hcq/oceanbase
1
0
Fork 0
Code Issues Pull Requests Actions 2 Packages Projects Releases Wiki Activity
Files
062a57e12a5454dea266ce4cf1296d0c9ffadb15
oceanbase/src/storage/ob_pg_sstable_mgr.cpp

1010 lines
35 KiB
C++
Raw Blame History

/**
* Copyright (c) 2021 OceanBase
* OceanBase CE is licensed under Mulan PubL v2.
* You can use this software according to the terms and conditions of the Mulan PubL v2.
* You may obtain a copy of Mulan PubL v2 at:
* http://license.coscl.org.cn/MulanPubL-2.0
* THIS SOFTWARE IS PROVIDED ON AN "AS IS" BASIS, WITHOUT WARRANTIES OF ANY KIND,
* EITHER EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO NON-INFRINGEMENT,
* MERCHANTABILITY OR FIT FOR A PARTICULAR PURPOSE.
* See the Mulan PubL v2 for more details.
*/
#define USING_LOG_PREFIX STORAGE
#include "storage/ob_pg_sstable_mgr.h"
#include "lib/oblog/ob_log_module.h"
#include "storage/ob_partition_log.h"
using namespace oceanbase::common;
using namespace oceanbase::blocksstable;
using namespace oceanbase::storage;
int ObPGSSTableMgr::TableGetFunctor::operator()(ObITable& table)
{
int ret = OB_SUCCESS;
if (OB_FAIL(table_handle_.set_table(&table))) {
LOG_WARN("fail to set table to handle", K(ret));
}
return ret;
}
int ObPGSSTableMgr::UnusedTableFunctor::init(const int64_t max_recycle_cnt)
{
int ret = OB_SUCCESS;
if (OB_UNLIKELY(is_inited_)) {
ret = OB_INIT_TWICE;
LOG_WARN("ObUnusedTableFunctor has not been inited", K(ret));
} else {
max_recycle_cnt_ = max_recycle_cnt;
is_inited_ = true;
}
return ret;
}
int ObPGSSTableMgr::UnusedTableFunctor::operator()(ObITable& table, bool& is_full)
{
int ret = OB_SUCCESS;
if (OB_UNLIKELY(!is_inited_)) {
ret = OB_NOT_INIT;
LOG_WARN("ObPGSSTableMgr has not been inited", K(ret));
} else if (0 == table.get_ref()) {
if (OB_FAIL(tables_.push_back(&table))) {
LOG_WARN("fail to push back table", K(ret));
} else {
is_full = tables_.count() >= max_recycle_cnt_;
}
}
return ret;
}
int ObPGSSTableMgr::UnusedTableFunctor::get_tables(common::ObIArray<ObITable*>& tables)
{
int ret = OB_SUCCESS;
tables.reset();
if (OB_UNLIKELY(!is_inited_)) {
ret = OB_NOT_INIT;
LOG_WARN("ObPGSSTableMgr has not been inited", K(ret));
} else {
for (int64_t i = 0; OB_SUCC(ret) && i < tables_.count(); ++i) {
ObITable* table = tables_.at(i);
if (OB_ISNULL(table)) {
ret = OB_ERR_SYS;
LOG_WARN("error sys, table must not be null", K(ret));
} else if (0 == table->get_ref()) {
if (OB_FAIL(tables.push_back(table))) {
LOG_WARN("fail to push back table", K(ret));
}
}
}
}
return ret;
}
int ObPGSSTableMgr::CheckAllTableUnusedFunctor::operator()(ObITable& table, bool& is_full)
{
int ret = OB_SUCCESS;
if (is_all_unused_) {
is_all_unused_ = 0 == table.get_ref();
}
is_full = is_all_unused_;
return ret;
}
ObPGSSTableMgr::ObPGSSTableMgr()
: is_inited_(false), pg_key_(), bucket_lock_(), table_map_(), file_handle_(), enable_write_log_(false)
{}
ObPGSSTableMgr::~ObPGSSTableMgr()
{
destroy();
}
void ObPGSSTableMgr::destroy()
{
if (is_inited_) {
{
ObBucketWLockAllGuard guard(bucket_lock_);
int tmp_ret = OB_SUCCESS;
