ClickHouse/dbms/src/Disks/DiskS3.cpp
2020-03-20 21:53:35 +03:00

736 lines
22 KiB
C++

#include "DiskS3.h"
#if USE_AWS_S3
# include "DiskFactory.h"
# include <random>
# include <utility>
# include <IO/ReadBufferFromFile.h>
# include <IO/ReadBufferFromS3.h>
# include <IO/ReadHelpers.h>
# include <IO/S3Common.h>
# include <IO/WriteBufferFromFile.h>
# include <IO/WriteBufferFromS3.h>
# include <IO/WriteHelpers.h>
# include <Poco/File.h>
# include <Common/checkStackSize.h>
# include <Common/quoteString.h>
# include <Common/thread_local_rng.h>
# include <aws/s3/model/CopyObjectRequest.h>
# include <aws/s3/model/DeleteObjectRequest.h>
# include <aws/s3/model/GetObjectRequest.h>
namespace DB
{
namespace ErrorCodes
{
extern const int LOGICAL_ERROR;
extern const int FILE_ALREADY_EXISTS;
extern const int PATH_ACCESS_DENIED;
extern const int CANNOT_SEEK_THROUGH_FILE;
extern const int UNKNOWN_FORMAT;
}
namespace
{
String getRandomName()
{
std::uniform_int_distribution<int> distribution('a', 'z');
String res(32, ' '); /// The number of bits of entropy should be not less than 128.
for (auto & c : res)
c = distribution(thread_local_rng);
return res;
}
template <typename Result, typename Error>
void throwIfError(Aws::Utils::Outcome<Result, Error> && response)
{
if (!response.IsSuccess())
{
const auto & err = response.GetError();
throw Exception(err.GetMessage(), static_cast<int>(err.GetErrorType()));
}
}
/**
* S3 metadata file layout:
* Number of S3 objects, Total size of all S3 objects.
* Each S3 object represents path where object located in S3 and size of object.
*/
struct Metadata
{
/// Metadata file version.
const UInt32 VERSION = 1;
using PathAndSize = std::pair<String, size_t>;
/// Path to metadata file on local FS.
String metadata_file_path;
/// S3 object references count.
UInt32 s3_objects_count;
/// Total size of all S3 objects.
size_t total_size;
/// S3 objects paths and their sizes.
std::vector<PathAndSize> s3_objects;
explicit Metadata(const Poco::File & file) : Metadata(file.path(), false) {}
/// Load metadata by path or create empty if `create` flag is set.
explicit Metadata(const String & file_path, bool create = false)
: metadata_file_path(file_path), s3_objects_count(0), total_size(0), s3_objects(0)
{
if (create)
return;
ReadBufferFromFile buf(file_path, 1024); /* reasonable buffer size for small file */
UInt32 version;
readIntText(version, buf);
if (version != VERSION)
throw Exception(
"Unknown metadata file version. Path: " + file_path + ", Version: " + std::to_string(version)
+ ", Expected version: " + std::to_string(VERSION),
ErrorCodes::UNKNOWN_FORMAT);
assertChar('\n', buf);
readIntText(s3_objects_count, buf);
assertChar('\t', buf);
readIntText(total_size, buf);
assertChar('\n', buf);
s3_objects.resize(s3_objects_count);
for (UInt32 i = 0; i < s3_objects_count; ++i)
{
String path;
size_t size;
readIntText(size, buf);
assertChar('\t', buf);
readEscapedString(path, buf);
assertChar('\n', buf);
s3_objects[i] = std::make_pair(path, size);
}
}
void addObject(const String & path, size_t size)
{
++s3_objects_count;
total_size += size;
s3_objects.emplace_back(path, size);
}
/// Fsync metadata file if 'sync' flag is set.
void save(bool sync = false)
{
WriteBufferFromFile buf(metadata_file_path, 1024);
writeIntText(VERSION, buf);
writeChar('\n', buf);
writeIntText(s3_objects_count, buf);
writeChar('\t', buf);
writeIntText(total_size, buf);
writeChar('\n', buf);
for (UInt32 i = 0; i < s3_objects_count; ++i)
{
auto path_and_size = s3_objects[i];
writeIntText(path_and_size.second, buf);
writeChar('\t', buf);
writeEscapedString(path_and_size.first, buf);
writeChar('\n', buf);
}
buf.finalize();
if (sync)
buf.sync();
}
};
/// Reads data from S3 using stored paths in metadata.
