rocksdb/test_util/testutil.h

816 lines
26 KiB
C
Raw Normal View History

// Copyright (c) 2011-present, Facebook, Inc. All rights reserved.
// This source code is licensed under both the GPLv2 (found in the
// COPYING file in the root directory) and Apache 2.0 License
// (found in the LICENSE.Apache file in the root directory).
//
// Copyright (c) 2011 The LevelDB Authors. All rights reserved.
// Use of this source code is governed by a BSD-style license that can be
// found in the LICENSE file. See the AUTHORS file for names of contributors.
#pragma once
#include <algorithm>
#include <deque>
#include <string>
#include <vector>
Introduce a new storage specific Env API (#5761) Summary: The current Env API encompasses both storage/file operations, as well as OS related operations. Most of the APIs return a Status, which does not have enough metadata about an error, such as whether its retry-able or not, scope (i.e fault domain) of the error etc., that may be required in order to properly handle a storage error. The file APIs also do not provide enough control over the IO SLA, such as timeout, prioritization, hinting about placement and redundancy etc. This PR separates out the file/storage APIs from Env into a new FileSystem class. The APIs are updated to return an IOStatus with metadata about the error, as well as to take an IOOptions structure as input in order to allow more control over the IO. The user can set both ```options.env``` and ```options.file_system``` to specify that RocksDB should use the former for OS related operations and the latter for storage operations. Internally, a ```CompositeEnvWrapper``` has been introduced that inherits from ```Env``` and redirects individual methods to either an ```Env``` implementation or the ```FileSystem``` as appropriate. When options are sanitized during ```DB::Open```, ```options.env``` is replaced with a newly allocated ```CompositeEnvWrapper``` instance if both env and file_system have been specified. This way, the rest of the RocksDB code can continue to function as before. This PR also ports PosixEnv to the new API by splitting it into two - PosixEnv and PosixFileSystem. PosixEnv is defined as a sub-class of CompositeEnvWrapper, and threading/time functions are overridden with Posix specific implementations in order to avoid an extra level of indirection. The ```CompositeEnvWrapper``` translates ```IOStatus``` return code to ```Status```, and sets the severity to ```kSoftError``` if the io_status is retryable. The error handling code in RocksDB can then recover the DB automatically. Pull Request resolved: https://github.com/facebook/rocksdb/pull/5761 Differential Revision: D18868376 Pulled By: anand1976 fbshipit-source-id: 39efe18a162ea746fabac6360ff529baba48486f
2019-12-13 23:47:08 +01:00
#include "env/composite_env_wrapper.h"
#include "file/writable_file_writer.h"
#include "rocksdb/compaction_filter.h"
#include "rocksdb/env.h"
#include "rocksdb/iterator.h"
#include "rocksdb/merge_operator.h"
#include "rocksdb/options.h"
#include "rocksdb/slice.h"
#include "rocksdb/table.h"
#include "table/block_based/block_based_table_factory.h"
#include "table/internal_iterator.h"
#include "table/plain/plain_table_factory.h"
LogAndApply() should fail if the column family has been dropped Summary: This patch finally fixes the ColumnFamilyTest.ReadDroppedColumnFamily test. The test has been failing very sporadically and it was hard to repro. However, I managed to write a new tests that reproes the failure deterministically. Here's what happens: 1. We start the flush for the column family 2. We check if the column family was dropped here: https://github.com/facebook/rocksdb/blob/a3fc49bfddcdb1ff29409aacd06c04df56c7a1d7/db/flush_job.cc#L149 3. This check goes through, ends up in InstallMemtableFlushResults() and it goes into LogAndApply() 4. At about this time, we start dropping the column family. Dropping the column family process gets to LogAndApply() at about the same time as LogAndApply() from flush process 5. Drop column family goes through LogAndApply() first, marking the column family as dropped. 6. Flush process gets woken up and gets a chance to write to the MANIFEST. However, this is where it gets stuck: https://github.com/facebook/rocksdb/blob/a3fc49bfddcdb1ff29409aacd06c04df56c7a1d7/db/version_set.cc#L1975 7. We see that the column family was dropped, so there is no need to write to the MANIFEST. We return OK. 8. Flush gets OK back from LogAndApply() and it deletes the memtable, thinking that the data is now safely persisted to sst file. The fix is pretty simple. Instead of OK, we return ShutdownInProgress. This is not really true, but we have been using this status code to also mean "this operation was canceled because the column family has been dropped". The fix is only one LOC. All other code is related to tests. I added a new test that reproes the failure. I also moved SleepingBackgroundTask to util/testutil.h (because I needed it in column_family_test for my new test). There's plenty of other places where we reimplement SleepingBackgroundTask, but I'll address that in a separate commit. Test Plan: 1. new test 2. make check 3. Make sure the ColumnFamilyTest.ReadDroppedColumnFamily doesn't fail on Travis: https://travis-ci.org/facebook/rocksdb/jobs/79952386 Reviewers: yhchiang, anthony, IslamAbdelRahman, kradhakrishnan, rven, sdong Reviewed By: sdong Subscribers: dhruba, leveldb Differential Revision: https://reviews.facebook.net/D46773
2015-09-15 20:28:44 +02:00
#include "util/mutexlock.h"
#include "util/random.h"
namespace ROCKSDB_NAMESPACE {
class SequentialFile;
class SequentialFileReader;
namespace test {
extern const uint32_t kDefaultFormatVersion;
extern const uint32_t kLatestFormatVersion;
// Store in *dst a random string of length "len" and return a Slice that
// references the generated data.
extern Slice RandomString(Random* rnd, int len, std::string* dst);
extern std::string RandomHumanReadableString(Random* rnd, int len);
// Return a random key with the specified length that may contain interesting
// characters (e.g. \x00, \xff, etc.).
enum RandomKeyType : char { RANDOM, LARGEST, SMALLEST, MIDDLE };
extern std::string RandomKey(Random* rnd, int len,
RandomKeyType type = RandomKeyType::RANDOM);
// Store in *dst a string of length "len" that will compress to
// "N*compressed_fraction" bytes and return a Slice that references
// the generated data.
extern Slice CompressibleString(Random* rnd, double compressed_fraction,
int len, std::string* dst);
// A wrapper that allows injection of errors.
