rgw: add 'inline_data' zone placement info option

Adds a new RGW zone placement info option to control whether
an object's first data chunk is placed in the head object.
This allows admins to make a tradeoff between optimizing for
PUT/GET performance vs. DELETE performance for some cluster
configurations.

Fixes: https://tracker.ceph.com/issues/57965
Signed-off-by: Cory Snyder <csnyder@iland.com>
(cherry picked from commit 9052ca420f003af4b89b568c50bd5083333d3010)

Conflicts:
	src/rgw/rgw_zone.cc
	src/rgw/rgw_zone_types.h

Cherry-pick notes:
- Conflicts due to rgw_zone_types.h moving to rgw_zone.h

diff --git a/doc/man/8/radosgw-admin.rst b/doc/man/8/radosgw-admin.rst
index 57d7fcfc902384dbbbdcf0355113de0a01a2b5d6..e099bba074df0753ca62ccfbb60614aa37b320e9 100644 (file)
--- a/doc/man/8/radosgw-admin.rst
+++ b/doc/man/8/radosgw-admin.rst
@@ -743,6 +743,10 @@ Options
 
    The placement target index type (normal, indexless, or #id).
 
+.. option:: --placement-inline-data=<true>
+
+   Whether the placement target is configured to store a data chunk inline in head objects.
+
 .. option:: --tier-type=<type>
 
    The zone tier type.

diff --git a/doc/radosgw/placement.rst b/doc/radosgw/placement.rst
index d799244da4e2fb06ec0017313016f8b22eb8f4c8..ae4316524d94a062d91f711f3c3f7be66dfed868 100644 (file)
--- a/doc/radosgw/placement.rst
+++ b/doc/radosgw/placement.rst
@@ -86,7 +86,8 @@ The zone placement configuration can be queried with:
                       }
                   },
                   "data_extra_pool": "default.rgw.buckets.non-ec",
-                  "index_type": 0
+                  "index_type": 0,
+                  "inline_data": true
               }
           }
       ],
@@ -132,7 +133,6 @@ Then provide the zone placement info for that target:
           the BlueStore DB is located on faster storage than bucket data since it eliminates the need to access
           slower devices synchronously while processing the client request. In that case, data associated with the deleted
           objects is removed asynchronously in the background by garbage collection.                                          
-
 .. _adding_a_storage_class:
 
 Adding a Storage Class

diff --git a/src/rgw/rgw_admin.cc b/src/rgw/rgw_admin.cc
index 097ad1c2f90b3167145dfe2e744a1171ba75bd8f..fd8188e2ddabb6719e447194f19ce687e72cbcf3 100644 (file)
--- a/src/rgw/rgw_admin.cc
+++ b/src/rgw/rgw_admin.cc
@@ -369,6 +369,9 @@ void usage()
   cout << "   --data-extra-pool=<pool>  placement target data extra (non-ec) pool\n";
   cout << "   --placement-index-type=<type>\n";
   cout << "                             placement target index type (normal, indexless, or #id)\n";
+  cout << "   --placement-inline-data=<true>\n";
+  cout << "                             set whether the placement target is configured to store a data\n";
+  cout << "                             chunk inline in head objects\n";
   cout << "   --compression=<type>      placement target compression type (plugin name or empty/none)\n";
   cout << "   --tier-type=<type>        zone tier type\n";
   cout << "   --tier-config=<k>=<v>[,...]\n";
@@ -3478,6 +3481,8 @@ int main(int argc, const char **argv)
   list<string> tags;
   list<string> tags_add;
   list<string> tags_rm;
+  int placement_inline_data = true;
+  bool placement_inline_data_specified = false;
 
   int64_t max_objects = -1;
   int64_t max_size = -1;
@@ -3515,6 +3520,7 @@ int main(int argc, const char **argv)
   int num_shards = 0;
   bool num_shards_specified = false;
   std::optional<int> bucket_index_max_shards;
+
   int max_concurrent_ios = 32;
   uint64_t orphan_stale_secs = (24 * 3600);
   int detail = false;
@@ -3843,6 +3849,9 @@ int main(int argc, const char **argv)
      // do nothing
     } else if (ceph_argparse_binary_flag(args, i, &inconsistent_index, NULL, "--inconsistent-index", (char*)NULL)) {
      // do nothing
+    } else if (ceph_argparse_binary_flag(args, i, &placement_inline_data, NULL, "--placement-inline-data", (char*)NULL)) {
+      placement_inline_data_specified = true;
+     // do nothing
     } else if (ceph_argparse_witharg(args, i, &val, "--caps", (char*)NULL)) {
       caps = val;
     } else if (ceph_argparse_witharg(args, i, &val, "--infile", (char*)NULL)) {
@@ -5977,6 +5986,9 @@ int main(int argc, const char **argv)
           if (index_type_specified) {
            info.index_type = placement_index_type;
           }
+          if (placement_inline_data_specified) {
+            info.inline_data = placement_inline_data;
+          }
 
           ret = check_pool_support_omap(info.get_data_extra_pool());
           if (ret < 0) {

diff --git a/src/rgw/rgw_putobj_processor.cc b/src/rgw/rgw_putobj_processor.cc
index 5c21c76a65b8d550b8ec5cd9b096c61dd14c12b1..f0045d6f1686a67ab57506c88847aedbe686a7db 100644 (file)
--- a/src/rgw/rgw_putobj_processor.cc
+++ b/src/rgw/rgw_putobj_processor.cc
@@ -18,6 +18,7 @@
 #include "rgw_multi.h"
 #include "rgw_compression.h"
 #include "services/svc_sys_obj.h"
+#include "services/svc_zone.h"
 #include "rgw_sal_rados.h"
 