UnusedTableFunctor finder;
ObArray<ObITable*> tables;
if (OB_SUCCESS != (tmp_ret = finder.init(INT64_MAX))) {
LOG_WARN("fail to init table map", K(tmp_ret));
} else if (OB_SUCCESS != (tmp_ret = table_map_.foreach (finder))) {
LOG_WARN("failed to get all tables", K(tmp_ret));
} else if (OB_SUCCESS != (tmp_ret = finder.get_tables(tables))) {
LOG_WARN("failed to get tables", K(tmp_ret));
} else {
for (int64_t i = 0; i < tables.count(); ++i) {
ObSSTable* sstable = static_cast<ObSSTable*>(tables.at(i));
free_sstable(sstable);
}
}
table_map_.destroy();
}
bucket_lock_.destroy();
pg_key_.reset();
file_handle_.reset();
enable_write_log_ = false;
is_inited_ = false;
}
}
int ObPGSSTableMgr::init(const ObPGKey& pg_key)
{
int ret = OB_SUCCESS;
if (OB_UNLIKELY(is_inited_)) {
ret = OB_INIT_TWICE;
LOG_WARN("ObPGSSTableMgr has been inited twice", K(ret));
} else if (OB_FAIL(
bucket_lock_.init(DEFAULT_HASH_BUCKET_COUNT, ObLatchIds::TABLE_MGR_MAP, ObModIds::OB_TABLE_MGR_MAP))) {
LOG_WARN("fail to init bucket lock", K(ret));
} else if (OB_FAIL(
table_map_.init(DEFAULT_HASH_BUCKET_COUNT, ObLatchIds::TABLE_MGR_MAP, ObModIds::OB_TABLE_MGR_MAP))) {
LOG_WARN("fail to init table map", K(ret));
} else {
pg_key_ = pg_key;
is_inited_ = true;
}
return ret;
}
int ObPGSSTableMgr::create_sstable(storage::ObCreateSSTableParamWithTable& param, ObTableHandle& handle)
{
int ret = OB_SUCCESS;
if (!is_inited_) {
ret = OB_NOT_INIT;
STORAGE_LOG(WARN, "not inited", K(ret));
} else {
ObCreatePartitionMeta schema;
ObCreateSSTableParamWithPartition p_param;
if (OB_FAIL(p_param.extract_from(param, schema))) {
STORAGE_LOG(WARN, "failed to extract partition param", K(param), K(ret));
} else {
ret = create_sstable(p_param, handle);
}
}
return ret;
}
int ObPGSSTableMgr::create_sstable(storage::ObCreateSSTableParamWithPartition& param, ObTableHandle& handle)
{
int ret = OB_SUCCESS;
int hash_ret = OB_SUCCESS;
ObSSTable* sstable = nullptr;
handle.reset();
ObBucketHashRLockGuard lock_guard(bucket_lock_, param.table_key_.hash());
param.pg_key_ = pg_key_;
if (OB_UNLIKELY(!is_inited_)) {
ret = OB_NOT_INIT;
LOG_WARN("ObPGSSTableMgr has not been inited", K(ret));
} else if (OB_UNLIKELY(!param.is_valid() || !ObITable::is_sstable(param.table_key_.table_type_))) {
ret = OB_INVALID_ARGUMENT;
LOG_WARN("invalid arguments", K(ret), K(param));
} else if (OB_HASH_NOT_EXIST != (hash_ret = table_map_.exist(param.table_key_))) {
if (OB_HASH_EXIST == hash_ret) {
ret = OB_ENTRY_EXIST;
LOG_WARN("sstable already exist", K(ret), K(param.table_key_));
} else {
ret = hash_ret;
LOG_WARN("fail to check sstable exist", K(ret), K(param.table_key_));
}
} else if (OB_FAIL(alloc_sstable(param.table_key_.get_tenant_id(), sstable))) {
LOG_WARN("fail to alloc table", K(ret), K(param));
} else if (OB_FAIL(sstable->init(param.table_key_))) {
LOG_WARN("fail to init sstable", K(ret), K(param));
} else if (OB_FAIL(sstable->open(param))) {
LOG_WARN("fail to open sstable", K(ret), K(param));
} else if (OB_FAIL(sstable->set_storage_file_handle(file_handle_))) {