class ErrorEnv : public EnvWrapper {
public:
bool writable_file_error_;
int num_writable_file_errors_;
ErrorEnv() : EnvWrapper(Env::Default()),
writable_file_error_(false),
num_writable_file_errors_(0) { }
virtual Status NewWritableFile(const std::string& fname,
std::unique_ptr<WritableFile>* result,
const EnvOptions& soptions) override {
result->reset();
if (writable_file_error_) {
++num_writable_file_errors_;
return Status::IOError(fname, "fake error");
}
return target()->NewWritableFile(fname, result, soptions);
}
};
#ifndef NDEBUG
// An internal comparator that just forward comparing results from the
// user comparator in it. Can be used to test entities that have no dependency
// on internal key structure but consumes InternalKeyComparator, like
// BlockBasedTable.
class PlainInternalKeyComparator : public InternalKeyComparator {
public:
explicit PlainInternalKeyComparator(const Comparator* c)
: InternalKeyComparator(c) {}
virtual ~PlainInternalKeyComparator() {}
virtual int Compare(const Slice& a, const Slice& b) const override {
return user_comparator()->Compare(a, b);
}
};
#endif
// A test comparator which compare two strings in this way:
// (1) first compare prefix of 8 bytes in alphabet order,
// (2) if two strings share the same prefix, sort the other part of the string
// in the reverse alphabet order.
// This helps simulate the case of compounded key of [entity][timestamp] and
// latest timestamp first.
class SimpleSuffixReverseComparator : public Comparator {
public:
SimpleSuffixReverseComparator() {}
virtual const char* Name() const override {
return "SimpleSuffixReverseComparator";
}
virtual int Compare(const Slice& a, const Slice& b) const override {
Slice prefix_a = Slice(a.data(), 8);
Slice prefix_b = Slice(b.data(), 8);
int prefix_comp = prefix_a.compare(prefix_b);
if (prefix_comp != 0) {
return prefix_comp;
} else {
Slice suffix_a = Slice(a.data() + 8, a.size() - 8);
Slice suffix_b = Slice(b.data() + 8, b.size() - 8);
return -(suffix_a.compare(suffix_b));
}
}
virtual void FindShortestSeparator(std::string* /*start*/,
const Slice& /*limit*/) const override {}
virtual void FindShortSuccessor(std::string* /*key*/) const override {}
};
// Returns a user key comparator that can be used for comparing two uint64_t
// slices. Instead of comparing slices byte-wise, it compares all the 8 bytes
// at once. Assumes same endian-ness is used though the database's lifetime.
// Symantics of comparison would differ from Bytewise comparator in little
// endian machines.
extern const Comparator* Uint64Comparator();
// Iterator over a vector of keys/values
class VectorIterator : public InternalIterator {
public:
explicit VectorIterator(const std::vector<std::string>& keys)
: keys_(keys), current_(keys.size()) {
std::sort(keys_.begin(), keys_.end());
values_.resize(keys.size());
}
VectorIterator(const std::vector<std::string>& keys,
const std::vector<std::string>& values)
: keys_(keys), values_(values), current_(keys.size()) {
assert(keys_.size() == values_.size());
}
virtual bool Valid() const override { return current_ < keys_.size(); }
virtual void SeekToFirst() override { current_ = 0; }
virtual void SeekToLast() override { current_ = keys_.size() - 1; }
virtual void Seek(const Slice& target) override {
current_ = std::lower_bound(keys_.begin(), keys_.end(), target.ToString()) -
keys_.begin();
}
virtual void SeekForPrev(const Slice& target) override {
current_ = std::upper_bound(keys_.begin(), keys_.end(), target.ToString()) -
keys_.begin();
if (!Valid()) {
SeekToLast();
} else {
Prev();
}
}
virtual void Next() override { current_++; }
virtual void Prev() override { current_--; }
virtual Slice key() const override { return Slice(keys_[current_]); }
virtual Slice value() const override { return Slice(values_[current_]); }
virtual Status status() const override { return Status::OK(); }
virtual bool IsKeyPinned() const override { return true; }
virtual bool IsValuePinned() const override { return true; }
private:
std::vector<std::string> keys_;
std::vector<std::string> values_;
size_t current_;
};
extern WritableFileWriter* GetWritableFileWriter(WritableFile* wf,
const std::string& fname);
extern RandomAccessFileReader* GetRandomAccessFileReader(RandomAccessFile* raf);
extern SequentialFileReader* GetSequentialFileReader(SequentialFile* se,
const std::string& fname);
class StringSink: public WritableFile {
public:
std::string contents_;
explicit StringSink(Slice* reader_contents = nullptr) :
WritableFile(),
contents_(""),
reader_contents_(reader_contents),
last_flush_(0) {
if (reader_contents_ != nullptr) {
*reader_contents_ = Slice(contents_.data(), 0);
}
}
const std::string& contents() const { return contents_; }
Refactor to support file_reader_writer on Windows. Summary. A change https://reviews.facebook.net/differential/diff/224721/ Has attempted to move common functionality out of platform dependent code to a new facility called file_reader_writer. This includes: - perf counters - Buffering - RateLimiting However, the change did not attempt to refactor Windows code. To mitigate, we introduce new quering interfaces such as UseOSBuffer(), GetRequiredBufferAlignment() and ReaderWriterForward() for pure forwarding where required. Introduce WritableFile got a new method Truncate(). This is to communicate to the file as to how much data it has on close. - When space is pre-allocated on Linux it is filled with zeros implicitly, no such thing exist on Windows so we must truncate file on close. - When operating in unbuffered mode the last page is filled with zeros but we still want to truncate. Previously, Close() would take care of it but now buffer management is shifted to the wrappers and the file has no idea about the file true size. This means that Close() on the wrapper level must always include Truncate() as well as wrapper __dtor should call Close() and against double Close(). Move buffered/unbuffered write logic to the wrapper. Utilize Aligned buffer class. Adjust tests and implement Truncate() where necessary. Come up with reasonable defaults for new virtual interfaces. Forward calls for RandomAccessReadAhead class to avoid double buffering and locking (double locking in unbuffered mode on WIndows).
