osd, PG: share the snaptrimq.size() within pg_stat_t

That way it will be unnecessary to go through all pgs separately
to find pgs with excessively long snap trim queues. And we don't need
to share snap trim queues itself, which may be large by itself.
As snap trim queues tend to be short and anything above 50 000
I consider absurdly large, the snaptrimq_len is capped at 2^32 to
save space in pg_stat_t.

Signed-off-by: Piotr Dałek <piotr.dalek@corp.ovh.com>
(cherry picked from commit 022d246f17b31d309525395831c84e9856533adb)

Conflicts:
	src/osd/osd_types.cc
     - needed a fix for snaptrimq_len that was placed after two other
new fields

diff --git a/src/mon/PGMap.cc b/src/mon/PGMap.cc
index 2373691a3f74973cfcdfffc550a3e73bbe5a4660..834da03debc922969051ca8a1a03211075b28f09 100644 (file)
--- a/src/mon/PGMap.cc
+++ b/src/mon/PGMap.cc
@@ -1785,6 +1785,7 @@ void PGMap::dump_pg_stats_plain(
     tab.define_column("SCRUB_STAMP", TextTable::LEFT, TextTable::RIGHT);
     tab.define_column("LAST_DEEP_SCRUB", TextTable::LEFT, TextTable::RIGHT);
     tab.define_column("DEEP_SCRUB_STAMP", TextTable::LEFT, TextTable::RIGHT);
+    tab.define_column("SNAPTRIMQ_LEN", TextTable::LEFT, TextTable::RIGHT);
   }
 
   for (auto i = pg_stats.begin();
@@ -1823,6 +1824,7 @@ void PGMap::dump_pg_stats_plain(
           << st.last_scrub_stamp
           << st.last_deep_scrub
           << st.last_deep_scrub_stamp
+          << st.snaptrimq_len
           << TextTable::endrow;
     }
   }

diff --git a/src/osd/PG.cc b/src/osd/PG.cc
index 139a9ac6d17941d835bfecbea14b2e2577d3f072..255e0d3751552cdea01701ff43d60566d8f3f7c2 100644 (file)
--- a/src/osd/PG.cc
+++ b/src/osd/PG.cc
@@ -2566,6 +2566,7 @@ void PG::_update_calc_stats()
   info.stats.ondisk_log_size = info.stats.log_size;
   info.stats.log_start = pg_log.get_tail();
   info.stats.ondisk_log_start = pg_log.get_tail();
+  info.stats.snaptrimq_len = snap_trimq.size();
 
   // If actingset is larger then upset we will have misplaced,
   // so we will report based on actingset size.

diff --git a/src/osd/osd_types.cc b/src/osd/osd_types.cc
index b22001af6f003d62a0d4cdecf30b6ac55f47b382..a8f50467affab3881106b54d329f4bafd0cb9a99 100644 (file)
--- a/src/osd/osd_types.cc
+++ b/src/osd/osd_types.cc
@@ -2287,6 +2287,7 @@ void pg_stat_t::dump(Formatter *f) const
   f->dump_bool("hitset_stats_invalid", hitset_stats_invalid);
   f->dump_bool("hitset_bytes_stats_invalid", hitset_bytes_stats_invalid);
   f->dump_bool("pin_stats_invalid", pin_stats_invalid);
+  f->dump_unsigned("snaptrimq_len", snaptrimq_len);
   stats.dump(f);
   f->open_array_section("up");
   for (vector<int32_t>::const_iterator p = up.begin(); p != up.end(); ++p)
@@ -2322,7 +2323,7 @@ void pg_stat_t::dump_brief(Formatter *f) const
 
 void pg_stat_t::encode(bufferlist &bl) const
 {
-  ENCODE_START(22, 22, bl);
+  ENCODE_START(23, 22, bl);
   ::encode(version, bl);
   ::encode(reported_seq, bl);
   ::encode(reported_epoch, bl);
@@ -2363,6 +2364,7 @@ void pg_stat_t::encode(bufferlist &bl) const
   ::encode(last_peered, bl);
   ::encode(last_became_peered, bl);
   ::encode(pin_stats_invalid, bl);
+  ::encode(snaptrimq_len, bl);
   ENCODE_FINISH(bl);
 }
 