LOG_WARN("fail to set storage file handle", K(ret));
} else if (OB_FAIL(handle.set_table(sstable))) {
LOG_WARN("fail to set table", K(ret));
}
if (OB_FAIL(ret) && nullptr != sstable) {
free_sstable(sstable);
sstable = nullptr;
handle.reset();
}
return ret;
}
int ObPGSSTableMgr::create_sstable(
const ObITable::TableKey& table_key, blocksstable::ObSSTableBaseMeta& meta, ObTableHandle& handle)
{
int ret = OB_SUCCESS;
int hash_ret = OB_SUCCESS;
handle.reset();
ObSSTable* sstable = nullptr;
meta.pg_key_ = pg_key_;
ObBucketHashRLockGuard lock_guard(bucket_lock_, table_key.hash());
if (OB_UNLIKELY(!is_inited_)) {
ret = OB_NOT_INIT;
LOG_WARN("ObPGSSTableMgr has not been inited", K(ret));
} else if (OB_UNLIKELY(!table_key.is_valid() || !meta.is_valid() || !ObITable::is_sstable(table_key.table_type_))) {
ret = OB_INVALID_ARGUMENT;
LOG_WARN("invalid arguments", K(ret), K(table_key), K(meta));
} else if (OB_HASH_NOT_EXIST != (hash_ret = table_map_.exist(table_key))) {
if (OB_HASH_EXIST == hash_ret) {
ret = OB_ENTRY_EXIST;
LOG_WARN("sstable already exist", K(ret), K(table_key));
} else {
ret = hash_ret;
LOG_WARN("fail to check sstable exist", K(ret), K(table_key));
}
} else if (OB_FAIL(alloc_sstable(table_key.get_tenant_id(), sstable))) {
LOG_WARN("fail to alloc table", K(ret), K(table_key));
} else if (OB_FAIL(sstable->init(table_key))) {
LOG_WARN("fail to init sstable", K(ret), K(table_key));
} else if (OB_FAIL(sstable->open(meta))) {
LOG_WARN("fail to open sstable", K(ret), K(meta));
} else if (OB_FAIL(sstable->set_storage_file_handle(file_handle_))) {
LOG_WARN("fail to set storage file handle", K(ret));
} else if (OB_FAIL(handle.set_table(sstable))) {
LOG_WARN("fail to set table", K(ret));
}
if (OB_FAIL(ret) && nullptr != sstable) {
free_sstable(sstable);
sstable = nullptr;
handle.reset();
}
return ret;
}
void ObPGSSTableMgr::free_sstable(ObSSTable* sstable)
{
if (nullptr != sstable) {
OB_DELETE_ALIGN32(ObSSTable, unused, sstable);
}
}
int ObPGSSTableMgr::acquire_sstable(const ObITable::TableKey& table_key, ObTableHandle& table_handle)
{
int ret = OB_SUCCESS;
TableGetFunctor get_functor(table_handle);
ObBucketHashRLockGuard lock_guard(bucket_lock_, table_key.hash());
if (OB_UNLIKELY(!is_inited_)) {
ret = OB_NOT_INIT;
LOG_WARN("ObPGSSTableMgr has not been inited", K(ret));
} else if (OB_UNLIKELY(!table_key.is_valid() || !ObITable::is_sstable(table_key.table_type_))) {
ret = OB_INVALID_ARGUMENT;
LOG_WARN("invalid argument", K(ret), K(table_key));
} else if (OB_FAIL(table_map_.get(table_key, get_functor))) {
if (OB_HASH_NOT_EXIST != ret) {
LOG_WARN("fail to get table", K(ret), K(table_key));
} else {
ret = OB_ENTRY_NOT_EXIST;
}
}
return ret;
}
int ObPGSSTableMgr::get_all_sstables(ObTablesHandle& handle)
{
int ret = OB_SUCCESS;
ObITable* table = nullptr;
const int64_t lock_start_time = ObTimeUtility::current_time();
TableMap::Iterator iter(table_map_);
const int64_t lock_end_time = ObTimeUtility::current_time();
if (OB_UNLIKELY(!is_inited_)) {
ret = OB_NOT_INIT;
