Merge commit 'adae78ea9940f4d44382967d1296e7db0b54a4de' into leveldb-squashed

6 files changed, 122 insertions, 30 deletions

diff --git a/src/leveldb/db/db_impl.cc b/src/leveldb/db/db_impl.cc
index c9de169f29..af02467b33 100644
--- a/src/leveldb/db/db_impl.cc
+++ b/src/leveldb/db/db_impl.cc
@@ -35,6 +35,8 @@
 
 namespace leveldb {
 
+const int kNumNonTableCacheFiles = 10;
+
 // Information kept for every waiting writer
 struct DBImpl::Writer {
   Status status;
@@ -92,9 +94,9 @@ Options SanitizeOptions(const std::string& dbname,
   Options result = src;
   result.comparator = icmp;
   result.filter_policy = (src.filter_policy != NULL) ? ipolicy : NULL;
-  ClipToRange(&result.max_open_files,            20,     50000);
-  ClipToRange(&result.write_buffer_size,         64<<10, 1<<30);
-  ClipToRange(&result.block_size,                1<<10,  4<<20);
+  ClipToRange(&result.max_open_files,    64 + kNumNonTableCacheFiles, 50000);
+  ClipToRange(&result.write_buffer_size, 64<<10,                      1<<30);
+  ClipToRange(&result.block_size,        1<<10,                       4<<20);
   if (result.info_log == NULL) {
     // Open a log file in the same directory as the db
     src.env->CreateDir(dbname);  // In case it does not exist
@@ -130,12 +132,13 @@ DBImpl::DBImpl(const Options& options, const std::string& dbname)
       log_(NULL),
       tmp_batch_(new WriteBatch),
       bg_compaction_scheduled_(false),
-      manual_compaction_(NULL) {
+      manual_compaction_(NULL),
+      consecutive_compaction_errors_(0) {
   mem_->Ref();
   has_imm_.Release_Store(NULL);
 
   // Reserve ten files or so for other uses and give the rest to TableCache.
-  const int table_cache_size = options.max_open_files - 10;
+  const int table_cache_size = options.max_open_files - kNumNonTableCacheFiles;
   table_cache_ = new TableCache(dbname_, &options_, table_cache_size);
 
   versions_ = new VersionSet(dbname_, &options_, table_cache_,
@@ -310,16 +313,24 @@ Status DBImpl::Recover(VersionEdit* edit) {
     if (!s.ok()) {
       return s;
     }
+    std::set<uint64_t> expected;
+    versions_->AddLiveFiles(&expected);
     uint64_t number;
     FileType type;
     std::vector<uint64_t> logs;
     for (size_t i = 0; i < filenames.size(); i++) {
-      if (ParseFileName(filenames[i], &number, &type)
-          && type == kLogFile
-          && ((number >= min_log) || (number == prev_log))) {
+      if (ParseFileName(filenames[i], &number, &type)) {
+        expected.erase(number);
+        if (type == kLogFile && ((number >= min_log) || (number == prev_log)))
         logs.push_back(number);
       }
     }
+    if (!expected.empty()) {
+      char buf[50];
+      snprintf(buf, sizeof(buf), "%d missing files; e.g.",
+               static_cast<int>(expected.size()));
+      return Status::Corruption(buf, TableFileName(dbname_, *(expected.begin())));
+    }
 