@@ -2416,6 +2418,9 @@ void pg_stat_t::decode(bufferlist::iterator &bl)
   ::decode(last_became_peered, bl);
   ::decode(tmp, bl);
   pin_stats_invalid = tmp;
+  if (struct_v >= 23) {
+    ::decode(snaptrimq_len, bl);
+  }
   DECODE_FINISH(bl);
 }
 
@@ -2447,6 +2452,7 @@ void pg_stat_t::generate_test_instances(list<pg_stat_t*>& o)
   a.last_deep_scrub = eversion_t(13, 14);
   a.last_deep_scrub_stamp = utime_t(15, 16);
   a.last_clean_scrub_stamp = utime_t(17, 18);
+  a.snaptrimq_len = 1048576;
   list<object_stat_collection_t*> l;
   object_stat_collection_t::generate_test_instances(l);
   a.stats = *l.back();
@@ -2509,7 +2515,8 @@ bool operator==(const pg_stat_t& l, const pg_stat_t& r)
     l.hitset_bytes_stats_invalid == r.hitset_bytes_stats_invalid &&
     l.up_primary == r.up_primary &&
     l.acting_primary == r.acting_primary &&
-    l.pin_stats_invalid == r.pin_stats_invalid;
+    l.pin_stats_invalid == r.pin_stats_invalid &&
+    l.snaptrimq_len == r.snaptrimq_len;
 }
 
 // -- pool_stat_t --

diff --git a/src/osd/osd_types.h b/src/osd/osd_types.h
index a820c8f6bfa70e8e63881310e5b038ca3fee868b..b2afa4658d105cf534ca20746664e0a071d07da1 100644 (file)
--- a/src/osd/osd_types.h
+++ b/src/osd/osd_types.h
@@ -1947,6 +1947,10 @@ struct pg_stat_t {
   int32_t up_primary;
   int32_t acting_primary;
 
+  // snaptrimq.size() is 64bit, but let's be serious - anything over 50k is
+  // absurd already, so cap it to 2^32 and save 4 bytes at  the same time
+  uint32_t snaptrimq_len;
+
   bool stats_invalid:1;
   /// true if num_objects_dirty is not accurate (because it was not
   /// maintained starting from pool creation)
@@ -1966,6 +1970,7 @@ struct pg_stat_t {
       mapping_epoch(0),
       up_primary(-1),
       acting_primary(-1),
+      snaptrimq_len(0),
       stats_invalid(false),
       dirty_stats_invalid(false),
       omap_stats_invalid(false),
@@ -1994,17 +1999,29 @@ struct pg_stat_t {
       log_size = f;
     if (ondisk_log_size < f)
       ondisk_log_size = f;
+    if (snaptrimq_len < f)
+      snaptrimq_len = f;
   }
 
   void add(const pg_stat_t& o) {
     stats.add(o.stats);
     log_size += o.log_size;
     ondisk_log_size += o.ondisk_log_size;
+    if (((uint64_t)snaptrimq_len + (uint64_t)o.snaptrimq_len) > (uint64_t)(1 << 31)) {
+      snaptrimq_len = 1 << 31;
+    } else {
+      snaptrimq_len += o.snaptrimq_len;
+    }
   }
   void sub(const pg_stat_t& o) {
     stats.sub(o.stats);
     log_size -= o.log_size;
     ondisk_log_size -= o.ondisk_log_size;
+    if (o.snaptrimq_len < snaptrimq_len) {
+      snaptrimq_len -= o.snaptrimq_len;
+    } else {
+      snaptrimq_len = 0;
+    }
   }
 
   bool is_acting_osd(int32_t osd, bool primary) const;