2015-09-11 18:57:02 +02:00
virtual Status Truncate(uint64_t size) override {
contents_.resize(static_cast<size_t>(size));
Refactor to support file_reader_writer on Windows. Summary. A change https://reviews.facebook.net/differential/diff/224721/ Has attempted to move common functionality out of platform dependent code to a new facility called file_reader_writer. This includes: - perf counters - Buffering - RateLimiting However, the change did not attempt to refactor Windows code. To mitigate, we introduce new quering interfaces such as UseOSBuffer(), GetRequiredBufferAlignment() and ReaderWriterForward() for pure forwarding where required. Introduce WritableFile got a new method Truncate(). This is to communicate to the file as to how much data it has on close. - When space is pre-allocated on Linux it is filled with zeros implicitly, no such thing exist on Windows so we must truncate file on close. - When operating in unbuffered mode the last page is filled with zeros but we still want to truncate. Previously, Close() would take care of it but now buffer management is shifted to the wrappers and the file has no idea about the file true size. This means that Close() on the wrapper level must always include Truncate() as well as wrapper __dtor should call Close() and against double Close(). Move buffered/unbuffered write logic to the wrapper. Utilize Aligned buffer class. Adjust tests and implement Truncate() where necessary. Come up with reasonable defaults for new virtual interfaces. Forward calls for RandomAccessReadAhead class to avoid double buffering and locking (double locking in unbuffered mode on WIndows).
2015-09-11 18:57:02 +02:00
return Status::OK();
}
virtual Status Close() override { return Status::OK(); }
virtual Status Flush() override {
if (reader_contents_ != nullptr) {
assert(reader_contents_->size() <= last_flush_);
size_t offset = last_flush_ - reader_contents_->size();
*reader_contents_ = Slice(
contents_.data() + offset,
contents_.size() - offset);
last_flush_ = contents_.size();
}
return Status::OK();
}
virtual Status Sync() override { return Status::OK(); }
virtual Status Append(const Slice& slice) override {
contents_.append(slice.data(), slice.size());
return Status::OK();
}
void Drop(size_t bytes) {
if (reader_contents_ != nullptr) {
contents_.resize(contents_.size() - bytes);
*reader_contents_ = Slice(
reader_contents_->data(), reader_contents_->size() - bytes);
last_flush_ = contents_.size();
}
}
private:
Slice* reader_contents_;
size_t last_flush_;
};
// A wrapper around a StringSink to give it a RandomRWFile interface
class RandomRWStringSink : public RandomRWFile {
public:
explicit RandomRWStringSink(StringSink* ss) : ss_(ss) {}
Status Write(uint64_t offset, const Slice& data) override {
if (offset + data.size() > ss_->contents_.size()) {
ss_->contents_.resize(static_cast<size_t>(offset) + data.size(), '\0');
}
char* pos = const_cast<char*>(ss_->contents_.data() + offset);
memcpy(pos, data.data(), data.size());
return Status::OK();
}
Status Read(uint64_t offset, size_t n, Slice* result,
char* /*scratch*/) const override {
*result = Slice(nullptr, 0);
if (offset < ss_->contents_.size()) {
size_t str_res_sz =
std::min(static_cast<size_t>(ss_->contents_.size() - offset), n);
*result = Slice(ss_->contents_.data() + offset, str_res_sz);
}
return Status::OK();
}
Status Flush() override { return Status::OK(); }
Status Sync() override { return Status::OK(); }
Status Close() override { return Status::OK(); }
const std::string& contents() const { return ss_->contents(); }
private:
StringSink* ss_;
};
// Like StringSink, this writes into a string. Unlink StringSink, it
// has some initial content and overwrites it, just like a recycled
// log file.
class OverwritingStringSink : public WritableFile {
public:
explicit OverwritingStringSink(Slice* reader_contents)
: WritableFile(),
contents_(""),
reader_contents_(reader_contents),
last_flush_(0) {}
const std::string& contents() const { return contents_; }
virtual Status Truncate(uint64_t size) override {
contents_.resize(static_cast<size_t>(size));
return Status::OK();
}
virtual Status Close() override { return Status::OK(); }
virtual Status Flush() override {
if (last_flush_ < contents_.size()) {
assert(reader_contents_->size() >= contents_.size());
memcpy((char*)reader_contents_->data() + last_flush_,
contents_.data() + last_flush_, contents_.size() - last_flush_);
last_flush_ = contents_.size();
}
return Status::OK();
}
virtual Status Sync() override { return Status::OK(); }
virtual Status Append(const Slice& slice) override {
contents_.append(slice.data(), slice.size());
return Status::OK();
}
void Drop(size_t bytes) {
contents_.resize(contents_.size() - bytes);
if (last_flush_ > contents_.size()) last_flush_ = contents_.size();
}
private:
std::string contents_;
Slice* reader_contents_;
size_t last_flush_;
};
class StringSource: public RandomAccessFile {
public:
explicit StringSource(const Slice& contents, uint64_t uniq_id = 0,
bool mmap = false)
: contents_(contents.data(), contents.size()),
uniq_id_(uniq_id),
mmap_(mmap),
total_reads_(0) {}
virtual ~StringSource() { }
uint64_t Size() const { return contents_.size(); }
virtual Status Read(uint64_t offset, size_t n, Slice* result,
char* scratch) const override {
total_reads_++;
if (offset > contents_.size()) {
return Status::InvalidArgument("invalid Read offset");
}
if (offset + n > contents_.size()) {
n = contents_.size() - static_cast<size_t>(offset);
}
if (!mmap_) {
memcpy(scratch, &contents_[static_cast<size_t>(offset)], n);
*result = Slice(scratch, n);
} else {
*result = Slice(&contents_[static_cast<size_t>(offset)], n);
}
return Status::OK();
}
virtual size_t GetUniqueId(char* id, size_t max_size) const override {
if (max_size < 20) {
return 0;
}
char* rid = id;
rid = EncodeVarint64(rid, uniq_id_);
rid = EncodeVarint64(rid, 0);
return static_cast<size_t>(rid-id);
}
int total_reads() const { return total_reads_; }
void set_total_reads(int tr) { total_reads_ = tr; }
private:
std::string contents_;
uint64_t uniq_id_;
bool mmap_;
mutable int total_reads_;
};