     // Recover in the order in which the logs were generated
     std::sort(logs.begin(), logs.end());
@@ -611,6 +622,7 @@ void DBImpl::BackgroundCall() {
     Status s = BackgroundCompaction();
     if (s.ok()) {
       // Success
+      consecutive_compaction_errors_ = 0;
     } else if (shutting_down_.Acquire_Load()) {
       // Error most likely due to shutdown; do not wait
     } else {
@@ -622,7 +634,12 @@ void DBImpl::BackgroundCall() {
       Log(options_.info_log, "Waiting after background compaction error: %s",
           s.ToString().c_str());
       mutex_.Unlock();
-      env_->SleepForMicroseconds(1000000);
+      ++consecutive_compaction_errors_;
+      int seconds_to_sleep = 1;
+      for (int i = 0; i < 3 && i < consecutive_compaction_errors_ - 1; ++i) {
+        seconds_to_sleep *= 2;
+      }
+      env_->SleepForMicroseconds(seconds_to_sleep * 1000000);
       mutex_.Lock();
     }
   }
@@ -805,6 +822,9 @@ Status DBImpl::FinishCompactionOutputFile(CompactionState* compact,
           (unsigned long long) output_number,
           (unsigned long long) current_entries,
           (unsigned long long) current_bytes);
+
+      // rate-limit compaction file creation with a 100ms pause
+      env_->SleepForMicroseconds(100000);
     }
   }
   return s;
@@ -1268,10 +1288,11 @@ Status DBImpl::MakeRoomForWrite(bool force) {
     } else if (imm_ != NULL) {
       // We have filled up the current memtable, but the previous
       // one is still being compacted, so we wait.
+      Log(options_.info_log, "Current memtable full; waiting...\n");
       bg_cv_.Wait();
     } else if (versions_->NumLevelFiles(0) >= config::kL0_StopWritesTrigger) {
       // There are too many level-0 files.
-      Log(options_.info_log, "waiting...\n");
+      Log(options_.info_log, "Too many L0 files; waiting...\n");
       bg_cv_.Wait();
     } else {
       // Attempt to switch to a new memtable and trigger compaction of old
diff --git a/src/leveldb/db/db_impl.h b/src/leveldb/db/db_impl.h
index bd29dd8055..3c8d711ae0 100644
--- a/src/leveldb/db/db_impl.h
+++ b/src/leveldb/db/db_impl.h
@@ -163,6 +163,7 @@ class DBImpl : public DB {
 
   // Have we encountered a background error in paranoid mode?
   Status bg_error_;
+  int consecutive_compaction_errors_;
 
   // Per level compaction stats.  stats_[level] stores the stats for
   // compactions that produced data for the specified "level".
diff --git a/src/leveldb/db/db_test.cc b/src/leveldb/db/db_test.cc
index 684ea3bdbc..49aae04dbd 100644
--- a/src/leveldb/db/db_test.cc
+++ b/src/leveldb/db/db_test.cc
@@ -33,8 +33,11 @@ class AtomicCounter {
  public:
   AtomicCounter() : count_(0) { }
   void Increment() {
+    IncrementBy(1);
+  }
+  void IncrementBy(int count) {
     MutexLock l(&mu_);
-    count_++;
+    count_ += count;
   }
   int Read() {
     MutexLock l(&mu_);
@@ -45,6 +48,10 @@ class AtomicCounter {
     count_ = 0;
   }
 };
+
+void DelayMilliseconds(int millis) {
+  Env::Default()->SleepForMicroseconds(millis * 1000);
+}
 }
 
 // Special Env used to delay background operations
@@ -69,6 +76,7 @@ class SpecialEnv : public EnvWrapper {
   AtomicCounter random_read_counter_;
 
   AtomicCounter sleep_counter_;
+  AtomicCounter sleep_time_counter_;
 
   explicit SpecialEnv(Env* base) : EnvWrapper(base) {
     delay_sstable_sync_.Release_Store(NULL);
@@ -103,7 +111,7 @@ class SpecialEnv : public EnvWrapper {
       Status Flush() { return base_->Flush(); }
       Status Sync() {
         while (env_->delay_sstable_sync_.Acquire_Load() != NULL) {
-          env_->SleepForMicroseconds(100000);
+          DelayMilliseconds(100);
         }
         return base_->Sync();
       }
@@ -174,8 +182,9 @@ class SpecialEnv : public EnvWrapper {
 
   virtual void SleepForMicroseconds(int micros) {
     sleep_counter_.Increment();
-    target()->SleepForMicroseconds(micros);
+    sleep_time_counter_.IncrementBy(micros);
   }
+
 };
 
