From c8070d4e192b3f3a35d7f6a8244891f6c71a6bb5 Mon Sep 17 00:00:00 2001
From: Xinze Chi <xinze@xsky.com>
Date: Mon, 9 Nov 2015 19:58:03 +0800
Subject: [PATCH] MOSDRepOpReply: Simple Messenger optimization

the origin idea is from https://github.com/ceph/ceph/pull/5211.

Signed-off-by: Xinze Chi <xinze@xsky.com>
---
 src/messages/MOSDRepOpReply.h | 69 +++++++++++++++++++++++++++--------
 src/messages/MOSDSubOpReply.h |  3 ++
 src/osd/ReplicatedBackend.cc  |  1 +
 3 files changed, 57 insertions(+), 16 deletions(-)

diff --git a/src/messages/MOSDRepOpReply.h b/src/messages/MOSDRepOpReply.h
index f0faa4c022994b..81bd47b73795ab 100644
--- a/src/messages/MOSDRepOpReply.h
+++ b/src/messages/MOSDRepOpReply.h
@@ -29,7 +29,7 @@
  */
 
 class MOSDRepOpReply : public Message {
-  static const int HEAD_VERSION = 1;
+  static const int HEAD_VERSION = 2;
   static const int COMPAT_VERSION = 1;
 public:
   epoch_t map_epoch;
@@ -46,20 +46,50 @@ class MOSDRepOpReply : public Message {
   // piggybacked osd state
   eversion_t last_complete_ondisk;
 
+  bufferlist::iterator p;
+  // Decoding flags. Decoding is only needed for messages catched by pipe reader.
+  bool partial_decode_needed;
+  bool final_decode_needed;
 
   virtual void decode_payload() {
-    bufferlist::iterator p = payload.begin();
-    ::decode(map_epoch, p);
-    ::decode(reqid, p);
-    ::decode(pgid, p);
+    p = payload.begin();
+    if (header.version < 2) {
+      ::decode(map_epoch, p);
+      ::decode(reqid, p);
+      ::decode(pgid, p);
+
+      ::decode(ack_type, p);
+      ::decode(result, p);
+      ::decode(last_complete_ondisk, p);
+
+      ::decode(from, p);
+    } else {
+      // new, v2 decode, splitted to partial and final
+      ::decode(map_epoch, p);
+      ::decode(reqid, p);
+      ::decode(pgid, p);
+    }
+    partial_decode_needed = false;
+  }
 
+  void finish_decode() {
+    assert(!partial_decode_needed); // partial decoding required
+    if (!final_decode_needed)
+      return; // Message is already final decoded
+    assert(header.version >= 2);
     ::decode(ack_type, p);
     ::decode(result, p);
     ::decode(last_complete_ondisk, p);
 
     ::decode(from, p);
+    final_decode_needed = false;
   }
   virtual void encode_payload(uint64_t features) {
+    if ((features & CEPH_FEATURE_NEW_OSDREPOP_ENCODING) == 0) {
+      header.version = 1;
+    } else {
+      header.version = HEAD_VERSION;
+    }
     ::encode(map_epoch, payload);
     ::encode(reqid, payload);
     ::encode(pgid, payload);
@@ -91,12 +121,16 @@ class MOSDRepOpReply : public Message {
     from(from),
     pgid(req->pgid.pgid, req->from.shard),
     ack_type(at),
-    result(result_) {
+    result(result_),
+    partial_decode_needed(false),
+    final_decode_needed(false) {
     set_tid(req->get_tid());
   }
   MOSDRepOpReply() 
     : Message(MSG_OSD_REPOPREPLY), map_epoch(0),  
-      ack_type(0), result(0) {}
+      ack_type(0), result(0),
+      partial_decode_needed(true),
+      final_decode_needed(true) {}
 private:
   ~MOSDRepOpReply() {}
 
@@ -104,15 +138,18 @@ class MOSDRepOpReply : public Message {
   const char *get_type_name() const { return "osd_repop_reply"; }
 
   void print(ostream& out) const {
-    out << "osd_repop_reply(" << reqid
-	<< " " << pgid;
-    if (ack_type & CEPH_OSD_FLAG_ONDISK)
-      out << " ondisk";
-    if (ack_type & CEPH_OSD_FLAG_ONNVRAM)
-      out << " onnvram";
-    if (ack_type & CEPH_OSD_FLAG_ACK)
-      out << " ack";
-    out << ", result = " << result;
+    out << "osd_repop_reply(";
+    if (!partial_decode_needed) {
+      out << reqid
+          << " " << pgid;
+      if (ack_type & CEPH_OSD_FLAG_ONDISK)
+        out << " ondisk";
+      if (ack_type & CEPH_OSD_FLAG_ONNVRAM)
+        out << " onnvram";
+      if (ack_type & CEPH_OSD_FLAG_ACK)
+        out << " ack";
+      out << ", result = " << result;
+    }
     out << ")";
   }
 
diff --git a/src/messages/MOSDSubOpReply.h b/src/messages/MOSDSubOpReply.h
index a084246e7f94e3..81d1b2836de376 100644
--- a/src/messages/MOSDSubOpReply.h
+++ b/src/messages/MOSDSubOpReply.h
@@ -85,6 +85,9 @@ class MOSDSubOpReply : public Message {
       pgid.shard = shard_id_t::NO_SHARD;
     }
   }
+
+  void finish_decode() { }
+
   virtual void encode_payload(uint64_t features) {
     ::encode(map_epoch, payload);
     ::encode(reqid, payload);
diff --git a/src/osd/ReplicatedBackend.cc b/src/osd/ReplicatedBackend.cc
index 63c7d93f1c5773..daa96e29172459 100644
--- a/src/osd/ReplicatedBackend.cc
+++ b/src/osd/ReplicatedBackend.cc
@@ -666,6 +666,7 @@ template<typename T, int MSGTYPE>
 void ReplicatedBackend::sub_op_modify_reply(OpRequestRef op)
 {
   T *r = static_cast<T *>(op->get_req());
+  r->finish_decode();
   assert(r->get_header().type == MSGTYPE);
   assert(MSGTYPE == MSG_OSD_SUBOPREPLY || MSGTYPE == MSG_OSD_REPOPREPLY);