Add soft and hard rate limit support

Summary:
This diff adds support for both soft and hard rate limiting. The following changes are included:

1) Options.rate_limit is renamed to Options.hard_rate_limit.
2) Options.rate_limit_delay_milliseconds is renamed to Options.rate_limit_delay_max_milliseconds.
3) Options.soft_rate_limit is added.
4) If the maximum compaction score is > hard_rate_limit and rate_limit_delay_max_milliseconds == 0, then writes are delayed by 1 ms at a time until the max compaction score falls below hard_rate_limit.
5) If the max compaction score is > soft_rate_limit but <= hard_rate_limit, then writes are delayed by 0-1 ms depending on how close we are to hard_rate_limit.
6) Users can disable 4 by setting hard_rate_limit = 0. They can add a limit to the maximum amount of time waited by setting rate_limit_delay_max_milliseconds > 0. Thus, the old behavior can be preserved by setting soft_rate_limit = 0, which is the default.

Test Plan:
make -j32 check
./db_stress

Reviewers: dhruba, haobo, MarkCallaghan

Reviewed By: dhruba

CC: leveldb

Differential Revision: https://reviews.facebook.net/D12003

db/db_bench.cc

@@ -264,13 +264,16 @@ static int FLAGS_stats_interval = 0;
 // than 0.
 static int FLAGS_stats_per_interval = 0;
 
+static double FLAGS_soft_rate_limit = 0;
+
 // When not equal to 0 this make threads sleep at each stats
 // reporting interval until the compaction score for all levels is
 // less than or equal to this value.
-static double FLAGS_rate_limit = 0;
+static double FLAGS_hard_rate_limit = 0;
 
-// When FLAGS_rate_limit is set then this is the max time a put will be stalled.
-static int FLAGS_rate_limit_delay_milliseconds = 1000;
+// When FLAGS_hard_rate_limit is set then this is the max time a put will be
+// stalled.
+static int FLAGS_rate_limit_delay_max_milliseconds = 1000;
 
 // Control maximum bytes of overlaps in grandparent (i.e., level+2) before we
 // stop building a single file in a level->level+1 compaction.
@@ -1146,8 +1149,10 @@ unique_ptr<char []> GenerateKeyFromInt(int v, const char* suffix = "")
     options.disable_seek_compaction = FLAGS_disable_seek_compaction;
     options.delete_obsolete_files_period_micros =
       FLAGS_delete_obsolete_files_period_micros;
-    options.rate_limit = FLAGS_rate_limit;
-    options.rate_limit_delay_milliseconds = FLAGS_rate_limit_delay_milliseconds;
+    options.soft_rate_limit = FLAGS_soft_rate_limit;
+    options.hard_rate_limit = FLAGS_hard_rate_limit;
+    options.rate_limit_delay_max_milliseconds =
+      FLAGS_rate_limit_delay_max_milliseconds;
     options.table_cache_numshardbits = FLAGS_table_cache_numshardbits;
     options.max_grandparent_overlap_factor =
       FLAGS_max_grandparent_overlap_factor;
@@ -2039,7 +2044,8 @@ int main(int argc, char** argv) {
     } else if (sscanf(argv[i], "--min_write_buffer_number_to_merge=%d%c",
                &n, &junk) == 1) {
       FLAGS_min_write_buffer_number_to_merge = n;
-    } else if (sscanf(argv[i], "--max_background_compactions=%d%c", &n, &junk) == 1) {
+    } else if (sscanf(argv[i], "--max_background_compactions=%d%c", &n, &junk)
+               == 1) {
       FLAGS_max_background_compactions = n;
     } else if (sscanf(argv[i], "--cache_size=%ld%c", &l, &junk) == 1) {
       FLAGS_cache_size = l;
@@ -2173,13 +2179,16 @@ int main(int argc, char** argv) {
     } else if (sscanf(argv[i], "--stats_per_interval=%d%c", &n, &junk) == 1
         && (n == 0 || n == 1)) {
       FLAGS_stats_per_interval = n;
-    } else if (sscanf(argv[i], "--rate_limit=%lf%c", &d, &junk) == 1 &&
+    } else if (sscanf(argv[i], "--soft_rate_limit=%lf%c", &d, &junk) == 1 &&
+               d > 0.0) {
+      FLAGS_soft_rate_limit = d;
+    } else if (sscanf(argv[i], "--hard_rate_limit=%lf%c", &d, &junk) == 1 &&
                d > 1.0) {
-      FLAGS_rate_limit = d;
+      FLAGS_hard_rate_limit = d;
     } else if (sscanf(argv[i],
-               "--rate_limit_delay_milliseconds=%d%c", &n, &junk) == 1
-        && n > 0) {
-      FLAGS_rate_limit_delay_milliseconds = n;
+               "--rate_limit_delay_max_milliseconds=%d%c", &n, &junk) == 1
+        && n >= 0) {
+      FLAGS_rate_limit_delay_max_milliseconds = n;
     } else if (sscanf(argv[i], "--readonly=%d%c", &n, &junk) == 1 &&
         (n == 0 || n ==1 )) {
       FLAGS_read_only = n;

db/db_impl.cc

@@ -154,6 +154,9 @@ Options SanitizeOptions(const std::string& dbname,
   if (result.max_mem_compaction_level >= result.num_levels) {
     result.max_mem_compaction_level = result.num_levels - 1;
   }
+  if (result.soft_rate_limit > result.hard_rate_limit) {
+    result.soft_rate_limit = result.hard_rate_limit;
+  }
   return result;
 }
 
