Merge branch 'master' into performance

.gitignore

@@ -19,5 +19,4 @@ sst_dump
 util/build_version.cc
 build_tools/VALGRIND_LOGS/
 coverage/COVERAGE_REPORT
-util/build_version.cc.tmp
 .gdbhistory

build_tools/build_detect_version

@@ -5,38 +5,18 @@
 # is then built as a regular source file as part of the compilation process.
 # One can run "strings executable_filename | grep _build_" to find the version of
 # the source that we used to build the executable file.
-#
 
-# create git version file
-VFILE=$PWD/util/build_version.cc.tmp
-trap "rm $VFILE" EXIT
+OUTFILE="$PWD/util/build_version.cc"
 
-# check to see if git is in the path
-which git > /dev/null
-
-if [ "$?" = 0 ]; then
-  env -i git rev-parse HEAD 2>&1 |
-  awk '
-  BEGIN {
-    print "#include \"build_version.h\"\n"
-  }
-  { print "const char* rocksdb_build_git_sha = \"rocksdb_build_git_sha:" $0"\";" }
-  ' >  ${VFILE}
-else
-  echo "git not found" |
-  awk '
-    BEGIN {
-      print "#include \"build_version.h\""
-    }
-    { print "const char* rocksdb_build_git_sha = \"rocksdb_build_git_sha:git not found\";" }
-    ' > ${VFILE}
+GIT_SHA=""
+if command -v git >/dev/null 2>&1; then
+    GIT_SHA=$(git rev-parse HEAD 2>/dev/null)
 fi
 
-echo "const char* rocksdb_build_git_datetime = \"rocksdb_build_git_datetime:$(date)\";" >> ${VFILE}
-echo "const char* rocksdb_build_compile_date = __DATE__;" >> ${VFILE}
-echo "const char* rocksdb_build_compile_time = __TIME__;" >> ${VFILE}
-
-OUTFILE=$PWD/util/build_version.cc
-if [ ! -e $OUTFILE ] || ! cmp -s $VFILE $OUTFILE; then
-    cp $VFILE $OUTFILE
-fi
+cat > "${OUTFILE}" <<EOF
+#include "build_version.h"
+const char* rocksdb_build_git_sha = "rocksdb_build_git_sha:${GIT_SHA}";
+const char* rocksdb_build_git_datetime = "rocksdb_build_git_datetime:$(date)";
+const char* rocksdb_build_compile_date = __DATE__;
+const char* rocksdb_build_compile_time = __TIME__;
+EOF

db/db_impl.cc

@@ -244,6 +244,7 @@ DBImpl::DBImpl(const Options& options, const std::string& dbname)
       super_version_(nullptr),
       tmp_batch_(),
       bg_compaction_scheduled_(0),
+      bg_manual_only_(0),
       bg_flush_scheduled_(0),
       bg_logstats_scheduled_(false),
       manual_compaction_(nullptr),
@@ -1608,45 +1609,44 @@ void DBImpl::TEST_CompactRange(int level, const Slice* begin,const Slice* end) {
 
   MutexLock l(&mutex_);
 
-  // When a manual compaction arrives, temporarily throttle down
-  // the number of background compaction threads to 1. This is
-  // needed to ensure that this manual compaction can compact
-  // any range of keys/files. We artificialy increase
-  // bg_compaction_scheduled_ by a large number, this causes
-  // the system to have a single background thread. Now,
-  // this manual compaction can progress without stomping
-  // on any other concurrent compactions.
-  const int LargeNumber = 10000000;
-  const int newvalue = options_.max_background_compactions-1;
-  bg_compaction_scheduled_ += LargeNumber;
-  while (bg_compaction_scheduled_ > LargeNumber) {
-    Log(options_.info_log, "Manual compaction request waiting for background threads to fall below 1");
+  // When a manual compaction arrives, temporarily disable scheduling of
+  // non-manual compactions and wait until the number of scheduled compaction
+  // jobs drops to zero. This is needed to ensure that this manual compaction
+  // can compact any range of keys/files.
+  //
+  // bg_manual_only_ is non-zero when at least one thread is inside
+  // TEST_CompactRange(), i.e. during that time no other compaction will
+  // get scheduled (see MaybeScheduleFlushOrCompaction).
+  //
+  // Note that the following loop doesn't stop more that one thread calling
+  // TEST_CompactRange() from getting to the second while loop below.
+  // However, only one of them will actually schedule compaction, while
+  // others will wait on a condition variable until it completes.
+
+  ++bg_manual_only_;
+  while (bg_compaction_scheduled_ > 0) {
+    Log(options_.info_log,
+        "Manual compaction waiting for all other scheduled background "
+        "compactions to finish");
     bg_cv_.Wait();
   }
+
   Log(options_.info_log, "Manual compaction starting");
 
