From 2b3761d7a247eaba12bbdb8e0fee4bd9cfc89041 Mon Sep 17 00:00:00 2001
From: "Yan, Zheng" <zyan@redhat.com>
Date: Thu, 27 Sep 2018 13:10:55 +0800
Subject: [PATCH] osdc: reduce ObjectCacher's memory fragments

Fixes: http://tracker.ceph.com/issues/36192
Signed-off-by: "Yan, Zheng" <zyan@redhat.com>
(cherry picked from commit 5d4f82117ed20f7a40259c3f5471e7a96a4c456f)

Conflicts:
	src/osdc/ObjectCacher.cc: trivial
---
 src/common/buffer.cc     | 26 ++++++++++++++++++++++++++
 src/include/buffer.h     |  1 +
 src/osdc/ObjectCacher.cc | 29 +++++++++++++++++++++++++----
 src/osdc/ObjectCacher.h  |  1 +
 4 files changed, 53 insertions(+), 4 deletions(-)

diff --git a/src/common/buffer.cc b/src/common/buffer.cc
index 1a8e9123b6c95..6449b4d04e669 100644
--- a/src/common/buffer.cc
+++ b/src/common/buffer.cc
@@ -1719,6 +1719,32 @@ public:
     }
   }
 
+  uint64_t buffer::list::get_wasted_space() const
+  {
+    if (_buffers.size() == 1)
+      return _buffers.back().wasted();
+
+    std::vector<const raw*> raw_vec;
+    raw_vec.reserve(_buffers.size());
+    for (const auto& p : _buffers)
+      raw_vec.push_back(p.get_raw());
+    std::sort(raw_vec.begin(), raw_vec.end());
+
+    uint64_t total = 0;
+    const raw *last = nullptr;
+    for (const auto r : raw_vec) {
+      if (r == last)
+	continue;
+      last = r;
+      total += r->len;
+    }
+    // If multiple buffers are sharing the same raw buffer and they overlap
+    // with each other, the wasted space will be underestimated.
+    if (total <= length())
+      return 0;
+    return total - length();
+  }
+
   void buffer::list::rebuild()
   {
     if (_len == 0) {
diff --git a/src/include/buffer.h b/src/include/buffer.h
index e110b74b2c0c0..473e69ff17b83 100644
--- a/src/include/buffer.h
+++ b/src/include/buffer.h
@@ -700,6 +700,7 @@ namespace buffer CEPH_BUFFER_API {
       return *this;
     }
 
+    uint64_t get_wasted_space() const;
     unsigned get_num_buffers() const { return _buffers.size(); }
     const ptr& front() const { return _buffers.front(); }
     const ptr& back() const { return _buffers.back(); }
diff --git a/src/osdc/ObjectCacher.cc b/src/osdc/ObjectCacher.cc
index 249f24636f680..cb1b0ca0cbfc9 100644
--- a/src/osdc/ObjectCacher.cc
+++ b/src/osdc/ObjectCacher.cc
@@ -12,7 +12,7 @@
 #include "include/assert.h"
 
 #define MAX_FLUSH_UNDER_LOCK 20  ///< max bh's we start writeback on
-#define BUFFER_MEMORY_WEIGHT 12   // memory usage of BufferHead, count in (1<<n)
+#define BUFFER_MEMORY_WEIGHT CEPH_PAGE_SHIFT  // memory usage of BufferHead, count in (1<<n)
 
 using std::chrono::seconds;
 				 /// while holding the lock
@@ -230,6 +230,20 @@ void ObjectCacher::Object::try_merge_bh(BufferHead *bh)
   ++p;
   if (p != data.end() && can_merge_bh(bh, p->second))
     merge_left(bh, p->second);
+
+  maybe_rebuild_buffer(bh);
+}
+
+void ObjectCacher::Object::maybe_rebuild_buffer(BufferHead *bh)
+{
+  auto& bl = bh->bl;
+  if (bl.get_num_buffers() <= 1)
+    return;
+
+  auto wasted = bl.get_wasted_space();
+  if (wasted * 2 > bl.length() &&
+      wasted > (1U << BUFFER_MEMORY_WEIGHT))
+    bl.rebuild();
 }
 
 /*
@@ -454,15 +468,18 @@ ObjectCacher::BufferHead *ObjectCacher::Object::map_write(ObjectExtent &ex,
         if (cur + max >= bh->end()) {
           // we want right bit (one splice)
           final = split(bh, cur);   // just split it, take right half.
+          maybe_rebuild_buffer(bh);
           replace_journal_tid(final, tid);
           ++p;
           assert(p->second == final);
         } else {
           // we want middle bit (two splices)
           final = split(bh, cur);
+          maybe_rebuild_buffer(bh);
           ++p;
           assert(p->second == final);
-          split(final, cur+max);
+          auto right = split(final, cur+max);
+          maybe_rebuild_buffer(right);
           replace_journal_tid(final, tid);
         }
       } else {
@@ -471,7 +488,8 @@ ObjectCacher::BufferHead *ObjectCacher::Object::map_write(ObjectExtent &ex,
           // whole bufferhead, piece of cake.
         } else {
           // we want left bit (one splice)
-          split(bh, cur + max);        // just split
+          auto right = split(bh, cur + max);        // just split
+          maybe_rebuild_buffer(right);
         }
         if (final) {
           oc->mark_dirty(bh);
@@ -549,6 +567,7 @@ void ObjectCacher::Object::truncate(loff_t s)
     // split bh at truncation point?
     if (bh->start() < s) {
       split(bh, s);
+      maybe_rebuild_buffer(bh);
       continue;
     }
 
@@ -586,13 +605,15 @@ void ObjectCacher::Object::discard(loff_t off, loff_t len,
     // split bh at truncation point?
     if (bh->start() < off) {
       split(bh, off);
+      maybe_rebuild_buffer(bh);
       ++p;
       continue;
     }
 
     assert(bh->start() >= off);
     if (bh->end() > off + len) {
-      split(bh, off + len);
+      auto right = split(bh, off + len);
+      maybe_rebuild_buffer(right);
     }
 
     ++p;
diff --git a/src/osdc/ObjectCacher.h b/src/osdc/ObjectCacher.h
index 60f049ef55d5b..81abb214c0737 100644
--- a/src/osdc/ObjectCacher.h
+++ b/src/osdc/ObjectCacher.h
@@ -343,6 +343,7 @@ class ObjectCacher {
     void merge_left(BufferHead *left, BufferHead *right);
     bool can_merge_bh(BufferHead *left, BufferHead *right);
     void try_merge_bh(BufferHead *bh);
+    void maybe_rebuild_buffer(BufferHead *bh);
 
     bool is_cached(loff_t off, loff_t len) const;
     bool include_all_cached_data(loff_t off, loff_t len);
-- 
2.47.3