TransactionLogIter should stall at the last record. Currently it errors out

Summary:
* Add a method to check if the log reader is at EOF.
* If we know a record has been flushed force the log_reader to believe it is not at EOF, using a new method UnMarkEof().

This does not work with MMpaed files.

Test Plan: added a unit test.

Reviewers: dhruba, heyongqiang

Reviewed By: heyongqiang

CC: leveldb

Differential Revision: https://reviews.facebook.net/D9567

db/db_test.cc

@@ -517,6 +517,26 @@ class DBTest {
     }
     return result;
   }
+
+  Options OptionsForLogIterTest() {
+    Options options = CurrentOptions();
+    options.create_if_missing = true;
+    options.WAL_ttl_seconds = 1000;
+    return options;
+  }
+
+  std::unique_ptr<TransactionLogIterator> OpenTransactionLogIter(
+    const SequenceNumber seq) {
+    unique_ptr<TransactionLogIterator> iter;
+    Status status = dbfull()->GetUpdatesSince(seq, &iter);
+    ASSERT_TRUE(status.ok());
+    ASSERT_TRUE(iter->Valid());
+    return std::move(iter);
+  }
+
+  std::string DummyString(size_t len, char c = 'a') {
+    return std::string(len, c);
+  }
 };
 
 TEST(DBTest, Empty) {
@@ -2591,21 +2611,9 @@ TEST(DBTest, WALArchival) {
 
 }
 
-TEST(DBTest, TransactionLogIterator) {
-  std::string value(1024, '1');
-  Options options = CurrentOptions();
-  options.create_if_missing = true;
-  options.WAL_ttl_seconds = 1000;
-  DestroyAndReopen(&options);
-  Put("key1", value);
-  Put("key2", value);
-  Put("key2", value);
-  ASSERT_EQ(dbfull()->GetLatestSequenceNumber(), 3U);
-  {
-    unique_ptr<TransactionLogIterator> iter;
-    Status status = dbfull()->GetUpdatesSince(0, &iter);
-    ASSERT_TRUE(status.ok());
-    ASSERT_TRUE(iter->Valid());
+void ExpectRecords(
+  const int expected_no_records,
+  std::unique_ptr<TransactionLogIterator>& iter) {
   int i = 0;
   SequenceNumber lastSequence = 0;
   while (iter->Valid()) {
@@ -2616,63 +2624,61 @@ TEST(DBTest, TransactionLogIterator) {
     ASSERT_TRUE(iter->status().ok());
     iter->Next();
   }
-    ASSERT_EQ(i, 3);
+  ASSERT_EQ(i, expected_no_records);
+}
+
+TEST(DBTest, TransactionLogIterator) {
+  Options options = OptionsForLogIterTest();
+  DestroyAndReopen(&options);
+  Put("key1", DummyString(1024));
+  Put("key2", DummyString(1024));
+  Put("key2", DummyString(1024));
+  ASSERT_EQ(dbfull()->GetLatestSequenceNumber(), 3U);
+  {
+    auto iter = OpenTransactionLogIter(0);
+    ExpectRecords(3, iter);
   }
   Reopen(&options);
   {
-    Put("key4", value);
-    Put("key5", value);
-    Put("key6", value);
+    Put("key4", DummyString(1024));
+    Put("key5", DummyString(1024));
+    Put("key6", DummyString(1024));
   }
   {
-    unique_ptr<TransactionLogIterator> iter;
-    Status status = dbfull()->GetUpdatesSince(0, &iter);
-    ASSERT_TRUE(status.ok());
-    ASSERT_TRUE(iter->Valid());
-    int i = 0;
-    SequenceNumber lastSequence = 0;
-    while (iter->Valid()) {
-      BatchResult res = iter->GetBatch();
-      ASSERT_TRUE(res.sequence > lastSequence);
-      lastSequence = res.sequence;
-      ASSERT_TRUE(iter->status().ok());
-      iter->Next();
-      ++i;
-    }
-    ASSERT_EQ(i, 6);
+    auto iter = OpenTransactionLogIter(0);
+    ExpectRecords(6, iter);
   }
 }
 
 TEST(DBTest, TransactionLogIteratorMoveOverZeroFiles) {
-  std::string value(1024, '1');
-  Options options = CurrentOptions();
-  options.create_if_missing = true;
-  options.WAL_ttl_seconds = 1000;
+  Options options = OptionsForLogIterTest();
   DestroyAndReopen(&options);
   // Do a plain Reopen.
-  Put("key1", value);
+  Put("key1", DummyString(1024));
   // Two reopens should create a zero record WAL file.
   Reopen(&options);
   Reopen(&options);
 
