options.delayed_write_rate use the rate of rate_limiter by default.

Summary:
It's hard for RocksDB to come up with a good default of delayed write rate. Use rate given by rate limiter if it is availalbe. This provides the I/O order of magnitude.
Closes facebook#2357

Differential Revision: D5115324

Pulled By: siying

fbshipit-source-id: 341065ad2211c981fc804011c0f0e59a50c7e754

DEFAULT_OPTIONS_HISTORY.md

@@ -1,5 +1,8 @@
 # RocksDB default options change log
 ## Unreleased
+* delayed_write_rate takes the rate given by rate_limiter if not specified.
+
+## 5.2
 * Change the default of delayed slowdown value to 16MB/s and further increase the L0 stop condition to 36 files.
 
 ## 5.0 (11/17/2016)

HISTORY.md

@@ -1,7 +1,8 @@
 # Rocksdb Change Log
 ## Unreleased
 ### Public API Change
-*Scheduling flushes and compactions in the same thread pool is no longer supported by setting `max_background_flushes=0`. Instead, users can achieve this by configuring their high-pri thread pool to have zero threads.
+* Scheduling flushes and compactions in the same thread pool is no longer supported by setting `max_background_flushes=0`. Instead, users can achieve this by configuring their high-pri thread pool to have zero threads.
+* options.delayed_write_rate by default take the value of options.rate_limiter rate.
 
 ### New Features
 * Change ticker/histogram statistics implementations to use core-local storage. This improves aggregation speed compared to our previous thread-local approach, particularly for applications with many threads.

db/db_impl_open.cc

@@ -18,6 +18,7 @@
 #include "db/builder.h"
 #include "options/options_helper.h"
 #include "rocksdb/wal_filter.h"
+#include "util/rate_limiter.h"
 #include "util/sst_file_manager_impl.h"
 #include "util/sync_point.h"
 
@@ -72,6 +73,15 @@ DBOptions SanitizeOptions(const std::string& dbname, const DBOptions& src) {
     }
   }
 
+  if (result.delayed_write_rate == 0) {
+    if (result.rate_limiter.get() != nullptr) {
+      result.delayed_write_rate = result.rate_limiter->GetBytesPerSecond();
+    }
+    if (result.delayed_write_rate == 0) {
+      result.delayed_write_rate = 16 * 1024 * 1024;
+    }
+  }
+
   if (result.WAL_ttl_seconds > 0 || result.WAL_size_limit_MB > 0) {
     result.recycle_log_file_num = false;
   }

db/db_options_test.cc

@@ -19,6 +19,7 @@
 #include "port/stack_trace.h"
 #include "rocksdb/cache.h"
 #include "rocksdb/convenience.h"
+#include "rocksdb/rate_limiter.h"
 #include "util/random.h"
 #include "util/sync_point.h"
 #include "util/testutil.h"
@@ -404,6 +405,17 @@ TEST_F(DBOptionsTest, MaxOpenFilesChange) {
   Close();
 }
 
+TEST_F(DBOptionsTest, SanitizeDelayedWriteRate) {
+  Options options;
+  options.delayed_write_rate = 0;
+  Reopen(options);
+  ASSERT_EQ(16 * 1024 * 1024, dbfull()->GetDBOptions().delayed_write_rate);
+
+  options.rate_limiter.reset(NewGenericRateLimiter(31 * 1024 * 1024));
+  Reopen(options);
+  ASSERT_EQ(31 * 1024 * 1024, dbfull()->GetDBOptions().delayed_write_rate);
+}
+
 #endif  // ROCKSDB_LITE
 
 }  // namespace rocksdb

include/rocksdb/options.h

@@ -746,10 +746,15 @@ struct DBOptions {
   // calculated using size of user write requests before compression.
   // RocksDB may decide to slow down more if the compaction still
   // gets behind further.
+  // If the value is 0, we will infer a value from `rater_limiter` value
+  // if it is not empty, or 16MB if `rater_limiter` is empty. Note that
+  // if users change the rate in `rate_limiter` after DB is opened,
+  // `delayed_write_rate` won't be adjusted.
+  //
   // Unit: byte per second.
   //
-  // Default: 16MB/s
-  uint64_t delayed_write_rate = 16 * 1024U * 1024U;
+  // Default: 0
+  uint64_t delayed_write_rate = 0;
 
   // By default, a single write thread queue is maintained. The thread gets
   // to the head of the queue becomes write batch group leader and responsible

include/rocksdb/rate_limiter.h

@@ -53,6 +53,8 @@ class RateLimiter {
   // Total # of requests that go though rate limiter
   virtual int64_t GetTotalRequests(
       const Env::IOPriority pri = Env::IO_TOTAL) const = 0;
+
+  virtual int64_t GetBytesPerSecond() const = 0;
 };
 
 // Create a RateLimiter object, which can be shared among RocksDB instances to

options/options.cc

@@ -486,8 +486,10 @@ DBOptions* DBOptions::OldDefaults(int rocksdb_major_version,
   if (rocksdb_major_version < 5 ||
       (rocksdb_major_version == 5 && rocksdb_minor_version < 2)) {
     delayed_write_rate = 2 * 1024U * 1024U;
+  } else if (rocksdb_major_version < 5 ||
+             (rocksdb_major_version == 5 && rocksdb_minor_version < 6)) {
+    delayed_write_rate = 16 * 1024U * 1024U;
   }
-
   max_open_files = 5000;
   base_background_compactions = -1;
   wal_recovery_mode = WALRecoveryMode::kTolerateCorruptedTailRecords;

util/rate_limiter.cc

@@ -32,6 +32,7 @@ GenericRateLimiter::GenericRateLimiter(int64_t rate_bytes_per_sec,
                                        int64_t refill_period_us,
                                        int32_t fairness)
     : refill_period_us_(refill_period_us),
+      rate_bytes_per_sec_(rate_bytes_per_sec),
       refill_bytes_per_period_(
           CalculateRefillBytesPerPeriod(rate_bytes_per_sec)),
       env_(Env::Default()),
@@ -68,6 +69,7 @@ GenericRateLimiter::~GenericRateLimiter() {
 // This API allows user to dynamically change rate limiter's bytes per second.
 void GenericRateLimiter::SetBytesPerSecond(int64_t bytes_per_second) {
   assert(bytes_per_second > 0);
+  rate_bytes_per_sec_ = bytes_per_second;
   refill_bytes_per_period_.store(
       CalculateRefillBytesPerPeriod(bytes_per_second),
       std::memory_order_relaxed);

util/rate_limiter.h

@@ -62,6 +62,10 @@ class GenericRateLimiter : public RateLimiter {
     return total_requests_[pri];
   }
 
+  virtual int64_t GetBytesPerSecond() const override {
+    return rate_bytes_per_sec_;
+  }
+
  private:
   void Refill();
   int64_t CalculateRefillBytesPerPeriod(int64_t rate_bytes_per_sec);
@@ -75,6 +79,8 @@ class GenericRateLimiter : public RateLimiter {
   const int64_t kMinRefillBytesPerPeriod = 100;
 
   const int64_t refill_period_us_;
+
+  int64_t rate_bytes_per_sec_;
   // This variable can be changed dynamically.
   std::atomic<int64_t> refill_bytes_per_period_;
   Env* const env_;