-  while (!manual.done) {
-    while (manual_compaction_ != nullptr) {
-      bg_cv_.Wait();
-    }
-    manual_compaction_ = &manual;
-    if (bg_compaction_scheduled_ == LargeNumber) {
-      bg_compaction_scheduled_ = newvalue;
-    }
-    MaybeScheduleFlushOrCompaction();
-    while (manual_compaction_ == &manual) {
+  while (!manual.done && !shutting_down_.Acquire_Load() && bg_error_.ok()) {
+    assert(bg_manual_only_ > 0);
+    if (manual_compaction_ != nullptr) {
+      // Running either this or some other manual compaction
       bg_cv_.Wait();
+    } else {
+      manual_compaction_ = &manual;
+      MaybeScheduleFlushOrCompaction();
     }
   }
-  assert(!manual.in_progress);
 
-  // wait till there are no background threads scheduled
-  bg_compaction_scheduled_ += LargeNumber;
-  while (bg_compaction_scheduled_ > LargeNumber + newvalue) {
-    Log(options_.info_log, "Manual compaction resetting background threads");
-    bg_cv_.Wait();
-  }
-  bg_compaction_scheduled_ = 0;
+  assert(!manual.in_progress);
+  assert(bg_manual_only_ > 0);
+  --bg_manual_only_;
 }
 
 Status DBImpl::FlushMemTable(const FlushOptions& options) {
@@ -1711,11 +1711,16 @@ void DBImpl::MaybeScheduleFlushOrCompaction() {
       env_->Schedule(&DBImpl::BGWorkFlush, this, Env::Priority::HIGH);
     }
 
+    // Schedule BGWorkCompaction if there's a compaction pending (or a memtable
+    // flush, but the HIGH pool is not enabled). Do it only if
+    // max_background_compactions hasn't been reached and, in case
+    // bg_manual_only_ > 0, if it's a manual compaction.
     if ((manual_compaction_ ||
          versions_->NeedsCompaction() ||
          (is_flush_pending && (options_.max_background_flushes <= 0))) &&
-        bg_compaction_scheduled_ < options_.max_background_compactions) {
-      // compaction needed, or memtable flush needed but HIGH pool not enabled.
+        bg_compaction_scheduled_ < options_.max_background_compactions &&
+        (!bg_manual_only_ || manual_compaction_)) {
+
       bg_compaction_scheduled_++;
       env_->Schedule(&DBImpl::BGWorkCompaction, this, Env::Priority::LOW);
     }

db/db_impl.h

@@ -389,9 +389,14 @@ class DBImpl : public DB {
   // part of ongoing compactions.
   std::set<uint64_t> pending_outputs_;
 
-  // count how many background compaction been scheduled or is running?
+  // count how many background compactions are running or have been scheduled
   int bg_compaction_scheduled_;
 
+  // If non-zero, MaybeScheduleFlushOrCompaction() will only schedule manual
+  // compactions (if manual_compaction_ is not null). This mechanism enables
+  // manual compactions to wait until all other compactions are finished.
+  int bg_manual_only_;
+
   // number of background memtable flush jobs, submitted to the HIGH pool
   int bg_flush_scheduled_;
 

util/autovector_test.cc

@@ -17,7 +17,7 @@ using namespace std;
 
 class AutoVectorTest { };
 
-const size_t kSize = 8;
+const unsigned long kSize = 8;
 TEST(AutoVectorTest, PushBackAndPopBack) {
   autovector<size_t, kSize> vec;
   ASSERT_TRUE(vec.empty());