@@ -2417,31 +2420,29 @@ WriteBatch* DBImpl::BuildBatchGroup(Writer** last_writer) {
 // This function computes the amount of time in microseconds by which a write
 // should be delayed based on the number of level-0 files according to the
 // following formula:
-// if num_level_files < level0_slowdown_writes_trigger, return 0;
-// if num_level_files >= level0_stop_writes_trigger, return 1000;
-// otherwise, let r = (num_level_files - level0_slowdown) /
-//                    (level0_stop - level0_slowdown)
+// if n < bottom, return 0;
+// if n >= top, return 1000;
+// otherwise, let r = (n - bottom) /
+//                    (top - bottom)
 //  and return r^2 * 1000.
 // The goal of this formula is to gradually increase the rate at which writes
 // are slowed. We also tried linear delay (r * 1000), but it seemed to do
 // slightly worse. There is no other particular reason for choosing quadratic.
-uint64_t DBImpl::SlowdownAmount(int num_level0_files) {
+uint64_t DBImpl::SlowdownAmount(int n, int top, int bottom) {
   uint64_t delay;
-  int stop_trigger = options_.level0_stop_writes_trigger;
-  int slowdown_trigger = options_.level0_slowdown_writes_trigger;
-  if (num_level0_files >= stop_trigger) {
+  if (n >= top) {
     delay = 1000;
   }
-  else if (num_level0_files < slowdown_trigger) {
+  else if (n < bottom) {
     delay = 0;
   }
   else {
     // If we are here, we know that:
-    //   slowdown_trigger <= num_level0_files < stop_trigger
+    //   level0_start_slowdown <= n < level0_slowdown
     // since the previous two conditions are false.
     float how_much =
-      (float) (num_level0_files - slowdown_trigger) /
-              (stop_trigger - slowdown_trigger);
+      (float) (n - bottom) /
+              (top - bottom);
     delay = how_much * how_much * 1000;
   }
   assert(delay <= 1000);
@@ -2454,7 +2455,8 @@ Status DBImpl::MakeRoomForWrite(bool force) {
   mutex_.AssertHeld();
   assert(!writers_.empty());
   bool allow_delay = !force;
-  bool allow_rate_limit_delay = !force;
+  bool allow_hard_rate_limit_delay = !force;
+  bool allow_soft_rate_limit_delay = !force;
   uint64_t rate_limit_delay_millis = 0;
   Status s;
   double score;
@@ -2478,7 +2480,11 @@ Status DBImpl::MakeRoomForWrite(bool force) {
       uint64_t delayed;
       {
         StopWatch sw(env_, options_.statistics, STALL_L0_SLOWDOWN_COUNT);
-        env_->SleepForMicroseconds(SlowdownAmount(versions_->NumLevelFiles(0)));
+        env_->SleepForMicroseconds(
+          SlowdownAmount(versions_->NumLevelFiles(0),
+                         options_.level0_slowdown_writes_trigger,
+                         options_.level0_stop_writes_trigger)
+        );
         delayed = sw.ElapsedMicros();
       }
       RecordTick(options_.statistics, STALL_L0_SLOWDOWN_MICROS, delayed);
@@ -2527,9 +2533,9 @@ Status DBImpl::MakeRoomForWrite(bool force) {
       stall_level0_num_files_ += stall;
       stall_level0_num_files_count_++;
     } else if (
-        allow_rate_limit_delay &&
-        options_.rate_limit > 1.0 &&
-        (score = versions_->MaxCompactionScore()) > options_.rate_limit) {
+        allow_hard_rate_limit_delay &&
+        options_.hard_rate_limit > 1.0 &&
+        (score = versions_->MaxCompactionScore()) > options_.hard_rate_limit) {
       // Delay a write when the compaction score for any level is too large.
       int max_level = versions_->MaxCompactionScoreLevel();
       mutex_.Unlock();
@@ -2545,14 +2551,29 @@ Status DBImpl::MakeRoomForWrite(bool force) {
       uint64_t rate_limit = std::max((delayed / 1000), (uint64_t) 1);
       rate_limit_delay_millis += rate_limit;
       RecordTick(options_.statistics, RATE_LIMIT_DELAY_MILLIS, rate_limit);
-      if (rate_limit_delay_millis >=
-          (unsigned)options_.rate_limit_delay_milliseconds) {
-        allow_rate_limit_delay = false;
+      if (options_.rate_limit_delay_max_milliseconds > 0 &&
+          rate_limit_delay_millis >=
+          (unsigned)options_.rate_limit_delay_max_milliseconds) {
+        allow_hard_rate_limit_delay = false;
       }
       // Log(options_.info_log,
       //    "delaying write %llu usecs for rate limits with max score %.2f\n",
       //    (long long unsigned int)delayed, score);
       mutex_.Lock();
+    } else if (
+        allow_soft_rate_limit_delay &&
+        options_.soft_rate_limit > 0.0 &&
+        (score = versions_->MaxCompactionScore()) > options_.soft_rate_limit) {
+      // Delay a write when the compaction score for any level is too large.
+      // TODO: add statistics
+      mutex_.Unlock();
+      env_->SleepForMicroseconds(SlowdownAmount(
+        score,
+        options_.soft_rate_limit,
+        options_.hard_rate_limit)
+      );
+      allow_soft_rate_limit_delay = false;
+      mutex_.Lock();
     } else {
       // Attempt to switch to a new memtable and trigger compaction of old
       DelayLoggingAndReset();

