objecter: resend linger ops on split

src/osd/osd_types.cc

@@ -2196,6 +2196,57 @@ void pg_interval_t::generate_test_instances(list<pg_interval_t*>& o)
   o.back()->maybe_went_rw = true;
 }
 
+bool pg_interval_t::is_new_interval(
+  int old_acting_primary,
+  int new_acting_primary,
+  const vector<int> &old_acting,
+  const vector<int> &new_acting,
+  int old_up_primary,
+  int new_up_primary,
+  const vector<int> &old_up,
+  const vector<int> &new_up,
+  int old_min_size,
+  int new_min_size,
+  unsigned old_pg_num,
+  unsigned new_pg_num,
+  pg_t pgid) {
+  return old_acting_primary != new_acting_primary ||
+    new_acting != old_acting ||
+    old_up_primary != new_up_primary ||
+    new_up != old_up ||
+    old_min_size != new_min_size ||
+    pgid.is_split(old_pg_num, new_pg_num, 0);
+}
+
+bool pg_interval_t::is_new_interval(
+  int old_acting_primary,
+  int new_acting_primary,
+  const vector<int> &old_acting,
+  const vector<int> &new_acting,
+  int old_up_primary,
+  int new_up_primary,
+  const vector<int> &old_up,
+  const vector<int> &new_up,
+  OSDMapRef osdmap,
+  OSDMapRef lastmap,
+  int64_t pool_id,
+  pg_t pgid) {
+  return !(lastmap->get_pools().count(pgid.pool())) ||
+    is_new_interval(old_acting_primary,
+		    new_acting_primary,
+		    old_acting,
+		    new_acting,
+		    old_up_primary,
+		    new_up_primary,
+		    old_up,
+		    new_up,
+		    lastmap->get_pools().find(pgid.pool())->second.min_size,
+		    osdmap->get_pools().find(pgid.pool())->second.min_size,
+		    lastmap->get_pg_num(pgid.pool()),
+		    osdmap->get_pg_num(pgid.pool()),
+		    pgid);
+}
+
 bool pg_interval_t::check_new_interval(
   int old_acting_primary,
   int new_acting_primary,
@@ -2218,15 +2269,19 @@ bool pg_interval_t::check_new_interval(
   //  NOTE: a change in the up set primary triggers an interval
   //  change, even though the interval members in the pg_interval_t
   //  do not change.
-  if (old_acting_primary != new_acting_primary ||
-      new_acting != old_acting ||
-      old_up_primary != new_up_primary ||
-      new_up != old_up ||
-      (!(lastmap->get_pools().count(pool_id))) ||
-      (lastmap->get_pools().find(pool_id)->second.min_size !=
-       osdmap->get_pools().find(pool_id)->second.min_size)  ||
-      pgid.is_split(lastmap->get_pg_num(pgid.pool()),
-        osdmap->get_pg_num(pgid.pool()), 0)) {
+  if (is_new_interval(
+	old_acting_primary,
+	new_acting_primary,
+	old_acting,
+	new_acting,
+	old_up_primary,
+	new_up_primary,
+	old_up,
+	new_up,
+	osdmap,
+	lastmap,
+	pool_id,
+	pgid)) {
     pg_interval_t& i = (*past_intervals)[same_interval_since];
     i.first = same_interval_since;
     i.last = osdmap->get_epoch() - 1;

src/osd/osd_types.h

@@ -1752,13 +1752,50 @@ struct pg_interval_t {
   void dump(Formatter *f) const;
   static void generate_test_instances(list<pg_interval_t*>& o);
 
+  /**
+   * Determines whether there is an interval change
+   */
+  static bool is_new_interval(
+    int old_acting_primary,
+    int new_acting_primary,
+    const vector<int> &old_acting,
+    const vector<int> &new_acting,
+    int old_up_primary,
+    int new_up_primary,
+    const vector<int> &old_up,
+    const vector<int> &new_up,
+    int old_min_size,
+    int new_min_size,
+    unsigned old_pg_num,
+    unsigned new_pg_num,
+    pg_t pgid
+    );
+
+  /**
+   * Determines whether there is an interval change
+   */
+  static bool is_new_interval(
+    int old_acting_primary,                     ///< [in] primary as of lastmap
+    int new_acting_primary,                     ///< [in] primary as of lastmap
+    const vector<int> &old_acting,              ///< [in] acting as of lastmap
+    const vector<int> &new_acting,              ///< [in] acting as of osdmap
+    int old_up_primary,                         ///< [in] up primary of lastmap
+    int new_up_primary,                         ///< [in] up primary of osdmap
+    const vector<int> &old_up,                  ///< [in] up as of lastmap
+    const vector<int> &new_up,                  ///< [in] up as of osdmap
+    ceph::shared_ptr<const OSDMap> osdmap,  ///< [in] current map
+    ceph::shared_ptr<const OSDMap> lastmap, ///< [in] last map
+    int64_t poolid,                             ///< [in] pool for pg
+    pg_t pgid                                   ///< [in] pgid for pg
+    );
+
   /**
    * Integrates a new map into *past_intervals, returns true
    * if an interval was closed out.
    */
   static bool check_new_interval(
     int old_acting_primary,                     ///< [in] primary as of lastmap
-    int new_acting_primary,                     ///< [in] primary as of lastmap
+    int new_acting_primary,                     ///< [in] primary as of osdmap
     const vector<int> &old_acting,              ///< [in] acting as of lastmap
     const vector<int> &new_acting,              ///< [in] acting as of osdmap
     int old_up_primary,                         ///< [in] up primary of lastmap

