crimson/os/seastore/rbm: write merged-able ool extents at once

This commit intens to write meraged-able ool extents, each of
which is located at consecutive address, to the RBM at once.

Signed-off-by: Myoungwon Oh <[email protected]>

src/crimson/os/seastore/extent_placement_manager.cc

@@ -1001,7 +1001,6 @@ RandomBlockOolWriter::do_write(
   DEBUGT("start with {} allocated extents",
          t, extents.size());
   std::vector<write_info_t> writes;
-  writes.reserve(extents.size());
   for (auto& ex : extents) {
     auto paddr = ex->get_paddr();
     assert(paddr.is_absolute());
@@ -1012,27 +1011,24 @@ RandomBlockOolWriter::do_write(
     auto& stats = t.get_ool_write_stats();
     stats.extents.num += 1;
     stats.extents.bytes += ex->get_length();
-    stats.num_records += 1;
-
     ex->prepare_write();
-    extent_len_t offset = 0;
+
     bufferptr bp;
     if (can_inplace_rewrite(t, ex)) {
       assert(ex->is_logical());
       auto r = ex->template cast<LogicalCachedExtent>()->get_modified_region();
       ceph_assert(r.has_value());
-      offset = p2align(r->offset, rbm->get_block_size());
+      extent_len_t offset = p2align(r->offset, rbm->get_block_size());
       extent_len_t len =
 	p2roundup(r->offset + r->len, rbm->get_block_size()) - offset;
       bp = ceph::bufferptr(ex->get_bptr(), offset, len);
+      paddr = ex->get_paddr() + offset;
     } else {
       bp = ex->get_bptr();
       auto& trans_stats = get_by_src(w_stats.stats_by_src, t.get_src());
-      ++(trans_stats.num_records);
       trans_stats.data_bytes += ex->get_length();
       w_stats.data_bytes += ex->get_length();
     }
-    writes.push_back(write_info_t{paddr + offset, std::move(bp), rbm});
 
     if (ex->is_initial_pending()) {
       t.mark_allocated_extent_ool(ex);
@@ -1043,11 +1039,52 @@ RandomBlockOolWriter::do_write(
     } else {
       ceph_assert("impossible");
     }
+
+    // TODO : allocate a consecutive address based on a transaction
+    if (writes.size() != 0 &&
+        writes.back().offset + writes.back().bp.length() == paddr) {
+      // We can write both the currrent extent and the previous one at once
+      // if the extents are located in a row
+      if (writes.back().mergeable_bps.size() == 0) {
+	 writes.back().mergeable_bps.push_back(writes.back().bp);
+      }
+      writes.back().mergeable_bps.push_back(ex->get_bptr());
+    } else {
+      // Write a single extent in the existing way
+      write_info_t w_info;
+      w_info.offset = paddr;
+      w_info.rbm = rbm;
+      w_info.bp = bp;
+      writes.push_back(w_info);
+    }
+    TRACE("current extent: base off {} len {},\
+      maybe-merged current extent: base off {} len {}",
+      paddr, ex->get_length(), writes.back().offset, writes.back().bp.length());
+  }
+
+  for (auto &w : writes) {
+    if (w.mergeable_bps.size() > 0) {
+      extent_len_t len = 0;
+      for (auto &b : w.mergeable_bps) {
+	len += b.length();
+      }
+      w.bp = ceph::bufferptr(ceph::buffer::create_page_aligned(len));
+      extent_len_t cursor = 0;
+      for (auto &b : w.mergeable_bps) {
+	w.bp.copy_in(cursor, b.length(), b.c_str());
+	cursor += b.length();
+      }
+      w.mergeable_bps.clear();
+    }
   }
 
   return trans_intr::make_interruptible(
     seastar::do_with(std::move(writes),
-      [](auto& writes) {
+      [&t, this](auto& writes) {
+      auto& stats = t.get_ool_write_stats();
+      stats.num_records += writes.size();
+      auto& trans_stats = get_by_src(w_stats.stats_by_src, t.get_src());
+      trans_stats.num_records += writes.size();
       return crimson::do_for_each(writes,
         [](auto& info) {
         return info.rbm->write(info.offset, info.bp

src/crimson/os/seastore/extent_placement_manager.h

@@ -194,6 +194,7 @@ class RandomBlockOolWriter : public ExtentOolWriter {
     paddr_t offset;
     ceph::bufferptr bp;
     RandomBlockManager* rbm;
+    std::list<ceph::bufferptr> mergeable_bps;
   };
   alloc_write_iertr::future<> do_write(
     Transaction& t,