class ReadIndirectBufferFromS3 final : public ReadBufferFromFileBase
{
public:
ReadIndirectBufferFromS3(
std::shared_ptr<Aws::S3::S3Client> client_ptr_, const String & bucket_, Metadata metadata_, size_t buf_size_)
: client_ptr(std::move(client_ptr_)),
bucket(bucket_),
metadata(std::move(metadata_)),
buf_size(buf_size_)
{
}
off_t seek(off_t offset_, int whence) override
{
if (whence == SEEK_CUR)
{
/// If position within current working buffer - shift pos.
if (working_buffer.size() && size_t(getPosition() + offset_) < absolute_position)
{
pos += offset_;
return getPosition();
}
else
{
absolute_position += offset_;
}
}
else if (whence == SEEK_SET)
{
/// If position within current working buffer - shift pos.
if (working_buffer.size() && size_t(offset_) >= absolute_position - working_buffer.size()
&& size_t(offset_) < absolute_position)
{
pos = working_buffer.end() - (absolute_position - offset_);
return getPosition();
}
else
{
absolute_position = offset_;
}
}
else
throw Exception("Only SEEK_SET or SEEK_CUR modes are allowed.", ErrorCodes::CANNOT_SEEK_THROUGH_FILE);
current_buf = initialize();
pos = working_buffer.end();
return absolute_position;
}
off_t getPosition() override { return absolute_position - available(); }
std::string getFileName() const override { return metadata.metadata_file_path; }
private:
std::unique_ptr<ReadBufferFromS3> initialize()
{
size_t offset = absolute_position;
for (UInt32 i = 0; i < metadata.s3_objects_count; ++i)
{
current_buf_idx = i;
auto [path, size] = metadata.s3_objects[i];
if (size > offset)
{
auto buf = std::make_unique<ReadBufferFromS3>(client_ptr, bucket, path, buf_size);
buf->seek(offset, SEEK_SET);
return buf;
}
offset -= size;
}
return nullptr;
}
bool nextImpl() override
{
/// Find first available buffer that fits to given offset.
if (!current_buf)
current_buf = initialize();
/// If current buffer has remaining data - use it.
if (current_buf && current_buf->next())
{
working_buffer = current_buf->buffer();
absolute_position += working_buffer.size();
return true;
}
/// If there is no available buffers - nothing to read.
if (current_buf_idx + 1 >= metadata.s3_objects_count)
return false;
++current_buf_idx;
auto path = metadata.s3_objects[current_buf_idx].first;
current_buf = std::make_unique<ReadBufferFromS3>(client_ptr, bucket, path, buf_size);
current_buf->next();
working_buffer = current_buf->buffer();
absolute_position += working_buffer.size();
return true;
}
private:
std::shared_ptr<Aws::S3::S3Client> client_ptr;
const String & bucket;
Metadata metadata;
size_t buf_size;
size_t absolute_position = 0;
UInt32 current_buf_idx = 0;
std::unique_ptr<ReadBufferFromS3> current_buf;
};
/// Stores data in S3 and adds the object key (S3 path) and object size to metadata file on local FS.
class WriteIndirectBufferFromS3 final : public WriteBufferFromFileBase
{
public:
WriteIndirectBufferFromS3(
std::shared_ptr<Aws::S3::S3Client> & client_ptr_,
const String & bucket_,
Metadata metadata_,
const String & s3_path_,
size_t min_upload_part_size,
size_t buf_size_)
: WriteBufferFromFileBase(buf_size_, nullptr, 0)
, impl(WriteBufferFromS3(client_ptr_, bucket_, s3_path_, min_upload_part_size, buf_size_))
, metadata(std::move(metadata_))
, s3_path(s3_path_)
{
}
~WriteIndirectBufferFromS3() override
{
try
{
finalize();
}
catch (...)