src/osdc/Objecter.cc

@@ -1483,9 +1483,29 @@ int Objecter::calc_target(op_target_t *t, bool any_change)
     if (ret == -ENOENT)
       return RECALC_OP_TARGET_POOL_DNE;
   }
-  int primary;
-  vector<int> acting;
-  osdmap->pg_to_acting_osds(pgid, &acting, &primary);
+
+  int min_size = pi->min_size;
+  unsigned pg_num = pi->get_pg_num();
+  int up_primary, acting_primary;
+  vector<int> up, acting;
+  osdmap->pg_to_up_acting_osds(pgid, &up, &up_primary,
+			       &acting, &acting_primary);
+  if (any_change && pg_interval_t::is_new_interval(
+          t->acting_primary,
+	  acting_primary,
+	  t->acting,
+	  acting,
+	  t->up_primary,
+	  up_primary,
+	  t->up,
+	  up,
+	  t->min_size,
+	  min_size,
+	  t->pg_num,
+	  pg_num,
+	  pi->raw_pg_to_pg(pgid))) {
+    force_resend = true;
+  }
 
   bool need_resend = false;
 
@@ -1497,15 +1517,20 @@ int Objecter::calc_target(op_target_t *t, bool any_change)
 
   if (t->pgid != pgid ||
       is_pg_changed(
-	t->primary, t->acting, primary, acting, t->used_replica || any_change) ||
+	t->acting_primary, t->acting, acting_primary, acting,
+	t->used_replica || any_change) ||
       force_resend) {
     t->pgid = pgid;
     t->acting = acting;
-    t->primary = primary;
-    ldout(cct, 10) << __func__ << " pgid " << pgid
-		   << " acting " << acting << dendl;
+    t->acting_primary = acting_primary;
+    t->up_primary = up_primary;
+    t->up = up;
+    t->min_size = min_size;
+    t->pg_num = pg_num;
+    ldout(cct, 10) << __func__ << " "
+		   << " pgid " << pgid << " acting " << acting << dendl;
     t->used_replica = false;
-    if (primary == -1) {
+    if (acting_primary == -1) {
       t->osd = -1;
     } else {
       int osd;
@@ -1541,7 +1566,7 @@ int Objecter::calc_target(op_target_t *t, bool any_change)
 	assert(best >= 0);
 	osd = acting[best];
       } else {
-	osd = primary;
+	osd = acting_primary;
       }
       t->osd = osd;
     }

src/osdc/Objecter.h

@@ -1072,8 +1072,12 @@ class Objecter : public md_config_obs_t {
     pg_t base_pgid;      ///< explciti pg target, if any
 
     pg_t pgid;           ///< last pg we mapped to
-    vector<int> acting;  ///< acting for last pg we mapped to
-    int primary;         ///< primary for last pg we mapped to
+    unsigned pg_num;     ///< last pg_num we mapped to
+    vector<int> up;      ///< set of up osds for last pg we mapped to
+    vector<int> acting;  ///< set of acting osds for last pg we mapped to
+    int up_primary;      ///< primary for last pg we mapped to based on the up set
+    int acting_primary;  ///< primary for last pg we mapped to based on the acting set
+    int min_size;        ///< the min size of the pool when were were last mapped
 
     bool used_replica;
     bool paused;
@@ -1085,7 +1089,10 @@ class Objecter : public md_config_obs_t {
 	base_oid(oid),
 	base_oloc(oloc),
 	precalc_pgid(false),
-	primary(-1),
+	pg_num(0),
+	up_primary(-1),
+	acting_primary(-1),
+	min_size(-1),
 	used_replica(false),
 	paused(false),
 	osd(-1)