 #define dout_subsys ceph_subsys_rgw
@@ -246,7 +247,12 @@ int AtomicObjectProcessor::prepare(optional_yield y)
   }
 
   if (same_pool) {
-    head_max_size = max_head_chunk_size;
+    RGWZonePlacementInfo placement_info;
+    if (!store->svc()->zone->get_zone_params().get_placement(head_obj->get_bucket()->get_placement_rule().name, &placement_info) || placement_info.inline_data) {
+      head_max_size = max_head_chunk_size;
+    } else {
+      head_max_size = 0;
+    }
     chunk_size = max_head_chunk_size;
   }
 

diff --git a/src/rgw/rgw_zone.cc b/src/rgw/rgw_zone.cc
index c8e8d09b8879a66adb8548e56b5138fa8627db8c..bbeea7efb84e1e1ad39c1d361e412e697de48466 100644 (file)
--- a/src/rgw/rgw_zone.cc
+++ b/src/rgw/rgw_zone.cc
@@ -2607,6 +2607,7 @@ void RGWZonePlacementInfo::dump(Formatter *f) const
   encode_json("storage_classes", storage_classes, f);
   encode_json("data_extra_pool", data_extra_pool, f);
   encode_json("index_type", (uint32_t)index_type, f);
+  encode_json("inline_data", inline_data, f);
 
   /* no real need for backward compatibility of compression_type and data_pool in here,
    * rather not clutter the output */
@@ -2619,6 +2620,7 @@ void RGWZonePlacementInfo::decode_json(JSONObj *obj)
   JSONDecoder::decode_json("data_extra_pool", data_extra_pool, obj);
   uint32_t it;
   JSONDecoder::decode_json("index_type", it, obj);
+  JSONDecoder::decode_json("inline_data", inline_data, obj);
   index_type = (rgw::BucketIndexType)it;
 
   /* backward compatibility, these are now defined in storage_classes */

diff --git a/src/rgw/rgw_zone.h b/src/rgw/rgw_zone.h
index 9e89cbf150a72183210415177d332757ba158b4e..f8ddcec5bda51f8a1285798467a0ce93f87fedfc 100644 (file)
--- a/src/rgw/rgw_zone.h
+++ b/src/rgw/rgw_zone.h
@@ -275,11 +275,12 @@ struct RGWZonePlacementInfo {
   rgw_pool data_extra_pool; /* if not set we should use data_pool */
   RGWZoneStorageClasses storage_classes;
   rgw::BucketIndexType index_type;
+  bool inline_data;
 
-  RGWZonePlacementInfo() : index_type(rgw::BucketIndexType::Normal) {}
+  RGWZonePlacementInfo() : index_type(rgw::BucketIndexType::Normal), inline_data(true) {}
 
   void encode(bufferlist& bl) const {
-    ENCODE_START(7, 1, bl);
+    ENCODE_START(8, 1, bl);
     encode(index_pool.to_str(), bl);
     rgw_pool standard_data_pool = get_data_pool(RGW_STORAGE_CLASS_STANDARD);
     encode(standard_data_pool.to_str(), bl);
@@ -288,11 +289,12 @@ struct RGWZonePlacementInfo {
     std::string standard_compression_type = get_compression_type(RGW_STORAGE_CLASS_STANDARD);
     encode(standard_compression_type, bl);
     encode(storage_classes, bl);
+    encode(inline_data, bl);
     ENCODE_FINISH(bl);
   }
 
   void decode(bufferlist::const_iterator& bl) {
-    DECODE_START(7, bl);
+    DECODE_START(8, bl);
     std::string index_pool_str;
     std::string data_pool_str;
     decode(index_pool_str, bl);
@@ -319,6 +321,9 @@ struct RGWZonePlacementInfo {
       storage_classes.set_storage_class(RGW_STORAGE_CLASS_STANDARD, &standard_data_pool,
                                         (!standard_compression_type.empty() ? &standard_compression_type : nullptr));
     }
+    if (struct_v >= 8) {
+      decode(inline_data, bl);
+    }
     DECODE_FINISH(bl);
   }
   const rgw_pool& get_data_extra_pool() const {

diff --git a/src/test/cli/radosgw-admin/help.t b/src/test/cli/radosgw-admin/help.t
index 6784b966ca4451939bc1f2f3dacf2182cda8218e..f6ba9650fddd6e5fbb478249c4550dd61f262e85 100644 (file)
--- a/src/test/cli/radosgw-admin/help.t
+++ b/src/test/cli/radosgw-admin/help.t
@@ -261,6 +261,9 @@
      --data-extra-pool=<pool>  placement target data extra (non-ec) pool
      --placement-index-type=<type>
                                placement target index type (normal, indexless, or #id)
+     --placement-inline-data=<true>
+                               set whether the placement target is configured to store a data
+                               chunk inline in head objects
      --compression=<type>      placement target compression type (plugin name or empty/none)
      --tier-type=<type>        zone tier type
      --tier-config=<k>=<v>[,...]