Introduce a new storage specific Env API (#5761) Summary: The current Env API encompasses both storage/file operations, as well as OS related operations. Most of the APIs return a Status, which does not have enough metadata about an error, such as whether its retry-able or not, scope (i.e fault domain) of the error etc., that may be required in order to properly handle a storage error. The file APIs also do not provide enough control over the IO SLA, such as timeout, prioritization, hinting about placement and redundancy etc. This PR separates out the file/storage APIs from Env into a new FileSystem class. The APIs are updated to return an IOStatus with metadata about the error, as well as to take an IOOptions structure as input in order to allow more control over the IO. The user can set both ```options.env``` and ```options.file_system``` to specify that RocksDB should use the former for OS related operations and the latter for storage operations. Internally, a ```CompositeEnvWrapper``` has been introduced that inherits from ```Env``` and redirects individual methods to either an ```Env``` implementation or the ```FileSystem``` as appropriate. When options are sanitized during ```DB::Open```, ```options.env``` is replaced with a newly allocated ```CompositeEnvWrapper``` instance if both env and file_system have been specified. This way, the rest of the RocksDB code can continue to function as before. This PR also ports PosixEnv to the new API by splitting it into two - PosixEnv and PosixFileSystem. PosixEnv is defined as a sub-class of CompositeEnvWrapper, and threading/time functions are overridden with Posix specific implementations in order to avoid an extra level of indirection. The ```CompositeEnvWrapper``` translates ```IOStatus``` return code to ```Status```, and sets the severity to ```kSoftError``` if the io_status is retryable. The error handling code in RocksDB can then recover the DB automatically. Pull Request resolved: https://github.com/facebook/rocksdb/pull/5761 Differential Revision: D18868376 Pulled By: anand1976 fbshipit-source-id: 39efe18a162ea746fabac6360ff529baba48486f
2019-12-13 23:47:08 +01:00
inline StringSink* GetStringSinkFromLegacyWriter(
const WritableFileWriter* writer) {
LegacyWritableFileWrapper* file =
static_cast<LegacyWritableFileWrapper*>(writer->writable_file());
return static_cast<StringSink*>(file->target());
}
class NullLogger : public Logger {
public:
using Logger::Logv;
virtual void Logv(const char* /*format*/, va_list /*ap*/) override {}
virtual size_t GetLogFileSize() const override { return 0; }
};
// Corrupts key by changing the type
extern void CorruptKeyType(InternalKey* ikey);
Support for SingleDelete() Summary: This patch fixes #7460559. It introduces SingleDelete as a new database operation. This operation can be used to delete keys that were never overwritten (no put following another put of the same key). If an overwritten key is single deleted the behavior is undefined. Single deletion of a non-existent key has no effect but multiple consecutive single deletions are not allowed (see limitations). In contrast to the conventional Delete() operation, the deletion entry is removed along with the value when the two are lined up in a compaction. Note: The semantics are similar to @igor's prototype that allowed to have this behavior on the granularity of a column family ( https://reviews.facebook.net/D42093 ). This new patch, however, is more aggressive when it comes to removing tombstones: It removes the SingleDelete together with the value whenever there is no snapshot between them while the older patch only did this when the sequence number of the deletion was older than the earliest snapshot. Most of the complex additions are in the Compaction Iterator, all other changes should be relatively straightforward. The patch also includes basic support for single deletions in db_stress and db_bench. Limitations: - Not compatible with cuckoo hash tables - Single deletions cannot be used in combination with merges and normal deletions on the same key (other keys are not affected by this) - Consecutive single deletions are currently not allowed (and older version of this patch supported this so it could be resurrected if needed) Test Plan: make all check Reviewers: yhchiang, sdong, rven, anthony, yoshinorim, igor Reviewed By: igor Subscribers: maykov, dhruba, leveldb Differential Revision: https://reviews.facebook.net/D43179
2015-09-17 20:42:56 +02:00
extern std::string KeyStr(const std::string& user_key,
const SequenceNumber& seq, const ValueType& t,
bool corrupt = false);
LogAndApply() should fail if the column family has been dropped Summary: This patch finally fixes the ColumnFamilyTest.ReadDroppedColumnFamily test. The test has been failing very sporadically and it was hard to repro. However, I managed to write a new tests that reproes the failure deterministically. Here's what happens: 1. We start the flush for the column family 2. We check if the column family was dropped here: https://github.com/facebook/rocksdb/blob/a3fc49bfddcdb1ff29409aacd06c04df56c7a1d7/db/flush_job.cc#L149 3. This check goes through, ends up in InstallMemtableFlushResults() and it goes into LogAndApply() 4. At about this time, we start dropping the column family. Dropping the column family process gets to LogAndApply() at about the same time as LogAndApply() from flush process 5. Drop column family goes through LogAndApply() first, marking the column family as dropped. 6. Flush process gets woken up and gets a chance to write to the MANIFEST. However, this is where it gets stuck: https://github.com/facebook/rocksdb/blob/a3fc49bfddcdb1ff29409aacd06c04df56c7a1d7/db/version_set.cc#L1975 7. We see that the column family was dropped, so there is no need to write to the MANIFEST. We return OK. 8. Flush gets OK back from LogAndApply() and it deletes the memtable, thinking that the data is now safely persisted to sst file. The fix is pretty simple. Instead of OK, we return ShutdownInProgress. This is not really true, but we have been using this status code to also mean "this operation was canceled because the column family has been dropped". The fix is only one LOC. All other code is related to tests. I added a new test that reproes the failure. I also moved SleepingBackgroundTask to util/testutil.h (because I needed it in column_family_test for my new test). There's plenty of other places where we reimplement SleepingBackgroundTask, but I'll address that in a separate commit. Test Plan: 1. new test 2. make check 3. Make sure the ColumnFamilyTest.ReadDroppedColumnFamily doesn't fail on Travis: https://travis-ci.org/facebook/rocksdb/jobs/79952386 Reviewers: yhchiang, anthony, IslamAbdelRahman, kradhakrishnan, rven, sdong Reviewed By: sdong Subscribers: dhruba, leveldb Differential Revision: https://reviews.facebook.net/D46773
2015-09-15 20:28:44 +02:00
class SleepingBackgroundTask {
public:
SleepingBackgroundTask()
: bg_cv_(&mutex_),
should_sleep_(true),
done_with_sleep_(false),
sleeping_(false) {}
bool IsSleeping() {
MutexLock l(&mutex_);
return sleeping_;
}
