rgw: assume obj write is a first write

if fails and object already exists then retry. This improves first obj
write performance on the expense of overwrites.

Fixes: http://tracker.ceph.com/issues/18622
Signed-off-by: Yehuda Sadeh <yehuda@redhat.com>

diff --git a/src/rgw/rgw_rados.cc b/src/rgw/rgw_rados.cc
index 7c9f6b881f3798f67867cef9c4e66c7a18faafee..ef1c95b7c1d816d8975bd8105c729ca83b17ba3f 100644 (file)
--- a/src/rgw/rgw_rados.cc
+++ b/src/rgw/rgw_rados.cc
@@ -6288,8 +6288,8 @@ int RGWRados::swift_versioning_restore(RGWObjectCtx& obj_ctx,
  * exclusive: create object exclusively
  * Returns: 0 on success, -ERR# otherwise.
  */
-int RGWRados::Object::Write::write_meta(uint64_t size, uint64_t accounted_size,
-                                        map<string, bufferlist>& attrs)
+int RGWRados::Object::Write::_do_write_meta(uint64_t size, uint64_t accounted_size,
+                                           map<string, bufferlist>& attrs, bool assume_noent)
 {
   rgw_bucket bucket;
   rgw_rados_ref ref;
@@ -6298,7 +6298,7 @@ int RGWRados::Object::Write::write_meta(uint64_t size, uint64_t accounted_size,
   ObjectWriteOperation op;
 
   RGWObjState *state;
-  int r = target->get_state(&state, false);
+  int r = target->get_state(&state, false, assume_noent);
   if (r < 0)
     return r;
 
@@ -6421,6 +6421,10 @@ int RGWRados::Object::Write::write_meta(uint64_t size, uint64_t accounted_size,
   if (r < 0) { /* we can expect to get -ECANCELED if object was replaced under,
                 or -ENOENT if was removed, or -EEXIST if it did not exist
                 before and now it does */
+    if (r == -EEXIST && assume_noent) {
+      target->invalidate_state();
+      return r;
+    }
     goto done_cancel;
   }
 
@@ -6486,7 +6490,10 @@ done_cancel:
    * should treat it as a success
    */
   if (meta.if_match == NULL && meta.if_nomatch == NULL) {
-    if (r == -ECANCELED || r == -ENOENT || r == -EEXIST) {
+    if (r == -ECANCELED || r == -ENOENT ||
+        (r == -EEXIST && !assume_noent)) /* if assume_noent, we want to send back error so that
+                                          * we'd be called again with assume_noent == false
+                                          */ {
       r = 0;
     }
   } else {
@@ -6516,6 +6523,23 @@ done_cancel:
   return r;
 }
 
+int RGWRados::Object::Write::write_meta(uint64_t size, uint64_t accounted_size,
+                                           map<string, bufferlist>& attrs)
+{
+  bool assume_noent = (meta.if_match == NULL && meta.if_nomatch == NULL);
+  int r;
+  if (assume_noent) {
+    r = _do_write_meta(size, accounted_size, attrs, assume_noent);
+    if (r == -EEXIST) {
+      assume_noent = false;
+    }
+  }
+  if (!assume_noent) {
+    r = _do_write_meta(size, accounted_size, attrs, assume_noent);
+  }
+  return r;
+}
+
 /** Write/overwrite a system object. */
 int RGWRados::put_system_obj_impl(rgw_obj& obj, uint64_t size, real_time *mtime,
               map<std::string, bufferlist>& attrs, int flags,
@@ -8503,7 +8527,7 @@ int RGWRados::get_system_obj_state(RGWObjectCtx *rctx, rgw_obj& obj, RGWObjState
   return ret;
 }
 
-int RGWRados::get_obj_state_impl(RGWObjectCtx *rctx, rgw_obj& obj, RGWObjState **state, bool follow_olh)
+int RGWRados::get_obj_state_impl(RGWObjectCtx *rctx, rgw_obj& obj, RGWObjState **state, bool follow_olh, bool assume_noent)
 {
   bool need_follow_olh = follow_olh && !obj.have_instance();
 
@@ -8519,7 +8543,11 @@ int RGWRados::get_obj_state_impl(RGWObjectCtx *rctx, rgw_obj& obj, RGWObjState *
 
   s->obj = obj;
 
