librbd: new rbd_journal_object_writethrough_until_flush option

When set to true, the journal will not attempt to batch appends until
after it receives the the first flush request from the user.

Signed-off-by: Jason Dillaman <dillaman@redhat.com>

3 files changed, 31 insertions, 4 deletions

diff --git a/src/librbd/Journal.cc b/src/librbd/Journal.cc
index 11b2f23fa0f..66abdf90207 100644
--- a/src/librbd/Journal.cc
+++ b/src/librbd/Journal.cc
@@ -715,6 +715,26 @@ void Journal<I>::flush_commit_position(Context *on_finish) {
 }
 
 template <typename I>
+void Journal<I>::user_flushed() {
+  if (m_state == STATE_READY && !m_user_flushed.exchange(true) &&
+      m_image_ctx.config.template get_val<bool>("rbd_journal_object_writethrough_until_flush")) {
+    Mutex::Locker locker(m_lock);
+    if (m_state == STATE_READY) {
+      CephContext *cct = m_image_ctx.cct;
+      ldout(cct, 5) << this << " " << __func__ << dendl;
+
+      ceph_assert(m_journaler != nullptr);
+      m_journaler->set_append_batch_options(
+        m_image_ctx.config.template get_val<uint64_t>("rbd_journal_object_flush_interval"),
+        m_image_ctx.config.template get_val<Option::size_t>("rbd_journal_object_flush_bytes"),
+        m_image_ctx.config.template get_val<double>("rbd_journal_object_flush_age"));
+    } else {
+      m_user_flushed = false;
+    }
+  }
+}
+
+template <typename I>
 uint64_t Journal<I>::append_write_event(uint64_t offset, size_t length,
                                         const bufferlist &bl,
                                         bool flush_entry) {
@@ -1172,10 +1192,12 @@ void Journal<I>::start_append() {
 
   m_journaler->start_append(
     m_image_ctx.config.template get_val<uint64_t>("rbd_journal_object_max_in_flight_appends"));
-  m_journaler->set_append_batch_options(
-    m_image_ctx.config.template get_val<uint64_t>("rbd_journal_object_flush_interval"),
-    m_image_ctx.config.template get_val<Option::size_t>("rbd_journal_object_flush_bytes"),
-    m_image_ctx.config.template get_val<double>("rbd_journal_object_flush_age"));
+  if (!m_image_ctx.config.template get_val<bool>("rbd_journal_object_writethrough_until_flush")) {
+    m_journaler->set_append_batch_options(
+      m_image_ctx.config.template get_val<uint64_t>("rbd_journal_object_flush_interval"),
+      m_image_ctx.config.template get_val<Option::size_t>("rbd_journal_object_flush_bytes"),
+      m_image_ctx.config.template get_val<double>("rbd_journal_object_flush_age"));
+  }
 
   transition_state(STATE_READY, 0);
 }
diff --git a/src/librbd/Journal.h b/src/librbd/Journal.h
index 88979b6252b..e63cc4a7137 100644
--- a/src/librbd/Journal.h
+++ b/src/librbd/Journal.h
@@ -130,6 +130,8 @@ public:
 
   void flush_commit_position(Context *on_finish);
 
+  void user_flushed();
+
   uint64_t append_write_event(uint64_t offset, size_t length,
                               const bufferlist &bl,
                               bool flush_entry);
@@ -291,6 +293,8 @@ private:
   uint64_t m_event_tid;
   Events m_events;
 
+  std::atomic<bool> m_user_flushed = false;
+
   std::atomic<uint64_t> m_op_tid = { 0 };
   TidToFutures m_op_futures;
 
diff --git a/src/librbd/io/ImageRequest.cc b/src/librbd/io/ImageRequest.cc
index ce064f1f07a..166c8702308 100644
--- a/src/librbd/io/ImageRequest.cc
+++ b/src/librbd/io/ImageRequest.cc
@@ -686,6 +686,7 @@ void ImageFlushRequest<I>::send_request() {
     ceph_assert(image_ctx.journal != NULL);
     journal_tid = image_ctx.journal->append_io_event(
       journal::EventEntry(journal::AioFlushEvent()), 0, 0, false, 0);
+    image_ctx.journal->user_flushed();
   }
 
   auto object_dispatch_spec = ObjectDispatchSpec::create_flush(