os/bluestore/BlueFS: do not dirty log file when size changes

We only need to dirty the log file if the allocate changes.  Replay is
smart enough to learn the file size as it goes.

Signed-off-by: Sage Weil <sage@redhat.com>

diff --git a/src/os/bluestore/BlueFS.cc b/src/os/bluestore/BlueFS.cc
index cf096869f767de9cd9a9eeff98125f85612cfb6c..0e0eb0c54c2f663004bff9928459849eecdbaf52 100644 (file)
--- a/src/os/bluestore/BlueFS.cc
+++ b/src/os/bluestore/BlueFS.cc
@@ -1110,7 +1110,12 @@ int BlueFS::_flush_range(FileWriter *h, uint64_t offset, uint64_t length)
   }
   if (h->file->fnode.size < offset + length) {
     h->file->fnode.size = offset + length;
-    must_dirty = true;
+    if (h->file->fnode.ino != 1) {
+      // we do not need to dirty the log file when the file size
+      // changes because replay is smart enough to discover it on its
+      // own.
+      must_dirty = true;
+    }
   }
   if (must_dirty) {
     h->file->fnode.mtime = ceph_clock_now(NULL);