LogAndApply() should fail if the column family has been dropped Summary: This patch finally fixes the ColumnFamilyTest.ReadDroppedColumnFamily test. The test has been failing very sporadically and it was hard to repro. However, I managed to write a new tests that reproes the failure deterministically. Here's what happens: 1. We start the flush for the column family 2. We check if the column family was dropped here: https://github.com/facebook/rocksdb/blob/a3fc49bfddcdb1ff29409aacd06c04df56c7a1d7/db/flush_job.cc#L149 3. This check goes through, ends up in InstallMemtableFlushResults() and it goes into LogAndApply() 4. At about this time, we start dropping the column family. Dropping the column family process gets to LogAndApply() at about the same time as LogAndApply() from flush process 5. Drop column family goes through LogAndApply() first, marking the column family as dropped. 6. Flush process gets woken up and gets a chance to write to the MANIFEST. However, this is where it gets stuck: https://github.com/facebook/rocksdb/blob/a3fc49bfddcdb1ff29409aacd06c04df56c7a1d7/db/version_set.cc#L1975 7. We see that the column family was dropped, so there is no need to write to the MANIFEST. We return OK. 8. Flush gets OK back from LogAndApply() and it deletes the memtable, thinking that the data is now safely persisted to sst file. The fix is pretty simple. Instead of OK, we return ShutdownInProgress. This is not really true, but we have been using this status code to also mean "this operation was canceled because the column family has been dropped". The fix is only one LOC. All other code is related to tests. I added a new test that reproes the failure. I also moved SleepingBackgroundTask to util/testutil.h (because I needed it in column_family_test for my new test). There's plenty of other places where we reimplement SleepingBackgroundTask, but I'll address that in a separate commit. Test Plan: 1. new test 2. make check 3. Make sure the ColumnFamilyTest.ReadDroppedColumnFamily doesn't fail on Travis: https://travis-ci.org/facebook/rocksdb/jobs/79952386 Reviewers: yhchiang, anthony, IslamAbdelRahman, kradhakrishnan, rven, sdong Reviewed By: sdong Subscribers: dhruba, leveldb Differential Revision: https://reviews.facebook.net/D46773
2015-09-15 20:28:44 +02:00
void DoSleep() {
MutexLock l(&mutex_);
sleeping_ = true;
bg_cv_.SignalAll();
LogAndApply() should fail if the column family has been dropped Summary: This patch finally fixes the ColumnFamilyTest.ReadDroppedColumnFamily test. The test has been failing very sporadically and it was hard to repro. However, I managed to write a new tests that reproes the failure deterministically. Here's what happens: 1. We start the flush for the column family 2. We check if the column family was dropped here: https://github.com/facebook/rocksdb/blob/a3fc49bfddcdb1ff29409aacd06c04df56c7a1d7/db/flush_job.cc#L149 3. This check goes through, ends up in InstallMemtableFlushResults() and it goes into LogAndApply() 4. At about this time, we start dropping the column family. Dropping the column family process gets to LogAndApply() at about the same time as LogAndApply() from flush process 5. Drop column family goes through LogAndApply() first, marking the column family as dropped. 6. Flush process gets woken up and gets a chance to write to the MANIFEST. However, this is where it gets stuck: https://github.com/facebook/rocksdb/blob/a3fc49bfddcdb1ff29409aacd06c04df56c7a1d7/db/version_set.cc#L1975 7. We see that the column family was dropped, so there is no need to write to the MANIFEST. We return OK. 8. Flush gets OK back from LogAndApply() and it deletes the memtable, thinking that the data is now safely persisted to sst file. The fix is pretty simple. Instead of OK, we return ShutdownInProgress. This is not really true, but we have been using this status code to also mean "this operation was canceled because the column family has been dropped". The fix is only one LOC. All other code is related to tests. I added a new test that reproes the failure. I also moved SleepingBackgroundTask to util/testutil.h (because I needed it in column_family_test for my new test). There's plenty of other places where we reimplement SleepingBackgroundTask, but I'll address that in a separate commit. Test Plan: 1. new test 2. make check 3. Make sure the ColumnFamilyTest.ReadDroppedColumnFamily doesn't fail on Travis: https://travis-ci.org/facebook/rocksdb/jobs/79952386 Reviewers: yhchiang, anthony, IslamAbdelRahman, kradhakrishnan, rven, sdong Reviewed By: sdong Subscribers: dhruba, leveldb Differential Revision: https://reviews.facebook.net/D46773
2015-09-15 20:28:44 +02:00
while (should_sleep_) {
bg_cv_.Wait();
}
sleeping_ = false;
LogAndApply() should fail if the column family has been dropped Summary: This patch finally fixes the ColumnFamilyTest.ReadDroppedColumnFamily test. The test has been failing very sporadically and it was hard to repro. However, I managed to write a new tests that reproes the failure deterministically. Here's what happens: 1. We start the flush for the column family 2. We check if the column family was dropped here: https://github.com/facebook/rocksdb/blob/a3fc49bfddcdb1ff29409aacd06c04df56c7a1d7/db/flush_job.cc#L149 3. This check goes through, ends up in InstallMemtableFlushResults() and it goes into LogAndApply() 4. At about this time, we start dropping the column family. Dropping the column family process gets to LogAndApply() at about the same time as LogAndApply() from flush process 5. Drop column family goes through LogAndApply() first, marking the column family as dropped. 6. Flush process gets woken up and gets a chance to write to the MANIFEST. However, this is where it gets stuck: https://github.com/facebook/rocksdb/blob/a3fc49bfddcdb1ff29409aacd06c04df56c7a1d7/db/version_set.cc#L1975 7. We see that the column family was dropped, so there is no need to write to the MANIFEST. We return OK. 8. Flush gets OK back from LogAndApply() and it deletes the memtable, thinking that the data is now safely persisted to sst file. The fix is pretty simple. Instead of OK, we return ShutdownInProgress. This is not really true, but we have been using this status code to also mean "this operation was canceled because the column family has been dropped". The fix is only one LOC. All other code is related to tests. I added a new test that reproes the failure. I also moved SleepingBackgroundTask to util/testutil.h (because I needed it in column_family_test for my new test). There's plenty of other places where we reimplement SleepingBackgroundTask, but I'll address that in a separate commit. Test Plan: 1. new test 2. make check 3. Make sure the ColumnFamilyTest.ReadDroppedColumnFamily doesn't fail on Travis: https://travis-ci.org/facebook/rocksdb/jobs/79952386 Reviewers: yhchiang, anthony, IslamAbdelRahman, kradhakrishnan, rven, sdong Reviewed By: sdong Subscribers: dhruba, leveldb Differential Revision: https://reviews.facebook.net/D46773
2015-09-15 20:28:44 +02:00
done_with_sleep_ = true;
bg_cv_.SignalAll();
}
void WaitUntilSleeping() {
MutexLock l(&mutex_);
while (!sleeping_ || !should_sleep_) {
bg_cv_.Wait();
}
}
// Waits for the status to change to sleeping,
// otherwise times out.
