forked from mindspore-Ecosystem/mindspore
!14136 Improve cache server spilling performance
From: @lixiachen Reviewed-by: @robingrosman,@pandoublefeng Signed-off-by: @pandoublefeng
This commit is contained in:
commit
f45890c3f6
|
@ -30,7 +30,8 @@ Status CachePool::DoServiceStart() {
|
|||
if (!root_.toString().empty()) {
|
||||
Path spill = GetSpillPath();
|
||||
RETURN_IF_NOT_OK(spill.CreateDirectories());
|
||||
sm_ = std::make_shared<StorageManager>(spill);
|
||||
auto &cs = CacheServer::GetInstance();
|
||||
sm_ = std::make_shared<StorageManager>(spill, cs.GetNumWorkers());
|
||||
RETURN_IF_NOT_OK(sm_->ServiceStart());
|
||||
MS_LOG(INFO) << "CachePool will use disk folder: " << spill.toString();
|
||||
}
|
||||
|
|
|
@ -127,11 +127,23 @@ Status StorageContainer::Insert(const std::vector<ReadableSlice> &buf, off64_t *
|
|||
addr_t addr = 0;
|
||||
RETURN_IF_NOT_OK(bs_->Alloc(sz, &bspd, &addr));
|
||||
*offset = static_cast<off64_t>(addr);
|
||||
// We will do piecewise copy of the data to disk.
|
||||
for (auto &v : buf) {
|
||||
RETURN_IF_NOT_OK(Write(v, addr));
|
||||
addr += v.GetSize();
|
||||
// We will do piecewise copy of the data to a large buffer
|
||||
std::string mem;
|
||||
try {
|
||||
mem.resize(sz);
|
||||
CHECK_FAIL_RETURN_UNEXPECTED(mem.capacity() >= sz, "Programming error");
|
||||
} catch (const std::bad_alloc &e) {
|
||||
return Status(StatusCode::kMDOutOfMemory);
|
||||
}
|
||||
WritableSlice all(mem.data(), sz);
|
||||
size_t pos = 0;
|
||||
for (auto &v : buf) {
|
||||
WritableSlice row_data(all, pos);
|
||||
RETURN_IF_NOT_OK(WritableSlice::Copy(&row_data, v));
|
||||
pos += v.GetSize();
|
||||
}
|
||||
// Write all data to disk at once
|
||||
RETURN_IF_NOT_OK(Write(all, addr));
|
||||
return Status::OK();
|
||||
}
|
||||
|
||||
|
|
|
@ -22,6 +22,7 @@
|
|||
#include "utils/ms_utils.h"
|
||||
#include "minddata/dataset/util/log_adapter.h"
|
||||
#include "minddata/dataset/util/path.h"
|
||||
#include "minddata/dataset/util/random.h"
|
||||
#include "minddata/dataset/util/services.h"
|
||||
|
||||
namespace mindspore {
|
||||
|
@ -37,7 +38,7 @@ std::string StorageManager::ConstructFileName(const std::string &prefix, int32_t
|
|||
return (base_name + "." + suffix);
|
||||
}
|
||||
|
||||
Status StorageManager::AddOneContainer() {
|
||||
Status StorageManager::AddOneContainer(int replaced_container_pos) {
|
||||
const std::string kPrefix = "IMG";
|
||||
const std::string kSuffix = "LB";
|
||||
Path container_name = root_ / ConstructFileName(kPrefix, file_id_, kSuffix);
|
||||
|
@ -45,13 +46,22 @@ Status StorageManager::AddOneContainer() {
|
|||
RETURN_IF_NOT_OK(StorageContainer::CreateStorageContainer(&sc, container_name.toString()));
|
||||
containers_.push_back(sc);
|
||||
file_id_++;
|
||||
if (replaced_container_pos >= 0) {
|
||||
writable_containers_pool_[replaced_container_pos] = containers_.size() - 1;
|
||||
} else {
|
||||
writable_containers_pool_.push_back(containers_.size() - 1);
|
||||
}
|
||||
return Status::OK();
|
||||
}
|
||||
|
||||
Status StorageManager::DoServiceStart() {
|
||||
containers_.reserve(1000);
|
||||
writable_containers_pool_.reserve(pool_size_);
|
||||
if (root_.IsDirectory()) {
|
||||
RETURN_IF_NOT_OK(AddOneContainer());
|
||||
// create multiple containers and store their index in a pool
|
||||
for (int i = 0; i < pool_size_; i++) {
|
||||
RETURN_IF_NOT_OK(AddOneContainer());
|
||||
}
|
||||
} else {
|
||||
RETURN_STATUS_UNEXPECTED("Not a directory");
|
||||
}
|
||||
|
@ -67,22 +77,25 @@ Status StorageManager::Write(key_type *key, const std::vector<ReadableSlice> &bu
|
|||
if (sz == 0) {
|
||||
RETURN_STATUS_UNEXPECTED("Unexpected 0 length");
|
||||
}
|
||||
auto mt = GetRandomDevice();
|
||||
std::shared_ptr<StorageContainer> cont;
|
||||
key_type out_key;
|
||||
value_type out_value;
|
||||
bool create_new_container = false;
|
||||
int old_container_pos = -1;
|
||||
size_t last_num_container = -1;
|
||||
do {
|
||||
SharedLock lock_s(&rw_lock_);
|
||||
size_t num_containers = containers_.size();
|
||||
if (create_new_container && (num_containers == last_num_container)) {
|
||||
// Upgrade to exclusvie lock.