LOG_WARN("ObPGSSTableMgr has not been inited", K(ret));
} else if (OB_FAIL(handle.reserve(table_map_.get_count()))) {
LOG_WARN("fail to reserve tables", K(ret));
} else {
while (OB_SUCC(ret)) {
if (OB_FAIL(iter.get_next(table))) {
if (OB_ITER_END != ret) {
LOG_WARN("fail to get next table", K(ret));
} else {
ret = OB_SUCCESS;
break;
}
} else if (OB_ISNULL(table)) {
ret = OB_ERR_SYS;
LOG_WARN("error sys, table must not be null", K(ret));
} else if (OB_FAIL(handle.add_table(table))) {
LOG_WARN("fail to add table", K(ret));
}
}
}
const int64_t end_time = ObTimeUtility::current_time();
const int64_t lock_time = lock_end_time - lock_start_time;
const int64_t get_time = end_time - lock_end_time;
FLOG_INFO("get_all_sstables cost time",
K_(pg_key),
K(lock_time),
K(get_time),
"total_cost_time",
lock_time + get_time,
"total_cnt",
handle.get_count());
return ret;
}
int ObPGSSTableMgr::write_add_sstable_log(ObSSTable& sstable)
{
int ret = OB_SUCCESS;
ObStorageFile* file = NULL;
if (OB_UNLIKELY(!is_inited_)) {
ret = OB_NOT_INIT;
LOG_WARN("ObPGSSTableMgr has not been inited", K(ret));
} else if (!enable_write_log_) {
ret = OB_ERR_SYS;
LOG_WARN("can not write slog during replay", K(ret));
} else if (OB_ISNULL(file = file_handle_.get_storage_file())) {
ret = OB_ERR_UNEXPECTED;
LOG_ERROR("data file ptr is null", K(ret), KP(file));
} else {
ObFullMacroBlockMeta full_meta;
const ObSSTable::ObSSTableGroupMacroBlocks& macro_blocks = sstable.get_total_macro_blocks();
int64_t subcmd = 0;
ObStorageLogAttribute log_attr(pg_key_.get_tenant_id(), file->get_file_id());
// write sstable data macro blocks and lob macro blocks
for (int64_t i = 0; OB_SUCC(ret) && i < macro_blocks.count(); ++i) {
const MacroBlockId& block_id = macro_blocks.at(i);
if (OB_FAIL(sstable.get_meta(block_id, full_meta))) {
LOG_WARN("fail to get macro block meta", K(ret), K(block_id));
} else if (!full_meta.is_valid()) {
ret = OB_ERR_SYS;
LOG_WARN("error sys, macro block meta must not be null", K(ret), K(block_id));
} else if (ObMacroBlockCommonHeader::SSTableData != full_meta.meta_->attr_ &&
ObMacroBlockCommonHeader::LobIndex != full_meta.meta_->attr_ &&
ObMacroBlockCommonHeader::LobData != full_meta.meta_->attr_) {
ret = OB_ERR_SYS;
LOG_WARN("error sys, macro block type is invalid", K(ret), K(full_meta));
} else {
ObPGMacroBlockMetaLogEntry meta_log_entry(pg_key_,
sstable.get_key(),
0 /*data file id*/,
block_id.disk_no(),
block_id,
const_cast<ObMacroBlockMetaV2&>(*full_meta.meta_));
subcmd = ObIRedoModule::gen_subcmd(OB_REDO_LOG_PARTITION, REDO_LOG_CHANGE_MACRO_META);
if (OB_FAIL(SLOGGER.write_log(subcmd, log_attr, meta_log_entry))) {
LOG_WARN("fail to write macro block meta slog", K(ret));
}
}
}
// write bloomfilter macro block meta log
if (OB_SUCC(ret) && sstable.has_bloom_filter_macro_block()) {
const MacroBlockId& block_id = sstable.get_bloom_filter_block_id();
if (OB_FAIL(sstable.get_meta(block_id, full_meta))) {
LOG_WARN("fail to get macro block meta", K(ret), K(block_id));
} else if (!full_meta.is_valid()) {