db/db_impl.h

@@ -166,7 +166,7 @@ class DBImpl : public DB {
   Status WriteLevel0Table(std::vector<MemTable*> &mems, VersionEdit* edit,
                                 uint64_t* filenumber);
 
-  uint64_t SlowdownAmount(int num_level0_files);
+  uint64_t SlowdownAmount(int n, int top, int bottom);
   Status MakeRoomForWrite(bool force /* compact even if there is room? */);
   WriteBatch* BuildBatchGroup(Writer** last_writer);
 

db/db_test.cc

@@ -286,8 +286,8 @@ class DBTest {
         options.purge_redundant_kvs_while_flush = !options.purge_redundant_kvs_while_flush;
         break;
       case kPerfOptions:
-        options.rate_limit = 2.0;
-        options.rate_limit_delay_milliseconds = 2;
+        options.hard_rate_limit = 2.0;
+        options.rate_limit_delay_max_milliseconds = 2;
         // TODO -- test more options
         break;
       case kDeletesFilterFirst:

include/leveldb/options.h

@@ -355,12 +355,22 @@ struct Options {
   // Default: 1000
   size_t keep_log_file_num;
 
-  // Puts are delayed when any level has a compaction score that
-  // exceeds rate_limit. This is ignored when <= 1.0.
-  double rate_limit;
+  // Puts are delayed 0-1 ms when any level has a compaction score that exceeds
+  // soft_rate_limit. This is ignored when == 0.0.
+  // CONSTRAINT: soft_rate_limit <= hard_rate_limit. If this constraint does not
+  // hold, RocksDB will set soft_rate_limit = hard_rate_limit
+  // Default: 0 (disabled)
+  double soft_rate_limit;
+
+  // Puts are delayed 1ms at a time when any level has a compaction score that
+  // exceeds hard_rate_limit. This is ignored when <= 1.0.
+  // Default: 0 (disabled)
+  double hard_rate_limit;
 
-  // Max time a put will be stalled when rate_limit is enforced
-  unsigned int rate_limit_delay_milliseconds;
+  // Max time a put will be stalled when hard_rate_limit is enforced. If 0, then
+  // there is no limit.
+  // Default: 1000
+  unsigned int rate_limit_delay_max_milliseconds;
 
   // manifest file is rolled over on reaching this limit.
   // The older manifest file be deleted.

util/options.cc

@@ -56,8 +56,9 @@ Options::Options()
       max_log_file_size(0),
       log_file_time_to_roll(0),
       keep_log_file_num(1000),
-      rate_limit(0.0),
-      rate_limit_delay_milliseconds(1000),
+      soft_rate_limit(0.0),
+      hard_rate_limit(0.0),
+      rate_limit_delay_max_milliseconds(1000),
       max_manifest_file_size(std::numeric_limits<uint64_t>::max()),
       no_block_cache(false),
       table_cache_numshardbits(4),
@@ -181,10 +182,10 @@ Options::Dump(Logger* log) const
         delete_obsolete_files_period_micros);
     Log(log,"             Options.max_background_compactions: %d",
         max_background_compactions);
-    Log(log,"                             Options.rate_limit: %.2f",
-        rate_limit);
-    Log(log,"          Options.rate_limit_delay_milliseconds: %d",
-        rate_limit_delay_milliseconds);
+    Log(log,"                      Options.hard_rate_limit: %.2f",
+        hard_rate_limit);
+    Log(log,"      Options.rate_limit_delay_max_milliseconds: %d",
+        rate_limit_delay_max_milliseconds);
     Log(log,"               Options.disable_auto_compactions: %d",
         disable_auto_compactions);
     Log(log,"                        Options.WAL_ttl_seconds: %ld",