Bugfixes: for Get(), don't hold mutex while writing log.

- Fix bug in Get: when it triggers a compaction, it could sometimes
  mark the compaction with the wrong level (if there was a gap
  in the set of levels examined for the Get).

- Do not hold mutex while writing to the log file or to the
  MANIFEST file.

  Added a new benchmark that runs a writer thread concurrently with
  reader threads.

  Percentiles
  ------------------------------
  micros/op: avg  median 99   99.9  99.99  99.999 max
  ------------------------------------------------------
  before:    42   38     110  225   32000  42000  48000
  after:     24   20     55   65    130    1100   7000

- Fixed race in optimized Get.  It should have been using the
  pinned memtables, not the current memtables.



git-svn-id: https://leveldb.googlecode.com/svn/trunk@50 62dab493-f737-651d-591e-8d6aee1b9529

db/db_impl.h

@@ -87,6 +87,11 @@ class DBImpl : public DB {
 
   Status WriteLevel0Table(MemTable* mem, VersionEdit* edit, Version* base);
 
+  // Only thread is allowed to log at a time.
+  struct LoggerId { };          // Opaque identifier for logging thread
+  void AcquireLoggingResponsibility(LoggerId* self);
+  void ReleaseLoggingResponsibility(LoggerId* self);
+
   Status MakeRoomForWrite(bool force /* compact even if there is room? */);
 
   struct CompactionState;
@@ -126,6 +131,8 @@ class DBImpl : public DB {
   WritableFile* logfile_;
   uint64_t logfile_number_;
   log::Writer* log_;
+  LoggerId* logger_;            // NULL, or the id of the current logging thread
+  port::CondVar logger_cv_;     // For threads waiting to log
   SnapshotList snapshots_;
 
   // Set of table files to protect from deletion because they are