{
tryLogCurrentException(__PRETTY_FUNCTION__);
}
}
void finalize() override
{
if (finalized)
return;
next();
impl.finalize();
metadata.addObject(s3_path, count());
metadata.save();
finalized = true;
}
void sync() override
{
if (finalized)
metadata.save(true);
}
std::string getFileName() const override { return metadata.metadata_file_path; }
private:
void nextImpl() override
{
/// Transfer current working buffer to WriteBufferFromS3.
impl.swap(*this);
/// Write actual data to S3.
impl.next();
/// Return back working buffer.
impl.swap(*this);
}
private:
WriteBufferFromS3 impl;
bool finalized = false;
Metadata metadata;
String s3_path;
};
}
class DiskS3DirectoryIterator final : public IDiskDirectoryIterator
{
public:
DiskS3DirectoryIterator(const String & full_path, const String & folder_path_) : iter(full_path), folder_path(folder_path_) {}
void next() override { ++iter; }
bool isValid() const override { return iter != Poco::DirectoryIterator(); }
String path() const override
{
if (iter->isDirectory())
return folder_path + iter.name() + '/';
else
return folder_path + iter.name();
}
String name() const override { return iter.name(); }
private:
Poco::DirectoryIterator iter;
String folder_path;
};
using DiskS3Ptr = std::shared_ptr<DiskS3>;
class DiskS3Reservation final : public IReservation
{
public:
DiskS3Reservation(const DiskS3Ptr & disk_, UInt64 size_)
: disk(disk_), size(size_), metric_increment(CurrentMetrics::DiskSpaceReservedForMerge, size_)
{
}
UInt64 getSize() const override { return size; }
DiskPtr getDisk() const override { return disk; }
void update(UInt64 new_size) override
{
std::lock_guard lock(disk->reservation_mutex);
disk->reserved_bytes -= size;
size = new_size;
disk->reserved_bytes += size;
}
~DiskS3Reservation() override;
private:
DiskS3Ptr disk;
UInt64 size;
CurrentMetrics::Increment metric_increment;
};
DiskS3::DiskS3(
String name_,
std::shared_ptr<Aws::S3::S3Client> client_,
String bucket_,
String s3_root_path_,
String metadata_path_,
size_t min_upload_part_size_)
: name(std::move(name_))
, client(std::move(client_))
, bucket(std::move(bucket_))
, s3_root_path(std::move(s3_root_path_))
, metadata_path(std::move(metadata_path_))
, min_upload_part_size(min_upload_part_size_)
{
}
ReservationPtr DiskS3::reserve(UInt64 bytes)
{
if (!tryReserve(bytes))
return {};
return std::make_unique<DiskS3Reservation>(std::static_pointer_cast<DiskS3>(shared_from_this()), bytes);
}
bool DiskS3::exists(const String & path) const
{
return Poco::File(metadata_path + path).exists();
}
bool DiskS3::isFile(const String & path) const
{
return Poco::File(metadata_path + path).isFile();
}
bool DiskS3::isDirectory(const String & path) const
{
return Poco::File(metadata_path + path).isDirectory();
}
size_t DiskS3::getFileSize(const String & path) const
{
Metadata metadata(metadata_path + path);
return metadata.total_size;
}
void DiskS3::createDirectory(const String & path)
{
Poco::File(metadata_path + path).createDirectory();
}
void DiskS3::createDirectories(const String & path)
{
Poco::File(metadata_path + path).createDirectories();
}
DiskDirectoryIteratorPtr DiskS3::iterateDirectory(const String & path)
{
return std::make_unique<DiskS3DirectoryIterator>(metadata_path + path, path);
}
void DiskS3::clearDirectory(const String & path)
{
for (auto it{iterateDirectory(path)}; it->isValid(); it->next())
if (isFile(it->path()))
remove(it->path());
}