|
||||
if (create_new_container && (num_containers == last_num_container) && (old_container_pos >= 0)) {
|
||||
// Upgrade to exclusive lock.
|
||||
lock_s.Upgrade();
|
||||
create_new_container = false;
|
||||
// Check again if someone has already added a
|
||||
// new container after we got the x lock
|
||||
if (containers_.size() == num_containers) {
|
||||
RETURN_IF_NOT_OK(AddOneContainer());
|
||||
// Create a new container and replace the full container in the pool with the newly created one
|
||||
RETURN_IF_NOT_OK(AddOneContainer(old_container_pos));
|
||||
}
|
||||
// Refresh how many containers there are.
|
||||
num_containers = containers_.size();
|
||||
|
@ -92,17 +105,21 @@ Status StorageManager::Write(key_type *key, const std::vector<ReadableSlice> &bu
|
|||
if (num_containers == 0) {
|
||||
RETURN_STATUS_UNEXPECTED("num_containers is zero");
|
||||
}
|
||||
// Go to the last container to insert.
|
||||
cont = containers_.at(num_containers - 1);
|
||||
// Pick a random container from the writable container pool to insert.
|
||||
std::uniform_int_distribution<int> distribution(0, pool_size_ - 1);
|
||||
int pos_in_pool = distribution(mt);
|
||||
int cont_index = writable_containers_pool_.at(pos_in_pool);
|
||||
cont = containers_.at(cont_index);
|
||||
off64_t offset;
|
||||
Status rc = cont->Insert(buf, &offset);
|
||||
if (rc.StatusCode() == StatusCode::kMDBuddySpaceFull) {
|
||||
create_new_container = true;
|
||||
old_container_pos = pos_in_pool;
|
||||
// Remember how many containers we saw. In the next iteration we will do a comparison to see
|
||||
// if someone has already created it.
|
||||
last_num_container = num_containers;
|
||||
} else if (rc.IsOk()) {
|
||||
out_value = std::make_pair(num_containers - 1, std::make_pair(offset, sz));
|
||||
out_value = std::make_pair(cont_index, std::make_pair(offset, sz));
|
||||
RETURN_IF_NOT_OK(index_.insert(out_value, &out_key));
|
||||
*key = out_key;
|
||||
break;
|
||||
|
@ -150,11 +167,15 @@ Status StorageManager::DoServiceStop() noexcept {
|
|||
}
|
||||
}
|
||||
containers_.clear();
|
||||
writable_containers_pool_.clear();
|
||||
file_id_ = 0;
|
||||
return rc1;
|
||||
}
|
||||
|
||||
StorageManager::StorageManager(const Path &root) : root_(root), file_id_(0), index_() {}
|
||||
StorageManager::StorageManager(const Path &root) : root_(root), pool_size_(1), file_id_(0), index_() {}
|
||||
|
||||
StorageManager::StorageManager(const Path &root, int pool_size)
|
||||
: root_(root), pool_size_(pool_size), file_id_(0), index_() {}
|
||||
|
||||
StorageManager::~StorageManager() { (void)StorageManager::DoServiceStop(); }
|
||||
|
||||
|
|
|
@ -35,12 +35,23 @@ namespace mindspore {
|
|||
namespace dataset {
|
||||
class StorageManager : public Service {
|
||||
public:
|
||||
using storage_index = AutoIndexObj<std::pair<int, std::pair<off_t, size_t>>>;
|
||||
// Use these traits for the B+ tree inside the StorageManager
|
||||
struct StorageBPlusTreeTraits {
|
||||
// This determines the limit of number of keys in a node.