// wait_time is in microseconds.
// Returns true when times out, false otherwise.
bool TimedWaitUntilSleeping(uint64_t wait_time) {
auto abs_time = Env::Default()->NowMicros() + wait_time;
MutexLock l(&mutex_);
while (!sleeping_ || !should_sleep_) {
if (bg_cv_.TimedWait(abs_time)) {
return true;
}
}
return false;
}
LogAndApply() should fail if the column family has been dropped Summary: This patch finally fixes the ColumnFamilyTest.ReadDroppedColumnFamily test. The test has been failing very sporadically and it was hard to repro. However, I managed to write a new tests that reproes the failure deterministically. Here's what happens: 1. We start the flush for the column family 2. We check if the column family was dropped here: https://github.com/facebook/rocksdb/blob/a3fc49bfddcdb1ff29409aacd06c04df56c7a1d7/db/flush_job.cc#L149 3. This check goes through, ends up in InstallMemtableFlushResults() and it goes into LogAndApply() 4. At about this time, we start dropping the column family. Dropping the column family process gets to LogAndApply() at about the same time as LogAndApply() from flush process 5. Drop column family goes through LogAndApply() first, marking the column family as dropped. 6. Flush process gets woken up and gets a chance to write to the MANIFEST. However, this is where it gets stuck: https://github.com/facebook/rocksdb/blob/a3fc49bfddcdb1ff29409aacd06c04df56c7a1d7/db/version_set.cc#L1975 7. We see that the column family was dropped, so there is no need to write to the MANIFEST. We return OK. 8. Flush gets OK back from LogAndApply() and it deletes the memtable, thinking that the data is now safely persisted to sst file. The fix is pretty simple. Instead of OK, we return ShutdownInProgress. This is not really true, but we have been using this status code to also mean "this operation was canceled because the column family has been dropped". The fix is only one LOC. All other code is related to tests. I added a new test that reproes the failure. I also moved SleepingBackgroundTask to util/testutil.h (because I needed it in column_family_test for my new test). There's plenty of other places where we reimplement SleepingBackgroundTask, but I'll address that in a separate commit. Test Plan: 1. new test 2. make check 3. Make sure the ColumnFamilyTest.ReadDroppedColumnFamily doesn't fail on Travis: https://travis-ci.org/facebook/rocksdb/jobs/79952386 Reviewers: yhchiang, anthony, IslamAbdelRahman, kradhakrishnan, rven, sdong Reviewed By: sdong Subscribers: dhruba, leveldb Differential Revision: https://reviews.facebook.net/D46773
2015-09-15 20:28:44 +02:00
void WakeUp() {
MutexLock l(&mutex_);
should_sleep_ = false;
bg_cv_.SignalAll();
}
void WaitUntilDone() {
MutexLock l(&mutex_);
while (!done_with_sleep_) {
bg_cv_.Wait();
}
}
// Similar to TimedWaitUntilSleeping.
// Waits until the task is done.
bool TimedWaitUntilDone(uint64_t wait_time) {
auto abs_time = Env::Default()->NowMicros() + wait_time;
MutexLock l(&mutex_);
while (!done_with_sleep_) {
if (bg_cv_.TimedWait(abs_time)) {
return true;
}
}
return false;
}
LogAndApply() should fail if the column family has been dropped Summary: This patch finally fixes the ColumnFamilyTest.ReadDroppedColumnFamily test. The test has been failing very sporadically and it was hard to repro. However, I managed to write a new tests that reproes the failure deterministically. Here's what happens: 1. We start the flush for the column family 2. We check if the column family was dropped here: https://github.com/facebook/rocksdb/blob/a3fc49bfddcdb1ff29409aacd06c04df56c7a1d7/db/flush_job.cc#L149 3. This check goes through, ends up in InstallMemtableFlushResults() and it goes into LogAndApply() 4. At about this time, we start dropping the column family. Dropping the column family process gets to LogAndApply() at about the same time as LogAndApply() from flush process 5. Drop column family goes through LogAndApply() first, marking the column family as dropped. 6. Flush process gets woken up and gets a chance to write to the MANIFEST. However, this is where it gets stuck: https://github.com/facebook/rocksdb/blob/a3fc49bfddcdb1ff29409aacd06c04df56c7a1d7/db/version_set.cc#L1975 7. We see that the column family was dropped, so there is no need to write to the MANIFEST. We return OK. 8. Flush gets OK back from LogAndApply() and it deletes the memtable, thinking that the data is now safely persisted to sst file. The fix is pretty simple. Instead of OK, we return ShutdownInProgress. This is not really true, but we have been using this status code to also mean "this operation was canceled because the column family has been dropped". The fix is only one LOC. All other code is related to tests. I added a new test that reproes the failure. I also moved SleepingBackgroundTask to util/testutil.h (because I needed it in column_family_test for my new test). There's plenty of other places where we reimplement SleepingBackgroundTask, but I'll address that in a separate commit. Test Plan: 1. new test 2. make check 3. Make sure the ColumnFamilyTest.ReadDroppedColumnFamily doesn't fail on Travis: https://travis-ci.org/facebook/rocksdb/jobs/79952386 Reviewers: yhchiang, anthony, IslamAbdelRahman, kradhakrishnan, rven, sdong Reviewed By: sdong Subscribers: dhruba, leveldb Differential Revision: https://reviews.facebook.net/D46773
