osd: add and utilize OSD_FIXED_COLLECTION_LIST feature

If all osds from upacting set have this feature set
the backend can use the new "fixed" collection_list method,
otherwise it fallbacks to the legacy method.

Signed-off-by: Mykola Golub <mgolub@suse.com>
(cherry picked from commit 8f9d335bc7cccb221ca7316ee0c8f22198d0f9ef)

Conflicts:
	src/include/ceph_features.h: trivial

diff --git a/src/include/ceph_features.h b/src/include/ceph_features.h
index 714948ae06db27fb51804b380deeb1db567409c5..9f3c012f7be8454e24a1253ee8b37fdf1b267194 100644 (file)
--- a/src/include/ceph_features.h
+++ b/src/include/ceph_features.h
@@ -120,7 +120,7 @@ DEFINE_CEPH_FEATURE(21, 2, RADOS_BACKOFF)    // overlap
 DEFINE_CEPH_FEATURE(21, 2, OSDMAP_PG_UPMAP)  // overlap
 DEFINE_CEPH_FEATURE(21, 2, CRUSH_CHOOSE_ARGS) // overlap
 DEFINE_CEPH_FEATURE_RETIRED(22, 1, BACKFILL_RESERVATION, JEWEL, LUMINOUS)
-
+DEFINE_CEPH_FEATURE(22, 2, OSD_FIXED_COLLECTION_LIST)
 DEFINE_CEPH_FEATURE(23, 1, MSG_AUTH)         // 3.19 req (unless nocephx_require_signatures)
 DEFINE_CEPH_FEATURE_RETIRED(24, 1, RECOVERY_RESERVATION, JEWEL, LUMINOUS)
 
@@ -246,6 +246,7 @@ DEFINE_CEPH_FEATURE_DEPRECATED(63, 1, RESERVED_BROKEN, LUMINOUS) // client-facin
         CEPH_FEATURE_OSD_PGLOG_HARDLIMIT | \
         CEPH_FEATUREMASK_SERVER_OCTOPUS | \
         CEPH_FEATUREMASK_OSD_REPOP_MLCOD | \
+        CEPH_FEATURE_OSD_FIXED_COLLECTION_LIST | \
         0ULL)
 
 #define CEPH_FEATURES_SUPPORTED_DEFAULT  CEPH_FEATURES_ALL

diff --git a/src/osd/PGBackend.cc b/src/osd/PGBackend.cc
index 06bb0f423f37f16efede853f3bf0804150775bae..be93941196734c73a10a99aa78061dd5763f361a 100644 (file)
--- a/src/osd/PGBackend.cc
+++ b/src/osd/PGBackend.cc
@@ -350,13 +350,24 @@ int PGBackend::objects_list_partial(
 
   while (!_next.is_max() && ls->size() < (unsigned)min) {
     vector<ghobject_t> objects;
-    r = store->collection_list(
-      ch,
-      _next,
-      ghobject_t::get_max(),
-      max - ls->size(),
-      &objects,
-      &_next);
+    if (HAVE_FEATURE(parent->min_upacting_features(),
+                     OSD_FIXED_COLLECTION_LIST)) {
+      r = store->collection_list(
+        ch,
+        _next,
+        ghobject_t::get_max(),
+        max - ls->size(),
+        &objects,
+        &_next);
+    } else {
+      r = store->collection_list_legacy(
+        ch,
+        _next,
+        ghobject_t::get_max(),
+        max - ls->size(),
+        &objects,
+        &_next);
+    }
     if (r != 0) {
       derr << __func__ << " list collection " << ch << " got: " << cpp_strerror(r) << dendl;
       break;
@@ -385,13 +396,25 @@ int PGBackend::objects_list_range(
 {
   ceph_assert(ls);
   vector<ghobject_t> objects;
-  int r = store->collection_list(
-    ch,
-    ghobject_t(start, ghobject_t::NO_GEN, get_parent()->whoami_shard().shard),
-    ghobject_t(end, ghobject_t::NO_GEN, get_parent()->whoami_shard().shard),
-    INT_MAX,
-    &objects,
-    NULL);
+  int r;
+  if (HAVE_FEATURE(parent->min_upacting_features(),
+                   OSD_FIXED_COLLECTION_LIST)) {
+    r = store->collection_list(
+      ch,
+      ghobject_t(start, ghobject_t::NO_GEN, get_parent()->whoami_shard().shard),
+      ghobject_t(end, ghobject_t::NO_GEN, get_parent()->whoami_shard().shard),
+      INT_MAX,
+      &objects,
+      NULL);
+  } else {
+    r = store->collection_list_legacy(
+      ch,
+      ghobject_t(start, ghobject_t::NO_GEN, get_parent()->whoami_shard().shard),
+      ghobject_t(end, ghobject_t::NO_GEN, get_parent()->whoami_shard().shard),
+      INT_MAX,
+      &objects,
+      NULL);
+  }
   ls->reserve(objects.size());
   for (vector<ghobject_t>::iterator i = objects.begin();
        i != objects.end();

diff --git a/src/osd/PGBackend.h b/src/osd/PGBackend.h
index 4a952c4ed8bd9ca4173a6a95623f070eacc5ad94..b4c598feb2208b79627777b9a931e82709ead4dc 100644 (file)
--- a/src/osd/PGBackend.h
+++ b/src/osd/PGBackend.h
@@ -273,6 +273,7 @@ typedef std::shared_ptr<const OSDMap> OSDMapRef;
      virtual spg_t primary_spg_t() const = 0;
      virtual pg_shard_t primary_shard() const = 0;
      virtual uint64_t min_peer_features() const = 0;
+     virtual uint64_t min_upacting_features() const = 0;
      virtual hobject_t get_temp_recovery_object(const hobject_t& target,
                                                eversion_t version) = 0;
 

diff --git a/src/osd/PrimaryLogPG.h b/src/osd/PrimaryLogPG.h
index afa54ab503adc21de938432b850fb7127fea3a72..9d1c12366a59d44a57aa4c09aba3c52b96f146a4 100644 (file)
--- a/src/osd/PrimaryLogPG.h
+++ b/src/osd/PrimaryLogPG.h
@@ -538,6 +538,9 @@ public:
   uint64_t min_peer_features() const override {
     return recovery_state.get_min_peer_features();
   }
+  uint64_t min_upacting_features() const override {
+    return recovery_state.get_min_upacting_features();
+  }
   void send_message_osd_cluster(
     int peer, Message *m, epoch_t from_epoch) override {
     osd->send_message_osd_cluster(peer, m, from_epoch);