osd/osd_types: pg_history_t: add prior_readable_until_ub

Signed-off-by: Sage Weil <sage@redhat.com>

diff --git a/src/osd/osd_types.cc b/src/osd/osd_types.cc
index f44676254d39730f78086cc44e3046c40aea0864..d2be580f4e9c3eea4d8fca864fc3c2e578fc8a48 100644 (file)
--- a/src/osd/osd_types.cc
+++ b/src/osd/osd_types.cc
@@ -3237,7 +3237,7 @@ void pool_stat_t::generate_test_instances(list<pool_stat_t*>& o)
 
 void pg_history_t::encode(ceph::buffer::list &bl) const
 {
-  ENCODE_START(9, 4, bl);
+  ENCODE_START(10, 4, bl);
   encode(epoch_created, bl);
   encode(last_epoch_started, bl);
   encode(last_epoch_clean, bl);
@@ -3254,12 +3254,13 @@ void pg_history_t::encode(ceph::buffer::list &bl) const
   encode(last_interval_started, bl);
   encode(last_interval_clean, bl);
   encode(epoch_pool_created, bl);
+  encode(prior_readable_until_ub, bl);
   ENCODE_FINISH(bl);
 }
 
 void pg_history_t::decode(ceph::buffer::list::const_iterator &bl)
 {
-  DECODE_START_LEGACY_COMPAT_LEN(9, 4, 4, bl);
+  DECODE_START_LEGACY_COMPAT_LEN(10, 4, 4, bl);
   decode(epoch_created, bl);
   decode(last_epoch_started, bl);
   if (struct_v >= 3)
@@ -3304,6 +3305,9 @@ void pg_history_t::decode(ceph::buffer::list::const_iterator &bl)
   } else {
     epoch_pool_created = epoch_created;
   }
+  if (struct_v >= 10) {
+    decode(prior_readable_until_ub, bl);
+  }
   DECODE_FINISH(bl);
 }
 
@@ -3325,6 +3329,9 @@ void pg_history_t::dump(Formatter *f) const
   f->dump_stream("last_deep_scrub") << last_deep_scrub;
   f->dump_stream("last_deep_scrub_stamp") << last_deep_scrub_stamp;
   f->dump_stream("last_clean_scrub_stamp") << last_clean_scrub_stamp;
+  f->dump_float(
+    "prior_readable_until_ub",
+    std::chrono::duration<double>(prior_readable_until_ub).count());
 }
 
 void pg_history_t::generate_test_instances(list<pg_history_t*>& o)
@@ -3338,6 +3345,7 @@ void pg_history_t::generate_test_instances(list<pg_history_t*>& o)
   o.back()->last_epoch_clean = 3;
   o.back()->last_interval_clean = 2;
   o.back()->last_epoch_split = 4;
+  o.back()->prior_readable_until_ub = make_timespan(3.1415);
   o.back()->same_up_since = 5;
   o.back()->same_interval_since = 6;
   o.back()->same_primary_since = 7;

diff --git a/src/osd/osd_types.h b/src/osd/osd_types.h
index dc37e96d62c36a183f820dec5e9205d85ef86ab5..52b511549f3f202f2966b4d3a6a460fca854e33c 100644 (file)
--- a/src/osd/osd_types.h
+++ b/src/osd/osd_types.h
@@ -2631,7 +2631,7 @@ struct pg_history_t {
   epoch_t last_interval_clean = 0;; // first epoch of last_epoch_clean interval
   epoch_t last_epoch_split = 0;;    // as parent or child
   epoch_t last_epoch_marked_full = 0;;  // pool or cluster
-  
+
   /**
    * In the event of a map discontinuity, same_*_since may reflect the first
    * map the osd has seen in the new map sequence rather than the actual start
@@ -2649,6 +2649,9 @@ struct pg_history_t {
   utime_t last_deep_scrub_stamp;
   utime_t last_clean_scrub_stamp;
 
+  /// upper bound on how long prior interval readable (relative to encode time)
+  ceph::timespan prior_readable_until_ub = ceph::timespan::zero();
+
   friend bool operator==(const pg_history_t& l, const pg_history_t& r) {
     return
       l.epoch_created == r.epoch_created &&
@@ -2666,7 +2669,8 @@ struct pg_history_t {
       l.last_deep_scrub == r.last_deep_scrub &&
       l.last_scrub_stamp == r.last_scrub_stamp &&
       l.last_deep_scrub_stamp == r.last_deep_scrub_stamp &&
-      l.last_clean_scrub_stamp == r.last_clean_scrub_stamp;
+      l.last_clean_scrub_stamp == r.last_clean_scrub_stamp &&
+      l.prior_readable_until_ub == r.prior_readable_until_ub;
   }
 
   pg_history_t() {}
@@ -2699,6 +2703,15 @@ struct pg_history_t {
     }
     if (last_interval_started < other.last_interval_started) {
       last_interval_started = other.last_interval_started;
+      // if we are learning about a newer *started* interval, our
+      // readable_until_ub is obsolete
+      prior_readable_until_ub = other.prior_readable_until_ub;
+      modified = true;
+    } else if (other.last_interval_started == last_interval_started &&
+              other.prior_readable_until_ub < prior_readable_until_ub) {
+      // if other is the *same* interval, than pull our upper bound in
+      // if they have a tighter bound.
+      prior_readable_until_ub = other.prior_readable_until_ub;
       modified = true;
     }
     if (last_epoch_clean < other.last_epoch_clean) {
@@ -2748,12 +2761,16 @@ struct pg_history_t {
 WRITE_CLASS_ENCODER(pg_history_t)
 
 inline std::ostream& operator<<(std::ostream& out, const pg_history_t& h) {
-  return out << "ec=" << h.epoch_created << "/" << h.epoch_pool_created
-            << " lis/c=" << h.last_interval_started
-            << "/" << h.last_interval_clean
-            << " les/c/f=" << h.last_epoch_started << "/" << h.last_epoch_clean
-            << "/" << h.last_epoch_marked_full
-            << " sis=" << h.same_interval_since;
+  out << "ec=" << h.epoch_created << "/" << h.epoch_pool_created
+      << " lis/c=" << h.last_interval_started
+      << "/" << h.last_interval_clean
+      << " les/c/f=" << h.last_epoch_started << "/" << h.last_epoch_clean
+      << "/" << h.last_epoch_marked_full
+      << " sis=" << h.same_interval_since;
+  if (h.prior_readable_until_ub != ceph::timespan::zero()) {
+    out << " pruub=" << h.prior_readable_until_ub;
+  }
+  return out;
 }