 class DBTest {
@@ -461,6 +470,20 @@ class DBTest {
     }
     return result;
   }
+
+  bool DeleteAnSSTFile() {
+    std::vector<std::string> filenames;
+    ASSERT_OK(env_->GetChildren(dbname_, &filenames));
+    uint64_t number;
+    FileType type;
+    for (size_t i = 0; i < filenames.size(); i++) {
+      if (ParseFileName(filenames[i], &number, &type) && type == kTableFile) {
+        ASSERT_OK(env_->DeleteFile(TableFileName(dbname_, number)));
+        return true;
+      }
+    }
+    return false;
+  }
 };
 
 TEST(DBTest, Empty) {
@@ -611,7 +634,7 @@ TEST(DBTest, GetEncountersEmptyLevel) {
     }
 
     // Step 4: Wait for compaction to finish
-    env_->SleepForMicroseconds(1000000);
+    DelayMilliseconds(1000);
 
     ASSERT_EQ(NumTableFilesAtLevel(0), 0);
   } while (ChangeOptions());
@@ -1295,7 +1318,7 @@ TEST(DBTest, L0_CompactionBug_Issue44_a) {
   Reopen();
   Reopen();
   ASSERT_EQ("(a->v)", Contents());
-  env_->SleepForMicroseconds(1000000);  // Wait for compaction to finish
+  DelayMilliseconds(1000);  // Wait for compaction to finish
   ASSERT_EQ("(a->v)", Contents());
 }
 
@@ -1311,7 +1334,7 @@ TEST(DBTest, L0_CompactionBug_Issue44_b) {
   Put("","");
   Reopen();
   Put("","");
-  env_->SleepForMicroseconds(1000000);  // Wait for compaction to finish
+  DelayMilliseconds(1000);  // Wait for compaction to finish
   Reopen();
   Put("d","dv");
   Reopen();
@@ -1321,7 +1344,7 @@ TEST(DBTest, L0_CompactionBug_Issue44_b) {
   Delete("b");
   Reopen();
   ASSERT_EQ("(->)(c->cv)", Contents());
-  env_->SleepForMicroseconds(1000000);  // Wait for compaction to finish
+  DelayMilliseconds(1000);  // Wait for compaction to finish
   ASSERT_EQ("(->)(c->cv)", Contents());
 }
 
@@ -1506,6 +1529,30 @@ TEST(DBTest, NoSpace) {
   ASSERT_GE(env_->sleep_counter_.Read(), 5);
 }
 
+TEST(DBTest, ExponentialBackoff) {
+  Options options = CurrentOptions();
+  options.env = env_;
+  Reopen(&options);
+
+  ASSERT_OK(Put("foo", "v1"));
+  ASSERT_EQ("v1", Get("foo"));
+  Compact("a", "z");
+  env_->non_writable_.Release_Store(env_);  // Force errors for new files
+  env_->sleep_counter_.Reset();
+  env_->sleep_time_counter_.Reset();
+  for (int i = 0; i < 5; i++) {
+    dbfull()->TEST_CompactRange(2, NULL, NULL);
+  }
+  env_->non_writable_.Release_Store(NULL);
+
+  // Wait for compaction to finish
+  DelayMilliseconds(1000);
+
+  ASSERT_GE(env_->sleep_counter_.Read(), 5);
+  ASSERT_LT(env_->sleep_counter_.Read(), 10);
+  ASSERT_GE(env_->sleep_time_counter_.Read(), 10e6);
+}
+
 TEST(DBTest, NonWritableFileSystem) {
   Options options = CurrentOptions();
   options.write_buffer_size = 1000;
@@ -1519,7 +1566,7 @@ TEST(DBTest, NonWritableFileSystem) {
     fprintf(stderr, "iter %d; errors %d\n", i, errors);
     if (!Put("foo", big).ok()) {
       errors++;
-      env_->SleepForMicroseconds(100000);
+      DelayMilliseconds(100);
     }
   }
   ASSERT_GT(errors, 0);
@@ -1567,6 +1614,24 @@ TEST(DBTest, ManifestWriteError) {
   }
 }
 