-  int r = RGWRados::raw_obj_stat(obj, &s->size, &s->mtime, &s->epoch, &s->attrset, (s->prefetch_data ? &s->data : NULL), NULL);
+  int r = -ENOENT;
+
+  if (!assume_noent) {
+    r = RGWRados::raw_obj_stat(obj, &s->size, &s->mtime, &s->epoch, &s->attrset, (s->prefetch_data ? &s->data : NULL), NULL);
+  }
   if (r == -ENOENT) {
     s->exists = false;
     s->has_attrs = true;
@@ -8645,12 +8673,12 @@ int RGWRados::get_obj_state_impl(RGWObjectCtx *rctx, rgw_obj& obj, RGWObjState *
   return 0;
 }
 
-int RGWRados::get_obj_state(RGWObjectCtx *rctx, rgw_obj& obj, RGWObjState **state, bool follow_olh)
+int RGWRados::get_obj_state(RGWObjectCtx *rctx, rgw_obj& obj, RGWObjState **state, bool follow_olh, bool assume_noent)
 {
   int ret;
 
   do {
-    ret = get_obj_state_impl(rctx, obj, state, follow_olh);
+    ret = get_obj_state_impl(rctx, obj, state, follow_olh, assume_noent);
   } while (ret == -EAGAIN);
 
   return ret;
@@ -8806,9 +8834,9 @@ int RGWRados::append_atomic_test(RGWObjectCtx *rctx, rgw_obj& obj,
   return 0;
 }
 
-int RGWRados::Object::get_state(RGWObjState **pstate, bool follow_olh)
+int RGWRados::Object::get_state(RGWObjState **pstate, bool follow_olh, bool assume_noent)
 {
-  return store->get_obj_state(&ctx, obj, pstate, follow_olh);
+  return store->get_obj_state(&ctx, obj, pstate, follow_olh, assume_noent);
 }
 
 void RGWRados::Object::invalidate_state()

diff --git a/src/rgw/rgw_rados.h b/src/rgw/rgw_rados.h
index 8cea3ad63e0c7d1f385b40a8cb9b5a8b190fbf5d..8192939075156e4c7dba80a867fe5735f3ed8be4 100644 (file)
--- a/src/rgw/rgw_rados.h
+++ b/src/rgw/rgw_rados.h
@@ -1966,7 +1966,7 @@ class RGWRados
   int get_olh_target_state(RGWObjectCtx& rctx, rgw_obj& obj, RGWObjState *olh_state,
                            RGWObjState **target_state);
   int get_system_obj_state_impl(RGWObjectCtx *rctx, rgw_obj& obj, RGWObjState **state, RGWObjVersionTracker *objv_tracker);
-  int get_obj_state_impl(RGWObjectCtx *rctx, rgw_obj& obj, RGWObjState **state, bool follow_olh);
+  int get_obj_state_impl(RGWObjectCtx *rctx, rgw_obj& obj, RGWObjState **state, bool follow_olh, bool assume_noent = false);
   int append_atomic_test(RGWObjectCtx *rctx, rgw_obj& obj,
                          librados::ObjectOperation& op, RGWObjState **state);
 
@@ -2354,7 +2354,7 @@ public:
     bool bs_initialized;
 
   protected:
-    int get_state(RGWObjState **pstate, bool follow_olh);
+    int get_state(RGWObjState **pstate, bool follow_olh, bool assume_noent = false);
     void invalidate_state();
 
     int prepare_atomic_modification(librados::ObjectWriteOperation& op, bool reset_obj, const string *ptag,
@@ -2459,6 +2459,9 @@ public:
 
       explicit Write(RGWRados::Object *_target) : target(_target) {}
 
+      int _do_write_meta(uint64_t size, uint64_t accounted_size,
+                     map<std::string, bufferlist>& attrs,
+                     bool assume_noent);
       int write_meta(uint64_t size, uint64_t accounted_size,
                      map<std::string, bufferlist>& attrs);
       int write_data(const char *data, uint64_t ofs, uint64_t len, bool exclusive);
@@ -2859,7 +2862,7 @@ public:
                         map<string, bufferlist>* rmattrs);
 
   int get_system_obj_state(RGWObjectCtx *rctx, rgw_obj& obj, RGWObjState **state, RGWObjVersionTracker *objv_tracker);
-  int get_obj_state(RGWObjectCtx *rctx, rgw_obj& obj, RGWObjState **state, bool follow_olh);
+  int get_obj_state(RGWObjectCtx *rctx, rgw_obj& obj, RGWObjState **state, bool follow_olh, bool assume_noent = false);
   int get_obj_state(RGWObjectCtx *rctx, rgw_obj& obj, RGWObjState **state) {
     return get_obj_state(rctx, obj, state, true);
   }