From e86ad68753407b841f54644d120242dcea1cd0e4 Mon Sep 17 00:00:00 2001
From: Sage Weil <sage@redhat.com>
Date: Fri, 23 Sep 2016 16:40:36 -0400
Subject: [PATCH 1/2] os/bluestore: prevent extent merging across shard
 boundaries

We cannot have a single extent span a shard boundary.  If we reach
a shard boundary, stop merging extents.

Note that it might be better to move the shard boundary, but it is
awkward to force that to happen at this point.

Signed-off-by: Sage Weil <sage@redhat.com>
---
 src/os/bluestore/BlueStore.cc | 28 +++++++++++++++++++++++++++-
 1 file changed, 27 insertions(+), 1 deletion(-)

diff --git a/src/os/bluestore/BlueStore.cc b/src/os/bluestore/BlueStore.cc
index 069f8a0ce100f..2d25969cc203b 100644
--- a/src/os/bluestore/BlueStore.cc
+++ b/src/os/bluestore/BlueStore.cc
@@ -1888,6 +1888,20 @@ int BlueStore::ExtentMap::compress_extent_map(uint64_t offset, uint64_t length)
   if (p != extent_map.begin()) {
     --p;  // start to the left of offset
   }
+
+  // identify the *next* shard
+  auto pshard = shards.begin();
+  while (pshard != shards.end() &&
+	 p->logical_offset >= pshard->offset) {
+    ++pshard;
+  }
+  uint64_t shard_end;
+  if (pshard != shards.end()) {
+    shard_end = pshard->offset;
+  } else {
+    shard_end = OBJECT_MAX_SIZE;
+  }
+
   auto n = p;
   for (++n; n != extent_map.end(); p = n++) {
     if (n->logical_offset > offset + length) {
@@ -1896,7 +1910,11 @@ int BlueStore::ExtentMap::compress_extent_map(uint64_t offset, uint64_t length)
     while (n != extent_map.end() &&
 	   p->logical_offset + p->length == n->logical_offset &&
 	   p->blob == n->blob &&
-	   p->blob_offset + p->length == n->blob_offset) {
+	   p->blob_offset + p->length == n->blob_offset &&
+	   n->logical_offset < shard_end) {
+      dout(20) << __func__ << " 0x" << std::hex << offset << "~" << length
+	       << " next shard 0x" << shard_end << std::dec
+	       << " merging " << *p << " and " << *n << dendl;
       p->length += n->length;
       rm(n++);
       ++removed;
@@ -1904,6 +1922,14 @@ int BlueStore::ExtentMap::compress_extent_map(uint64_t offset, uint64_t length)
     if (n == extent_map.end()) {
       break;
     }
+    if (n->logical_offset >= shard_end) {
+      ++pshard;
+      if (pshard != shards.end()) {
+	shard_end = pshard->offset;
+      } else {
+	shard_end = OBJECT_MAX_SIZE;
+      }
+    }
   }
   return removed;
 }

From 4ece5b70d2de4a771b090eb20682e532fb59c09d Mon Sep 17 00:00:00 2001
From: Sage Weil <sage@redhat.com>
Date: Fri, 23 Sep 2016 16:59:35 -0400
Subject: [PATCH 2/2] ceph_test_objectstore: add a test

Signed-off-by: Sage Weil <sage@redhat.com>
---
 src/test/objectstore/store_test.cc | 49 ++++++++++++++++++++++++++++++
 1 file changed, 49 insertions(+)

diff --git a/src/test/objectstore/store_test.cc b/src/test/objectstore/store_test.cc
index 54ece21aefb37..d906ecb1c043b 100644
--- a/src/test/objectstore/store_test.cc
+++ b/src/test/objectstore/store_test.cc
@@ -3951,6 +3951,55 @@ TEST_P(StoreTest, SyntheticMatrixSharding) {
   do_matrix(m, store);
 }
 
+TEST_P(StoreTest, ZipperPatternSharded) {
+  if(string(GetParam()) != "bluestore")
+    return;
+  g_conf->set_val("bluestore_min_alloc_size", "4096");
+  g_ceph_context->_conf->apply_changes(NULL);
+  int r = store->umount();
+  ASSERT_EQ(r, 0);
+  r = store->mount(); //to force min_alloc_size update
+  ASSERT_EQ(r, 0);
+
+  ObjectStore::Sequencer osr("test");
+  coll_t cid;
+  ghobject_t a(hobject_t(sobject_t("Object 1", CEPH_NOSNAP)));
+  {
+    ObjectStore::Transaction t;
+    t.create_collection(cid, 0);
+    cerr << "Creating collection " << cid << std::endl;
+    r = apply_transaction(store, &osr, std::move(t));
+    ASSERT_EQ(r, 0);
+  }
+  bufferlist bl;
+  int len = 4096;
+  bufferptr bp(len);
+  bp.zero();
+  bl.append(bp);
+  for (int i=0; i<1000; ++i) {
+    ObjectStore::Transaction t;
+    t.write(cid, a, i*2*len, len, bl, 0);
+    r = apply_transaction(store, &osr, std::move(t));
+    ASSERT_EQ(r, 0);
+  }
+  for (int i=0; i<1000; ++i) {
+    ObjectStore::Transaction t;
+    t.write(cid, a, i*2*len + 1, len, bl, 0);
+    r = apply_transaction(store, &osr, std::move(t));
+    ASSERT_EQ(r, 0);
+  }
+  {
+    ObjectStore::Transaction t;
+    t.remove(cid, a);
+    t.remove_collection(cid);
+    cerr << "Cleaning" << std::endl;
+    r = apply_transaction(store, &osr, std::move(t));
+    ASSERT_EQ(r, 0);
+  }
+  g_conf->set_val("bluestore_min_alloc_size", "0");
+  g_ceph_context->_conf->apply_changes(NULL);
+}
+
 TEST_P(StoreTest, SyntheticMatrixCsumAlgorithm) {
   if (string(GetParam()) != "bluestore")
     return;