os/bluestore: Add fault_range_ex

Add fault_range_ex, a sibling to fault_range.
The new function does exactly the same as the original.
The difference is the returned range <start, end> that reflects
range encompassed by shards affected.

Signed-off-by: Adam Kupczyk <akupczyk@ibm.com>

diff --git a/src/os/bluestore/BlueStore.cc b/src/os/bluestore/BlueStore.cc
index 7dca7f49f1e51817c81dd02a9947ca2668bbf210..cd36ad6eabe24a4c21c795404944507bbae57ecc 100644 (file)
--- a/src/os/bluestore/BlueStore.cc
+++ b/src/os/bluestore/BlueStore.cc
@@ -4178,6 +4178,29 @@ void BlueStore::ExtentMap::init_shards(bool loaded, bool dirty)
   }
 }
 
+std::pair<uint32_t, uint32_t> BlueStore::ExtentMap::fault_range_ex(
+  KeyValueDB *db,
+  uint32_t offset,
+  uint32_t length)
+{
+  dout(30) << __func__ << " 0x" << std::hex << offset << "~" << length
+          << std::dec << dendl;
+  if (shards.size() == 0) {
+    // no sharding yet; everyting is loaded
+    return {0, OBJECT_MAX_SIZE};
+  }
+  auto start = seek_shard(offset);
+  auto last = seek_shard(offset + length);
+  maybe_load_shard(db, start, last);
+  uint32_t left_bound = shards[start].shard_info->offset;
+  uint32_t right_bound =  (size_t)last + 1 < shards.size() ?
+                          shards[last + 1].shard_info->offset : OBJECT_MAX_SIZE;
+  dout(20) << __func__ << " start=" << start << " last=" << last
+    << " -> 0x" << std::hex << left_bound << "~" << right_bound
+    << std::dec << dendl;
+  return {left_bound, right_bound};
+}
+
 void BlueStore::ExtentMap::fault_range(
   KeyValueDB *db,
   uint32_t offset,
@@ -4191,6 +4214,14 @@ void BlueStore::ExtentMap::fault_range(
   }
   auto start = seek_shard(offset);
   auto last = seek_shard(offset + length);
+  maybe_load_shard(db, start, last);
+}
+
+void BlueStore::ExtentMap::maybe_load_shard(
+  KeyValueDB *db,
+  int start,
+  int last)
+{
   ceph_assert(last >= start);
   ceph_assert(start >= 0);
 
@@ -4216,9 +4247,10 @@ void BlueStore::ExtentMap::fault_range(
       );
       p->extents = decode_some(v);
       p->loaded = true;
-      dout(20) << __func__ << " open shard 0x" << std::hex
-              << p->shard_info->offset
-              << " for range 0x" << offset << "~" << length << std::dec
+      uint32_t shard_end =
+        (size_t)start + 1 < shards.size() ? (p + 1)->shard_info->offset : OBJECT_MAX_SIZE;
+      dout(20) << __func__ << " open shard for range 0x"
+               << std::hex << p->shard_info->offset << "~" << shard_end << std::dec
               << " (" << v.length() << " bytes)" << dendl;
       ceph_assert(p->dirty == false);
       ceph_assert(v.length() == p->shard_info->bytes);
@@ -17621,10 +17653,11 @@ int BlueStore::_do_write_v2(
   if (bl.length() != length) {
     bl.splice(length, bl.length() - length);
   }
+  BlueStore::Writer wr(this, txc, &wctx, o);
   uint64_t start = p2align(offset, min_alloc_size);
   uint64_t end = p2roundup(offset + length, min_alloc_size);
-  o->extent_map.fault_range(db, start, end - start);
-  BlueStore::Writer wr(this, txc, &wctx, o);
+  std::tie(wr.left_shard_bound, wr.right_shard_bound) =
+    o->extent_map.fault_range_ex(db, start, end - start);
   wr.do_write(offset, bl);
   return r;
 }

diff --git a/src/os/bluestore/BlueStore.h b/src/os/bluestore/BlueStore.h
index b535d16bba56f291092cae14b9ad1020c5265056..9ca48cea44136e96fcc0ce5dfa9d11a20d3b4398 100644 (file)
--- a/src/os/bluestore/BlueStore.h
+++ b/src/os/bluestore/BlueStore.h
@@ -1149,6 +1149,16 @@ public:
     /// ensure that a range of the map is loaded
     void fault_range(KeyValueDB *db,
                     uint32_t offset, uint32_t length);
+    /// ensure that a range of the map is loaded
+    /// return range that is encompassed by affected shards
+    std::pair<uint32_t, uint32_t> fault_range_ex(
+      KeyValueDB *db,
+      uint32_t offset,
+      uint32_t length);
+    void maybe_load_shard(
+      KeyValueDB *db,
+      int begin_shard,
+      int end_shard);
 
     /// ensure a range of the map is marked dirty
     void dirty_range(uint32_t offset, uint32_t length);

diff --git a/src/os/bluestore/Writer.cc b/src/os/bluestore/Writer.cc
index b2d3195dc9e2a7c6e1d550f329947587c03bbc04..a41cab3cb87297c77ab46551f126c2c01e27988b 100644 (file)
--- a/src/os/bluestore/Writer.cc
+++ b/src/os/bluestore/Writer.cc
@@ -732,6 +732,7 @@ void BlueStore::Writer::_try_reuse_allocated_l(
   blob_data_t& bd)            // modified when consumed
 {
   uint32_t search_stop = p2align(logical_offset, (uint32_t)wctx->target_blob_size);
+  search_stop = std::max(left_shard_bound, search_stop);
   uint32_t au_size = bstore->min_alloc_size;
   uint32_t block_size = bstore->block_size;
   ceph_assert(!bd.is_compressed());
@@ -812,6 +813,7 @@ void BlueStore::Writer::_try_reuse_allocated_r(
   uint32_t block_size = bstore->block_size;
   uint32_t blob_size = wctx->target_blob_size;
   uint32_t search_end = p2roundup(end_offset, blob_size);
+  search_end = std::min(right_shard_bound, search_end);
   ceph_assert(!bd.is_compressed());
   ceph_assert(p2phase<uint32_t>(end_offset, au_size) != 0);
   BlueStore::ExtentMap& emap = onode->extent_map;

diff --git a/src/os/bluestore/Writer.h b/src/os/bluestore/Writer.h
index 82b164befbb6c530a737b6faec8b52698f86ea0d..f12d25c2f1bfab52ae3c27ebad0510647c5a31a7 100644 (file)
--- a/src/os/bluestore/Writer.h
+++ b/src/os/bluestore/Writer.h
@@ -50,7 +50,8 @@ public:
     virtual bufferlist read(uint32_t object_offset, uint32_t object_length) = 0;
   };
   Writer(BlueStore* bstore, TransContext* txc, WriteContext* wctx, OnodeRef o)
-    :bstore(bstore), txc(txc), wctx(wctx), onode(o) {
+    :left_shard_bound(0), right_shard_bound(OBJECT_MAX_SIZE)
+    , bstore(bstore), txc(txc), wctx(wctx), onode(o) {
       pp_mode = debug_level_to_pp_mode(bstore->cct);
     }
 public:
@@ -67,6 +68,8 @@ public:
   read_divertor* test_read_divertor = nullptr;
   std::vector<BlobRef> pruned_blobs;
   volatile_statfs statfs_delta;
+  uint32_t left_shard_bound;  // if sharding is in effect,
+  uint32_t right_shard_bound; // do not cross this line
 
 private:
   BlueStore* bstore;