ret = OB_ERR_SYS;
LOG_WARN("error sys, macro block meta must not be null", K(ret), K(block_id));
} else if (ObMacroBlockCommonHeader::BloomFilterData != full_meta.meta_->attr_) {
ret = OB_ERR_SYS;
LOG_WARN("error sys, macro block type is invalid", K(ret), K(full_meta));
} else {
ObPGMacroBlockMetaLogEntry meta_log_entry(pg_key_,
sstable.get_key(),
0 /*data file id*/,
block_id.disk_no(),
block_id,
const_cast<ObMacroBlockMetaV2&>(*full_meta.meta_));
subcmd = ObIRedoModule::gen_subcmd(OB_REDO_LOG_PARTITION, REDO_LOG_CHANGE_MACRO_META);
if (OB_FAIL(SLOGGER.write_log(subcmd, log_attr, meta_log_entry))) {
LOG_WARN("fail to write log", K(ret));
}
}
}
// write add sstable log
if (OB_SUCC(ret)) {
ObAddSSTableLogEntry sstable_log_entry(pg_key_, sstable);
int64_t subcmd = ObIRedoModule::gen_subcmd(OB_REDO_LOG_PARTITION, REDO_LOG_ADD_SSTABLE);
if (OB_FAIL(SLOGGER.write_log(subcmd, log_attr, sstable_log_entry))) {
LOG_WARN("fail to write sstable log", K(ret));
} else {
FLOG_INFO("write add sstable log", K(pg_key_), K(sstable));
}
}
}
return ret;
}
int ObPGSSTableMgr::write_remove_sstable_log(const ObITable::TableKey& table_key)
{
int ret = OB_SUCCESS;
ObStorageFile* file = NULL;
if (OB_UNLIKELY(!is_inited_)) {
ret = OB_NOT_INIT;
LOG_WARN("ObPGSSTableMgr has not been inited", K(ret));
} else if (!enable_write_log_) {
ret = OB_ERR_SYS;
LOG_WARN("can not write slog during replay", K(ret));
} else if (OB_UNLIKELY(!table_key.is_valid() || !ObITable::is_sstable(table_key.table_type_))) {
ret = OB_INVALID_ARGUMENT;
LOG_WARN("invalid arguments", K(ret), K(table_key));
} else if (OB_ISNULL(file = file_handle_.get_storage_file())) {
ret = OB_ERR_UNEXPECTED;
LOG_ERROR("data file ptr is null", K(ret), KP(file));
} else {
ObRemoveSSTableLogEntry sstable_log_entry(pg_key_, table_key);
ObStorageLogAttribute log_attr(pg_key_.get_tenant_id(), file->get_file_id());
int64_t subcmd = ObIRedoModule::gen_subcmd(OB_REDO_LOG_PARTITION, REDO_LOG_REMOVE_SSTABLE);
if (OB_FAIL(SLOGGER.write_log(subcmd, log_attr, sstable_log_entry))) {
LOG_WARN("fail to write sstable log", K(ret));
} else {
FLOG_INFO("write remove sstable log", K(pg_key_), K(table_key));
}
}
return ret;
}
int ObPGSSTableMgr::check_can_add_sstables(ObTablesHandle& tables_handle)
{
int ret = OB_SUCCESS;
if (OB_UNLIKELY(!is_inited_)) {
ret = OB_NOT_INIT;
LOG_WARN("ObPGSSTableMgr has not been inited", K(ret));
} else {
int hash_ret = OB_SUCCESS;
for (int64_t i = 0; OB_SUCC(ret) && i < tables_handle.get_count(); ++i) {
ObITable* table = tables_handle.get_table(i);
if (OB_ISNULL(table)) {
ret = OB_ERR_UNEXPECTED;
LOG_WARN("error unexpected, table must not be null", K(ret));
} else if (!table->is_sstable()) {
ret = OB_INVALID_ARGUMENT;
LOG_WARN("invalid arguments", K(ret), K(*table));
} else if (OB_HASH_NOT_EXIST != (hash_ret = table_map_.exist(table->get_key()))) {
if (OB_HASH_EXIST == hash_ret) {
ret = OB_ENTRY_EXIST;
LOG_WARN("sstable already exist in table map", K(ret), K(table->get_key()));
} else {
ret = hash_ret;