2015-09-15 20:28:44 +02:00
bool WokenUp() {
MutexLock l(&mutex_);
return should_sleep_ == false;
}
void Reset() {
MutexLock l(&mutex_);
should_sleep_ = true;
done_with_sleep_ = false;
}
static void DoSleepTask(void* arg) {
reinterpret_cast<SleepingBackgroundTask*>(arg)->DoSleep();
}
private:
port::Mutex mutex_;
port::CondVar bg_cv_; // Signalled when background work finishes
bool should_sleep_;
bool done_with_sleep_;
bool sleeping_;
LogAndApply() should fail if the column family has been dropped Summary: This patch finally fixes the ColumnFamilyTest.ReadDroppedColumnFamily test. The test has been failing very sporadically and it was hard to repro. However, I managed to write a new tests that reproes the failure deterministically. Here's what happens: 1. We start the flush for the column family 2. We check if the column family was dropped here: https://github.com/facebook/rocksdb/blob/a3fc49bfddcdb1ff29409aacd06c04df56c7a1d7/db/flush_job.cc#L149 3. This check goes through, ends up in InstallMemtableFlushResults() and it goes into LogAndApply() 4. At about this time, we start dropping the column family. Dropping the column family process gets to LogAndApply() at about the same time as LogAndApply() from flush process 5. Drop column family goes through LogAndApply() first, marking the column family as dropped. 6. Flush process gets woken up and gets a chance to write to the MANIFEST. However, this is where it gets stuck: https://github.com/facebook/rocksdb/blob/a3fc49bfddcdb1ff29409aacd06c04df56c7a1d7/db/version_set.cc#L1975 7. We see that the column family was dropped, so there is no need to write to the MANIFEST. We return OK. 8. Flush gets OK back from LogAndApply() and it deletes the memtable, thinking that the data is now safely persisted to sst file. The fix is pretty simple. Instead of OK, we return ShutdownInProgress. This is not really true, but we have been using this status code to also mean "this operation was canceled because the column family has been dropped". The fix is only one LOC. All other code is related to tests. I added a new test that reproes the failure. I also moved SleepingBackgroundTask to util/testutil.h (because I needed it in column_family_test for my new test). There's plenty of other places where we reimplement SleepingBackgroundTask, but I'll address that in a separate commit. Test Plan: 1. new test 2. make check 3. Make sure the ColumnFamilyTest.ReadDroppedColumnFamily doesn't fail on Travis: https://travis-ci.org/facebook/rocksdb/jobs/79952386 Reviewers: yhchiang, anthony, IslamAbdelRahman, kradhakrishnan, rven, sdong Reviewed By: sdong Subscribers: dhruba, leveldb Differential Revision: https://reviews.facebook.net/D46773
2015-09-15 20:28:44 +02:00
};
// Filters merge operands and values that are equal to `num`.
class FilterNumber : public CompactionFilter {
public:
explicit FilterNumber(uint64_t num) : num_(num) {}
std::string last_merge_operand_key() { return last_merge_operand_key_; }
bool Filter(int /*level*/, const ROCKSDB_NAMESPACE::Slice& /*key*/,
const ROCKSDB_NAMESPACE::Slice& value, std::string* /*new_value*/,
bool* /*value_changed*/) const override {
if (value.size() == sizeof(uint64_t)) {
return num_ == DecodeFixed64(value.data());
}
return true;
}
bool FilterMergeOperand(
int /*level*/, const ROCKSDB_NAMESPACE::Slice& key,
const ROCKSDB_NAMESPACE::Slice& value) const override {
last_merge_operand_key_ = key.ToString();
if (value.size() == sizeof(uint64_t)) {
return num_ == DecodeFixed64(value.data());
}
return true;
}
const char* Name() const override { return "FilterBadMergeOperand"; }
private:
mutable std::string last_merge_operand_key_;
uint64_t num_;
};
inline std::string EncodeInt(uint64_t x) {
std::string result;
PutFixed64(&result, x);
return result;
}
class SeqStringSource : public SequentialFile {
public:
SeqStringSource(const std::string& data, std::atomic<int>* read_count)
: data_(data), offset_(0), read_count_(read_count) {}
~SeqStringSource() override {}
Status Read(size_t n, Slice* result, char* scratch) override {
std::string output;
if (offset_ < data_.size()) {
n = std::min(data_.size() - offset_, n);
memcpy(scratch, data_.data() + offset_, n);
offset_ += n;
*result = Slice(scratch, n);
} else {
return Status::InvalidArgument(
"Attemp to read when it already reached eof.");
}
(*read_count_)++;
return Status::OK();
}
Status Skip(uint64_t n) override {
if (offset_ >= data_.size()) {
return Status::InvalidArgument(
"Attemp to read when it already reached eof.");
}
// TODO(yhchiang): Currently doesn't handle the overflow case.