-  Put("key2", value);
-  unique_ptr<TransactionLogIterator> iter;
-  Status status = dbfull()->GetUpdatesSince(0, &iter);
-  ASSERT_TRUE(status.ok());
+  Put("key2", DummyString(1024));
+
+  auto iter = OpenTransactionLogIter(0);
+  ExpectRecords(2, iter);
+}
+
+TEST(DBTest, TransactionLogIteratorStallAtLastRecord) {
+  Options options = OptionsForLogIterTest();
+  DestroyAndReopen(&options);
+  Put("key1", DummyString(1024));
+  auto iter = OpenTransactionLogIter(0);
+  ASSERT_OK(iter->status());
   ASSERT_TRUE(iter->Valid());
-  ASSERT_TRUE(status.ok());
-  ASSERT_TRUE(iter->Valid());
-  int i = 0;
-  SequenceNumber lastSequence = 0;
-  while (iter->Valid()) {
-    BatchResult res = iter->GetBatch();
-    ASSERT_TRUE(res.sequence > lastSequence);
-    lastSequence = res.sequence;
-    ASSERT_TRUE(iter->status().ok());
   iter->Next();
-    ++i;
-  }
-  ASSERT_EQ(i, 2);
+  ASSERT_TRUE(!iter->Valid());
+  ASSERT_OK(iter->status());
+  Put("key2", DummyString(1024));
+  iter->Next();
+  ASSERT_OK(iter->status());
+  ASSERT_TRUE(iter->Valid());
 }
 
 TEST(DBTest, ReadCompaction) {

db/log_reader.h

@@ -59,6 +59,17 @@ class Reader {
   // Undefined before the first call to ReadRecord.
   uint64_t LastRecordOffset();
 
+  // returns true if the reader has encountered an eof condition.
+  bool IsEOF() {
+    return eof_;
+  }
+
+  // when we know more data has been written to the file. we can use this
+  // function to force the reader to look again in the file.
+  void UnmarkEOF() {
+    eof_ = false;
+  }
+
   SequentialFile* file() { return file_.get(); }
 
  private:

db/transaction_log_iterator_impl.cc

@@ -1,6 +1,7 @@
 #include "db/transaction_log_iterator_impl.h"
 #include "db/write_batch_internal.h"
 #include "db/filename.h"
+
 namespace leveldb {
 
 TransactionLogIteratorImpl::TransactionLogIteratorImpl(
@@ -103,7 +104,7 @@ void TransactionLogIteratorImpl::Next() {
       }
       UpdateCurrentWriteBatch(record);
       if (currentSequence_ >= sequenceNumber_) {
-        assert(currentSequence_ < *lastFlushedSequence_);
+        assert(currentSequence_ <= *lastFlushedSequence_);
         isValid_ = true;
         currentLogReader_ = std::move(reader);
         break;
@@ -120,15 +121,11 @@ void TransactionLogIteratorImpl::Next() {
 LOOK_NEXT_FILE:
     assert(currentLogReader_);
     bool openNextFile = true;
-
-    if (currentSequence_ == *lastFlushedSequence_) {
-      // The last update has been read. and next is being called.
-      isValid_ = false;
-      currentStatus_ = Status::OK();
+    if (currentSequence_ < *lastFlushedSequence_) {
+      if (currentLogReader_->IsEOF()) {
+        currentLogReader_->UnmarkEOF();
       }
-
-    while (currentLogReader_->ReadRecord(&record, &scratch) &&
-           currentSequence_ < *lastFlushedSequence_) {
+      while (currentLogReader_->ReadRecord(&record, &scratch)) {
         if (record.size() < 12) {
           reporter.Corruption(
             record.size(), Status::Corruption("log record too small"));
@@ -139,6 +136,7 @@ LOOK_NEXT_FILE:
           break;
         }
       }
+    }
 
     if (openNextFile) {
       if (currentFileIndex_ < files_->size() - 1) {
@@ -154,6 +152,10 @@ LOOK_NEXT_FILE:
         currentLogReader_.reset(
           new log::Reader(std::move(file), &reporter, true, 0));
         goto LOOK_NEXT_FILE;
+      } else if (currentSequence_ == *lastFlushedSequence_) {
+        // The last update has been read. and next is being called.
+        isValid_ = false;
+        currentStatus_ = Status::OK();
       } else {
         //  LOOKED AT FILES. WE ARE DONE HERE.
         isValid_ = false;
@@ -169,6 +171,7 @@ void TransactionLogIteratorImpl::UpdateCurrentWriteBatch(const Slice& record) {
   WriteBatchInternal::SetContents(batch, record);
   currentSequence_ = WriteBatchInternal::Sequence(batch);
   currentBatch_.reset(batch);
+  isValid_ = true;
 }
 
 }  //  namespace leveldb