (info.stats.stats.sum.num_objects_degraded)
if (info.stats.stats.sum.num_objects_degraded) {
state_set(PG_STATE_DEGRADED);
} else {
state_clear(PG_STATE_DEGRADED);
}
queue_peering_event(
PGPeeringEventRef(
@ -2107,7 +2110,7 @@ bool PG::set_force_recovery(bool b)
if (!deleting) {
if (b) {
if (!(state & PG_STATE_FORCED_RECOVERY) &&
(state & (PG_STATE_DEGRADED |
(state & (PG_STATE_DEGRADED | // XXX: Will this check be messed up?
PG_STATE_RECOVERY_WAIT |
PG_STATE_RECOVERING))) {
dout(20) << __func__ << " set" << dendl;
@ -2133,7 +2136,7 @@ bool PG::set_force_backfill(bool b)
if (!deleting) {
if (b) {
if (!(state & PG_STATE_FORCED_RECOVERY) &&
(state & (PG_STATE_DEGRADED |
(state & (PG_STATE_DEGRADED | // XXX: Will this check be messed up?
PG_STATE_BACKFILL_WAIT |
PG_STATE_BACKFILLING))) {
dout(10) << __func__ << " set" << dendl;
@ -2693,8 +2696,10 @@ void PG::_update_calc_stats()
info.stats.stats.sum.num_objects_degraded = 0;
info.stats.stats.sum.num_objects_unfound = 0;
info.stats.stats.sum.num_objects_misplaced = 0;
if (!is_clean() && (is_peered() || is_activating())) {
dout(20) << __func__ << " not clean" << dendl;
if ((is_remapped() || is_undersized() || !is_clean()) && (is_peered() || is_activating())) {
dout(20) << __func__ << " actingset " << actingset << " upset "
<< upset << " actingbackfill " << actingbackfill << dendl;
dout(20) << __func__ << " acting " << acting << " up " << up << dendl;
assert(!actingbackfill.empty());
@ -2731,12 +2736,11 @@ void PG::_update_calc_stats()
// all other peers track missing accurately.
if (is_backfill_targets(peer.first)) {
missing = std::max((int64_t)0, num_objects - peer.second.stats.stats.sum.num_objects);
if (missing < 0) missing = 0;
} else {
if (peer_missing.count(peer.first)) {
missing = peer_missing[peer.first].num_missing();
} else {
dout(20) << __func__ << " no peer_mssing found for " << peer.first << dendl;
dout(20) << __func__ << " no peer_missing found for " << peer.first << dendl;
}
}
if (upset.count(peer.first)) {
@ -2780,6 +2784,11 @@ void PG::_update_calc_stats()
degraded += m->first;
}
}
if (target > upset.size()) {
int64_t undersize_degraded = (num_objects * (target - upset.size()));
degraded += undersize_degraded;
dout(20) << __func__ << " undersize degraded " << undersize_degraded << dendl;
}
dout(20) << __func__ << " degraded " << degraded << dendl;
dout(20) << __func__ << " misplaced " << misplaced << dendl;
@ -2845,6 +2854,11 @@ void PG::publish_stats_to_osd()
}
_update_calc_stats();
if (info.stats.stats.sum.num_objects_degraded) {
state_set(PG_STATE_DEGRADED);
} else {
state_clear(PG_STATE_DEGRADED);
}
_update_blocked_by();
pg_stat_t pre_publish = info.stats;
@ -7149,7 +7163,6 @@ PG::RecoveryState::Recovered::Recovered(my_context ctx)
assert(!pg->actingbackfill.empty());
if (pg->get_osdmap()->get_pg_size(pg->info.pgid.pgid) <=
pg->actingbackfill.size()) {
pg->state_clear(PG_STATE_DEGRADED);
pg->state_clear(PG_STATE_FORCED_BACKFILL | PG_STATE_FORCED_RECOVERY);
pg->publish_stats_to_osd();
}
@ -7379,16 +7392,6 @@ boost::statechart::result PG::RecoveryState::Active::react(const AdvMap& advmap)
}
bool need_publish = false;
pg->_update_calc_stats(); // Too bad this is called again from publish_stats_to_osd()
int64_t degraded = pg->info.stats.stats.sum.num_objects_degraded;
if (!pg->state_test(PG_STATE_DEGRADED) && degraded) {
need_publish = true;
pg->state_set(PG_STATE_DEGRADED);
} else if (pg->state_test(PG_STATE_DEGRADED) && !degraded) {
need_publish = true;
pg->state_clear(PG_STATE_DEGRADED);
}
/* Check for changes in pool size (if the acting set changed as a result,
* this does not matter) */
if (advmap.lastmap->get_pg_size(pg->info.pgid.pgid) !=
@ -7398,7 +7401,8 @@ boost::statechart::result PG::RecoveryState::Active::react(const AdvMap& advmap)
} else {
pg->state_set(PG_STATE_UNDERSIZED);
}
need_publish = true; // undersized may have changed
// degraded changes will be detected by call from publish_stats_to_osd()
need_publish = true;
}
// if we haven't reported our PG stats in a long time, do so now.

1
src/osd/PG.h
View File

@ -2688,6 +2688,7 @@ protected:
bool is_degraded() const { return state_test(PG_STATE_DEGRADED); }
bool is_undersized() const { return state_test(PG_STATE_UNDERSIZED); }
bool is_scrubbing() const { return state_test(PG_STATE_SCRUBBING); }
bool is_remapped() const { return state_test(PG_STATE_REMAPPED); }
bool is_peered() const {
return state_test(PG_STATE_ACTIVE) || state_test(PG_STATE_PEERED);
}

3
src/osd/PrimaryLogPG.cc
View File

@ -12025,9 +12025,6 @@ bool PrimaryLogPG::start_recovery_ops(
if (state_test(PG_STATE_RECOVERING)) {
state_clear(PG_STATE_RECOVERING);
state_clear(PG_STATE_FORCED_RECOVERY);
if (get_osdmap()->get_pg_size(info.pgid.pgid) <= acting.size()) {
state_clear(PG_STATE_DEGRADED);
}
if (needs_backfill()) {
dout(10) << "recovery done, queuing backfill" << dendl;
queue_peering_event(