void DiskS3::moveFile(const String & from_path, const String & to_path)
{
if (exists(to_path))
throw Exception("File already exists: " + to_path, ErrorCodes::FILE_ALREADY_EXISTS);
Poco::File(metadata_path + from_path).renameTo(metadata_path + to_path);
}
void DiskS3::replaceFile(const String & from_path, const String & to_path)
{
Poco::File from_file(metadata_path + from_path);
Poco::File to_file(metadata_path + to_path);
if (to_file.exists())
{
Poco::File tmp_file(metadata_path + to_path + ".old");
to_file.renameTo(tmp_file.path());
from_file.renameTo(metadata_path + to_path);
remove(to_path + ".old");
}
else
from_file.renameTo(to_file.path());
}
void DiskS3::copyFile(const String & from_path, const String & to_path)
{
if (exists(to_path))
remove(to_path);
Metadata from(metadata_path + from_path);
Metadata to(metadata_path + to_path, true);
for (const auto & [path, size] : from.s3_objects)
{
auto new_path = s3_root_path + getRandomName();
Aws::S3::Model::CopyObjectRequest req;
req.SetCopySource(bucket + "/" + path);
req.SetBucket(bucket);
req.SetKey(new_path);
throwIfError(client->CopyObject(req));
to.addObject(new_path, size);
}
to.save();
}
std::unique_ptr<ReadBufferFromFileBase> DiskS3::readFile(const String & path, size_t buf_size, size_t, size_t, size_t) const
{
Metadata metadata(metadata_path + path);
LOG_DEBUG(
&Logger::get("DiskS3"),
"Read from file by path: " << backQuote(metadata_path + path) << " Existing S3 objects: " << metadata.s3_objects_count);
return std::make_unique<ReadIndirectBufferFromS3>(client, bucket, metadata, buf_size);
}
std::unique_ptr<WriteBufferFromFileBase> DiskS3::writeFile(const String & path, size_t buf_size, WriteMode mode, size_t, size_t)
{
bool exist = exists(path);
/// Path to store new S3 object.
auto s3_path = s3_root_path + getRandomName();
if (!exist || mode == WriteMode::Rewrite)
{
/// If metadata file exists - remove and create new.
if (exist)
remove(path);
Metadata metadata(metadata_path + path, true);
/// Save empty metadata to disk to have ability to get file size while buffer is not finalized.
metadata.save();
LOG_DEBUG(&Logger::get("DiskS3"), "Write to file by path: " << backQuote(metadata_path + path) << " New S3 path: " << s3_path);
return std::make_unique<WriteIndirectBufferFromS3>(client, bucket, metadata, s3_path, min_upload_part_size, buf_size);
}
else
{
Metadata metadata(metadata_path + path);
LOG_DEBUG(
&Logger::get("DiskS3"),
"Append to file by path: " << backQuote(metadata_path + path) << " New S3 path: " << s3_path
<< " Existing S3 objects: " << metadata.s3_objects_count);
return std::make_unique<WriteIndirectBufferFromS3>(client, bucket, metadata, s3_path, min_upload_part_size, buf_size);
}
}
void DiskS3::remove(const String & path)
{
LOG_DEBUG(&Logger::get("DiskS3"), "Remove file by path: " << backQuote(metadata_path + path));
Poco::File file(metadata_path + path);
if (file.isFile())
{
Metadata metadata(file);
for (UInt32 i = 0; i < metadata.s3_objects_count; ++i)
{
auto s3_path = metadata.s3_objects[i].first;
/// TODO: Make operation idempotent. Do not throw exception if key is already deleted.
Aws::S3::Model::DeleteObjectRequest request;
request.SetBucket(bucket);
request.SetKey(s3_path);
throwIfError(client->DeleteObject(request));
}
}
file.remove();
}
void DiskS3::removeRecursive(const String & path)
{
checkStackSize(); /// This is needed to prevent stack overflow in case of cyclic symlinks.