+TEST(DBTest, MissingSSTFile) {
+  ASSERT_OK(Put("foo", "bar"));
+  ASSERT_EQ("bar", Get("foo"));
+
+  // Dump the memtable to disk.
+  dbfull()->TEST_CompactMemTable();
+  ASSERT_EQ("bar", Get("foo"));
+
+  Close();
+  ASSERT_TRUE(DeleteAnSSTFile());
+  Options options = CurrentOptions();
+  options.paranoid_checks = true;
+  Status s = TryReopen(&options);
+  ASSERT_TRUE(!s.ok());
+  ASSERT_TRUE(s.ToString().find("issing") != std::string::npos)
+      << s.ToString();
+}
+
 TEST(DBTest, FilesDeletedAfterCompaction) {
   ASSERT_OK(Put("foo", "v2"));
   Compact("a", "z");
@@ -1711,13 +1776,13 @@ TEST(DBTest, MultiThreaded) {
     }
 
     // Let them run for a while
-    env_->SleepForMicroseconds(kTestSeconds * 1000000);
+    DelayMilliseconds(kTestSeconds * 1000);
 
     // Stop the threads and wait for them to finish
     mt.stop.Release_Store(&mt);
     for (int id = 0; id < kNumThreads; id++) {
       while (mt.thread_done[id].Acquire_Load() == NULL) {
-        env_->SleepForMicroseconds(100000);
+        DelayMilliseconds(100);
       }
     }
   } while (ChangeOptions());
diff --git a/src/leveldb/db/dbformat.cc b/src/leveldb/db/dbformat.cc
index 28e11b398d..20a7ca4462 100644
--- a/src/leveldb/db/dbformat.cc
+++ b/src/leveldb/db/dbformat.cc
@@ -26,7 +26,7 @@ std::string ParsedInternalKey::DebugString() const {
            (unsigned long long) sequence,
            int(type));
   std::string result = "'";
-  result += user_key.ToString();
+  result += EscapeString(user_key.ToString());
   result += buf;
   return result;
 }
diff --git a/src/leveldb/db/filename_test.cc b/src/leveldb/db/filename_test.cc
index 47353d6c9a..5a26da4728 100644
--- a/src/leveldb/db/filename_test.cc
+++ b/src/leveldb/db/filename_test.cc
@@ -70,7 +70,7 @@ TEST(FileNameTest, Parse) {
   for (int i = 0; i < sizeof(errors) / sizeof(errors[0]); i++) {
     std::string f = errors[i];
     ASSERT_TRUE(!ParseFileName(f, &number, &type)) << f;
-  };
+  }
 }
 
 TEST(FileNameTest, Construction) {
diff --git a/src/leveldb/db/version_set.cc b/src/leveldb/db/version_set.cc
index 7d0a5de2b9..4fd1ddef21 100644
--- a/src/leveldb/db/version_set.cc
+++ b/src/leveldb/db/version_set.cc
@@ -1331,14 +1331,19 @@ Compaction* VersionSet::CompactRange(
   }
 
   // Avoid compacting too much in one shot in case the range is large.
-  const uint64_t limit = MaxFileSizeForLevel(level);
-  uint64_t total = 0;
-  for (size_t i = 0; i < inputs.size(); i++) {
-    uint64_t s = inputs[i]->file_size;
-    total += s;
-    if (total >= limit) {
-      inputs.resize(i + 1);
-      break;
+  // But we cannot do this for level-0 since level-0 files can overlap
+  // and we must not pick one file and drop another older file if the
+  // two files overlap.
+  if (level > 0) {
+    const uint64_t limit = MaxFileSizeForLevel(level);
+    uint64_t total = 0;
+    for (size_t i = 0; i < inputs.size(); i++) {
+      uint64_t s = inputs[i]->file_size;
+      total += s;
+      if (total >= limit) {
+        inputs.resize(i + 1);
+        break;
+      }
     }
   }