offset_ += static_cast<size_t>(n);
return Status::OK();
}
private:
std::string data_;
size_t offset_;
std::atomic<int>* read_count_;
};
class StringEnv : public EnvWrapper {
public:
class StringSink : public WritableFile {
public:
explicit StringSink(std::string* contents)
: WritableFile(), contents_(contents) {}
virtual Status Truncate(uint64_t size) override {
contents_->resize(static_cast<size_t>(size));
return Status::OK();
}
virtual Status Close() override { return Status::OK(); }
virtual Status Flush() override { return Status::OK(); }
virtual Status Sync() override { return Status::OK(); }
virtual Status Append(const Slice& slice) override {
contents_->append(slice.data(), slice.size());
return Status::OK();
}
private:
std::string* contents_;
};
explicit StringEnv(Env* t) : EnvWrapper(t) {}
~StringEnv() override {}
const std::string& GetContent(const std::string& f) { return files_[f]; }
const Status WriteToNewFile(const std::string& file_name,
const std::string& content) {
std::unique_ptr<WritableFile> r;
auto s = NewWritableFile(file_name, &r, EnvOptions());
if (!s.ok()) {
return s;
}
r->Append(content);
r->Flush();
r->Close();
assert(files_[file_name] == content);
return Status::OK();
}
// The following text is boilerplate that forwards all methods to target()
Status NewSequentialFile(const std::string& f,
std::unique_ptr<SequentialFile>* r,
const EnvOptions& /*options*/) override {
auto iter = files_.find(f);
if (iter == files_.end()) {
return Status::NotFound("The specified file does not exist", f);
}
r->reset(new SeqStringSource(iter->second, &num_seq_file_read_));
return Status::OK();
}
Status NewRandomAccessFile(const std::string& /*f*/,
std::unique_ptr<RandomAccessFile>* /*r*/,
const EnvOptions& /*options*/) override {
return Status::NotSupported();
}
Status NewWritableFile(const std::string& f,
std::unique_ptr<WritableFile>* r,
const EnvOptions& /*options*/) override {
auto iter = files_.find(f);
if (iter != files_.end()) {
return Status::IOError("The specified file already exists", f);
}
r->reset(new StringSink(&files_[f]));
return Status::OK();
}
virtual Status NewDirectory(
const std::string& /*name*/,
std::unique_ptr<Directory>* /*result*/) override {
return Status::NotSupported();
}
Status FileExists(const std::string& f) override {
if (files_.find(f) == files_.end()) {
return Status::NotFound();
}
return Status::OK();
}
Status GetChildren(const std::string& /*dir*/,
std::vector<std::string>* /*r*/) override {
return Status::NotSupported();
}
Status DeleteFile(const std::string& f) override {
files_.erase(f);
return Status::OK();
}
Status CreateDir(const std::string& /*d*/) override {
return Status::NotSupported();
}
Status CreateDirIfMissing(const std::string& /*d*/) override {
return Status::NotSupported();
}
Status DeleteDir(const std::string& /*d*/) override {
return Status::NotSupported();
}
Status GetFileSize(const std::string& f, uint64_t* s) override {
auto iter = files_.find(f);
if (iter == files_.end()) {
return Status::NotFound("The specified file does not exist:", f);
}
*s = iter->second.size();
return Status::OK();
}
Status GetFileModificationTime(const std::string& /*fname*/,
uint64_t* /*file_mtime*/) override {
return Status::NotSupported();
}
Status RenameFile(const std::string& /*s*/,
const std::string& /*t*/) override {
return Status::NotSupported();
}
Status LinkFile(const std::string& /*s*/,
const std::string& /*t*/) override {
return Status::NotSupported();
}
Status LockFile(const std::string& /*f*/, FileLock** /*l*/) override {
return Status::NotSupported();
}
Status UnlockFile(FileLock* /*l*/) override {
return Status::NotSupported();
}
std::atomic<int> num_seq_file_read_;
protected:
std::unordered_map<std::string, std::string> files_;
};
// Randomly initialize the given DBOptions
void RandomInitDBOptions(DBOptions* db_opt, Random* rnd);
// Randomly initialize the given ColumnFamilyOptions
// Note that the caller is responsible for releasing non-null
// cf_opt->compaction_filter.
void RandomInitCFOptions(ColumnFamilyOptions* cf_opt, DBOptions&, Random* rnd);
// A dummy merge operator which can change its name
class ChanglingMergeOperator : public MergeOperator {
public:
explicit ChanglingMergeOperator(const std::string& name)
: name_(name + "MergeOperator") {}
~ChanglingMergeOperator() {}
void SetName(const std::string& name) { name_ = name; }
virtual bool FullMergeV2(const MergeOperationInput& /*merge_in*/,
MergeOperationOutput* /*merge_out*/) const override {
return false;
}
virtual bool PartialMergeMulti(const Slice& /*key*/,
const std::deque<Slice>& /*operand_list*/,
std::string* /*new_value*/,
Logger* /*logger*/) const override {
return false;
}
virtual const char* Name() const override { return name_.c_str(); }
protected:
std::string name_;
};
// Returns a dummy merge operator with random name.
MergeOperator* RandomMergeOperator(Random* rnd);
// A dummy compaction filter which can change its name
class ChanglingCompactionFilter : public CompactionFilter {
public:
explicit ChanglingCompactionFilter(const std::string& name)
: name_(name + "CompactionFilter") {}
~ChanglingCompactionFilter() {}
void SetName(const std::string& name) { name_ = name; }
bool Filter(int /*level*/, const Slice& /*key*/,
const Slice& /*existing_value*/, std::string* /*new_value*/,
bool* /*value_changed*/) const override {
return false;
}
const char* Name() const override { return name_.c_str(); }
private:
std::string name_;
};
// Returns a dummy compaction filter with a random name.
CompactionFilter* RandomCompactionFilter(Random* rnd);
// A dummy compaction filter factory which can change its name
class ChanglingCompactionFilterFactory : public CompactionFilterFactory {
public:
explicit ChanglingCompactionFilterFactory(const std::string& name)
: name_(name + "CompactionFilterFactory") {}
~ChanglingCompactionFilterFactory() {}
void SetName(const std::string& name) { name_ = name; }
std::unique_ptr<CompactionFilter> CreateCompactionFilter(
const CompactionFilter::Context& /*context*/) override {
return std::unique_ptr<CompactionFilter>();
}
// Returns a name that identifies this compaction filter factory.
const char* Name() const override { return name_.c_str(); }
protected:
std::string name_;
};
extern const Comparator* ComparatorWithU64Ts();
CompressionType RandomCompressionType(Random* rnd);
void RandomCompressionTypeVector(const size_t count,
std::vector<CompressionType>* types,
Random* rnd);
CompactionFilterFactory* RandomCompactionFilterFactory(Random* rnd);
const SliceTransform* RandomSliceTransform(Random* rnd, int pre_defined = -1);
TableFactory* RandomTableFactory(Random* rnd, int pre_defined = -1);
std::string RandomName(Random* rnd, const size_t len);
Status DestroyDir(Env* env, const std::string& dir);
bool IsDirectIOSupported(Env* env, const std::string& dir);
// Return the number of lines where a given pattern was found in a file.
size_t GetLinesCount(const std::string& fname, const std::string& pattern);
// TEST_TMPDIR may be set to /dev/shm in Makefile,
// but /dev/shm does not support direct IO.
// Tries to set TEST_TMPDIR to a directory supporting direct IO.
void ResetTmpDirForDirectIO();
// Sets up sync points to mock direct IO instead of actually issuing direct IO
// to the file system.
void SetupSyncPointsToMockDirectIO();
void CorruptFile(const std::string& fname, int offset, int bytes_to_corrupt);
} // namespace test
} // namespace ROCKSDB_NAMESPACE