Poco::File file(metadata_path + path);
if (file.isFile())
{
remove(path);
}
else
{
for (auto it{iterateDirectory(path)}; it->isValid(); it->next())
removeRecursive(it->path());
file.remove();
}
}
bool DiskS3::tryReserve(UInt64 bytes)
{
std::lock_guard lock(reservation_mutex);
if (bytes == 0)
{
LOG_DEBUG(&Logger::get("DiskS3"), "Reserving 0 bytes on s3 disk " << backQuote(name));
++reservation_count;
return true;
}
auto available_space = getAvailableSpace();
UInt64 unreserved_space = available_space - std::min(available_space, reserved_bytes);
if (unreserved_space >= bytes)
{
LOG_DEBUG(
&Logger::get("DiskS3"),
"Reserving " << formatReadableSizeWithBinarySuffix(bytes) << " on disk " << backQuote(name) << ", having unreserved "
<< formatReadableSizeWithBinarySuffix(unreserved_space) << ".");
++reservation_count;
reserved_bytes += bytes;
return true;
}
return false;
}
void DiskS3::listFiles(const String & path, std::vector<String> & file_names)
{
for (auto it = iterateDirectory(path); it->isValid(); it->next())
file_names.push_back(it->name());
}
void DiskS3::setLastModified(const String & path, const Poco::Timestamp & timestamp)
{
Poco::File(metadata_path + path).setLastModified(timestamp);
}
Poco::Timestamp DiskS3::getLastModified(const String & path)
{
return Poco::File(metadata_path + path).getLastModified();
}
void DiskS3::createHardLink(const String & src_path, const String & dst_path)
{
/**
* TODO: Replace with optimal implementation:
* Store links into a list in metadata file.
* Hardlink creation is adding new link to list and just metadata file copy.
*/
copyFile(src_path, dst_path);
}
void DiskS3::createFile(const String & path)
{
/// Create empty metadata file.
Metadata metadata(metadata_path + path, true);
metadata.save();
}
void DiskS3::setReadOnly(const String & path)
{
Poco::File(metadata_path + path).setReadOnly(true);
}
DiskS3Reservation::~DiskS3Reservation()
{
try
{
std::lock_guard lock(disk->reservation_mutex);
if (disk->reserved_bytes < size)
{
disk->reserved_bytes = 0;
LOG_ERROR(&Logger::get("DiskLocal"), "Unbalanced reservations size for disk '" + disk->getName() + "'.");
}
else
{
disk->reserved_bytes -= size;
}
if (disk->reservation_count == 0)
LOG_ERROR(&Logger::get("DiskLocal"), "Unbalanced reservation count for disk '" + disk->getName() + "'.");
else
--disk->reservation_count;
}
catch (...)
{
tryLogCurrentException(__PRETTY_FUNCTION__);
}
}
namespace
{
void checkWriteAccess(IDisk & disk)
{
auto file = disk.writeFile("test_acl", DBMS_DEFAULT_BUFFER_SIZE, WriteMode::Rewrite);
file->write("test", 4);
}
void checkReadAccess(const String & disk_name, IDisk & disk)
{
auto file = disk.readFile("test_acl", DBMS_DEFAULT_BUFFER_SIZE);
String buf(4, '0');
file->readStrict(buf.data(), 4);
if (buf != "test")
throw Exception("No read access to S3 bucket in disk " + disk_name, ErrorCodes::PATH_ACCESS_DENIED);
}
void checkRemoveAccess(IDisk & disk)
{
disk.remove("test_acl");
}
}
void registerDiskS3(DiskFactory & factory)
{
auto creator = [](const String & name,
const Poco::Util::AbstractConfiguration & config,
const String & config_prefix,
const Context & context) -> DiskPtr {
Poco::File disk{context.getPath() + "disks/" + name};
disk.createDirectories();
S3::URI uri(Poco::URI(config.getString(config_prefix + ".endpoint")));
auto client = S3::ClientFactory::instance().create(
uri.endpoint,
config.getString(config_prefix + ".access_key_id", ""),
config.getString(config_prefix + ".secret_access_key", ""));
if (uri.key.back() != '/')
throw Exception("S3 path must ends with '/', but '" + uri.key + "' doesn't.", ErrorCodes::LOGICAL_ERROR);
String metadata_path = context.getPath() + "disks/" + name + "/";
auto s3disk
= std::make_shared<DiskS3>(name, client, uri.bucket, uri.key, metadata_path, context.getSettingsRef().s3_min_upload_part_size);
/// This code is used only to check access to the corresponding disk.
checkWriteAccess(*s3disk);
checkReadAccess(name, *s3disk);
checkRemoveAccess(*s3disk);
return s3disk;
};
factory.registerDiskType("s3", creator);
}
}
#endif