|
||||
using slot_type = uint16_t;
|
||||
// Number of slots in each leaf of the tree.
|
||||
static constexpr slot_type kLeafSlots = 512;
|
||||
// Number of slots in each inner node of the tree
|
||||
static constexpr slot_type kInnerSlots = 256;
|
||||
};
|
||||
using value_type = std::pair<int, std::pair<off_t, size_t>>;
|
||||
using storage_index = AutoIndexObj<value_type, std::allocator<value_type>, StorageBPlusTreeTraits>;
|
||||
using key_type = storage_index::key_type;
|
||||
using value_type = storage_index::value_type;
|
||||
|
||||
explicit StorageManager(const Path &);
|
||||
|
||||
StorageManager(const Path &root, int pool_size);
|
||||
|
||||
~StorageManager() override;
|
||||
|
||||
StorageManager(const StorageManager &) = delete;
|
||||
|
@ -63,12 +74,19 @@ class StorageManager : public Service {
|
|||
int file_id_;
|
||||
RWLock rw_lock_;
|
||||
storage_index index_;
|
||||
std::vector<int> writable_containers_pool_;
|
||||
int pool_size_;
|
||||
|
||||
std::string GetBaseName(const std::string &prefix, int32_t file_id);
|
||||
|
||||
std::string ConstructFileName(const std::string &prefix, int32_t file_id, const std::string &suffix);
|
||||
|
||||
Status AddOneContainer();
|
||||
/// \brief Add a new storage container
|
||||
/// The newly-created container is going to be added into a pool of writable containers.
|
||||
/// \param replaced_container_pos If provided, will use the newly created container to replace the corresponding old
|
||||
/// container in the pool. If not provided, will just append the newly created container to the end of the pool.
|
||||
/// \return Status object
|
||||
Status AddOneContainer(int replaced_container_pos = -1);
|
||||
};
|
||||
} // namespace dataset
|
||||
} // namespace mindspore
|
||||
|
|
|
@ -17,6 +17,7 @@
|
|||
#define MINDSPORE_CCSRC_MINDDATA_DATASET_UTIL_AUTO_INDEX_H_
|
||||
|
||||
#include <atomic>
|
||||
#include <functional>
|
||||
#include <memory>
|
||||
#include <utility>
|
||||
#include <vector>
|
||||
|
@ -30,16 +31,16 @@ namespace dataset {
|
|||
/// Use minKey() function to query the min key.
|
||||
/// Use maxKey() function to query the max key.
|
||||
/// @tparam T
|
||||
template <typename T, typename A = std::allocator<T>>
|
||||
class AutoIndexObj : public BPlusTree<int64_t, T, A> {
|
||||
template <typename V, typename A = std::allocator<V>, typename T = BPlusTreeTraits>
|
||||
class AutoIndexObj : public BPlusTree<int64_t, V, A, std::less<int64_t>, T> {
|
||||
public:
|
||||
using my_tree = BPlusTree<int64_t, T, A>;
|
||||
using my_tree = BPlusTree<int64_t, V, A, std::less<int64_t>, T>;
|
||||
using key_type = typename my_tree::key_type;
|
||||
using value_type = typename my_tree::value_type;
|
||||
|
||||
AutoIndexObj() : my_tree::BPlusTree(), inx_(kMinKey) {}
|
||||
|
||||
explicit AutoIndexObj(const Allocator<T> &alloc) : my_tree::BPlusTree(alloc), inx_(kMinKey) {}
|
||||
explicit AutoIndexObj(const Allocator<V> &alloc) : my_tree::BPlusTree(alloc), inx_(kMinKey) {}
|
||||
|
||||
~AutoIndexObj() = default;
|
||||
|
||||
|
|
Loading…
Reference in New Issue