crimson/os/seastore/EPM: coalesce writes using the same writer

After integrated with RecordSubmitter, Writer is able to coalesce
concurrent writes into record groups. There is no need to distribute
writes to multiple Writers for both large and small write sizes.

Signed-off-by: Yingxin Cheng <yingxin.cheng@intel.com>

diff --git a/src/common/options/crimson.yaml.in b/src/common/options/crimson.yaml.in
index 7f906a889589742db2afd2f0ceeac5d21cfeafd7..ee9ee653ec786ab2d0edea0e1576cc5bd1da2d81 100644 (file)
--- a/src/common/options/crimson.yaml.in
+++ b/src/common/options/crimson.yaml.in
@@ -40,11 +40,6 @@ options:
   default: true
   see_also:
   - seastore_device_size
-- name: seastore_init_write_segments_num_per_device
-  type: uint
-  level: dev
-  desc: Initial number of segments for writing new extents per device
-  default: 6
 - name: seastore_journal_batch_capacity
   type: uint
   level: dev

diff --git a/src/crimson/os/seastore/extent_placement_manager.cc b/src/crimson/os/seastore/extent_placement_manager.cc
index fcb9cb3947859e84a9201397b693dcff05338110..660e8d554f4c32c53f6364c15abd93657fbb30ad 100644 (file)
--- a/src/crimson/os/seastore/extent_placement_manager.cc
+++ b/src/crimson/os/seastore/extent_placement_manager.cc
@@ -12,22 +12,16 @@ namespace crimson::os::seastore {
 SegmentedAllocator::SegmentedAllocator(
   SegmentProvider& sp,
   SegmentManager& sm)
-  : rewriter(sp, sm)
+  : cold_writer{"COLD", sp, sm},
+    rewrite_writer{"REWRITE", sp, sm}
 {
-  std::generate_n(
-    std::back_inserter(writers),
-    crimson::common::get_conf<uint64_t>(
-      "seastore_init_write_segments_num_per_device"),
-    [&] {
-      return Writer{sp, sm};
-    }
-  );
 }
 
 SegmentedAllocator::Writer::Writer(
+  std::string name,
   SegmentProvider& sp,
   SegmentManager& sm)
-  : segment_allocator("OOL", segment_type_t::OOL, sp, sm),
+  : segment_allocator(name, segment_type_t::OOL, sp, sm),
     record_submitter(crimson::common::get_conf<uint64_t>(
                        "seastore_journal_iodepth_limit"),
                      crimson::common::get_conf<uint64_t>(

diff --git a/src/crimson/os/seastore/extent_placement_manager.h b/src/crimson/os/seastore/extent_placement_manager.h
index 263dc0cf42181d631f067e4657a5aba524495987..0d1c4e49cfb517168762e7100b8bfe09667897fb 100644 (file)
--- a/src/crimson/os/seastore/extent_placement_manager.h
+++ b/src/crimson/os/seastore/extent_placement_manager.h
@@ -75,7 +75,7 @@ class SegmentProvider;
 class SegmentedAllocator : public ExtentAllocator {
   class Writer : public ExtentOolWriter {
   public:
-    Writer(SegmentProvider& sp, SegmentManager& sm);
+    Writer(std::string name, SegmentProvider& sp, SegmentManager& sm);
     Writer(Writer &&) = default;
 
     open_ertr::future<> open() final;
@@ -112,16 +112,20 @@ public:
     SegmentManager& sm);
 
   Writer &get_writer(placement_hint_t hint) {
-    if (hint == placement_hint_t::REWRITE) {
-      return rewriter;
+    assert(hint >= placement_hint_t::COLD);
+    assert(hint < placement_hint_t::NUM_HINTS);
+    if (hint == placement_hint_t::COLD) {
+      return cold_writer;
     } else {
-      return writers[std::rand() % writers.size()];
+      assert(hint == placement_hint_t::REWRITE);
+      return rewrite_writer;
     }
   }
 
   open_ertr::future<> open() {
-    return crimson::do_for_each(writers, [](auto& writer) {
-      return writer.open();
+    return cold_writer.open(
+    ).safe_then([this] {
+      return rewrite_writer.open();
     });
   }
 
@@ -146,13 +150,17 @@ public:
   }
 
   stop_ertr::future<> stop() {
-    return crimson::do_for_each(writers, [](auto& writer) {
-      return writer.stop();
+    return cold_writer.stop(
+    ).safe_then([this] {
+      return rewrite_writer.stop();
     });
   }
 private:
-  Writer rewriter;
-  std::vector<Writer> writers;
+  // TODO:
+  // - hot_writer
+  // - a map of hint -> writer
+  Writer cold_writer;
+  Writer rewrite_writer;
 };
 
 class ExtentPlacementManager {
@@ -202,7 +210,7 @@ public:
               std::move(bp)};
     }
 
-    // FIXME: set delay for COLD extent when the record overhead is low
+    // FIXME: set delay for COLD extent and improve GC
     // NOTE: delay means to delay the decision about whether to write the
     // extent as inline or out-of-line extents.
     bool delay = (hint > placement_hint_t::COLD &&