LOG_WARN("fail to get from table map", K(ret), K(table->get_key()));
}
}
}
}
return ret;
}
int ObPGSSTableMgr::add_sstables(const bool in_slog_trans, ObTablesHandle& tables_handle)
{
int ret = OB_SUCCESS;
ObArray<TableNode*> node_array;
TableNode* table_node = nullptr;
const bool is_write_lock = true;
int64_t lsn = 0;
if (OB_UNLIKELY(!is_inited_)) {
ret = OB_NOT_INIT;
LOG_WARN("ObPGSSTableMgr has not been inited", K(ret));
} else if (OB_UNLIKELY(!file_handle_.is_valid())) {
ret = OB_ERR_UNEXPECTED;
LOG_ERROR("storage file handle is invalid", K(ret), K(file_handle_));
} else if (OB_FAIL(file_handle_.get_storage_file()->fsync())) {
LOG_WARN("fail to fsync", K(ret), "file", *(file_handle_.get_storage_file()));
}
ObMultiBucketLockGuard lock_guard(bucket_lock_, is_write_lock);
if (OB_FAIL(ret)) {
} else if (OB_FAIL(lock_multi_bucket(lock_guard, tables_handle.get_tables()))) {
LOG_WARN("fail to lock multi bucket", K(ret));
} else if (OB_FAIL(check_can_add_sstables(tables_handle))) {
LOG_WARN("fail to check can add sstables", K(ret));
} else if (OB_FAIL(node_array.reserve(tables_handle.get_count()))) {
LOG_WARN("fail to reverse node array", K(ret));
} else if (!in_slog_trans && OB_FAIL(SLOGGER.begin(OB_LOG_TABLE_MGR))) {
LOG_WARN("fail to begin slog trans", K(ret));
} else {
for (int64_t i = 0; OB_SUCC(ret) && i < tables_handle.get_count(); ++i) {
ObSSTable* sstable = static_cast<ObSSTable*>(tables_handle.get_table(i));
if (OB_ISNULL(sstable)) {
ret = OB_INVALID_ARGUMENT;
LOG_WARN("invalid arguments", K(ret));
} else if (OB_FAIL(write_add_sstable_log(*sstable))) {
LOG_WARN("fail to write sstable log", K(ret));
} else if (OB_ISNULL(table_node = table_map_.alloc_node(*sstable))) {
ret = OB_ALLOCATE_MEMORY_FAILED;
LOG_WARN("fail to alloc memory for table node", K(ret));
} else if (OB_FAIL(node_array.push_back(table_node))) {
LOG_WARN("fail to push back table node", K(ret));
table_map_.free_node(table_node);
}
}
if (!in_slog_trans) {
if (OB_SUCC(ret)) {
if (OB_FAIL(SLOGGER.commit(lsn))) {
LOG_WARN("fail to commit slog", K(ret));
}
} else {
int tmp_ret = OB_SUCCESS;
if (OB_SUCCESS != (tmp_ret = SLOGGER.abort())) {
LOG_WARN("fail to abort slog", K(tmp_ret));
} else {
LOG_WARN("fail to add sstables", K(ret), K(lsn), K(tables_handle));
}
}
}
if (OB_SUCC(ret)) {
for (int64_t i = 0; OB_SUCC(ret) && i < node_array.count(); ++i) {
if (OB_FAIL(table_map_.put(*node_array.at(i)))) {
LOG_WARN("fail to put table map", K(ret));
ob_abort();
}
}
FLOG_INFO("finish add sstables", K(ret), K(lsn), K(tables_handle));
}
}
if (OB_FAIL(ret)) {
for (int64_t i = 0; i < node_array.count(); ++i) {
table_map_.free_node(node_array.at(i));
}
}
return ret;
}
int ObPGSSTableMgr::add_sstable(const bool in_slog_trans, ObTableHandle& table_handle)
{
int ret = OB_SUCCESS;
ObTablesHandle tables_handle;
if (OB_UNLIKELY(!is_inited_)) {
ret = OB_NOT_INIT;
LOG_WARN("ObPGSSTableMgr has not been inited", K(ret));
} else if (OB_FAIL(tables_handle.add_table(table_handle))) {
LOG_WARN("fail to add table", K(ret), K(table_handle));
} else if (OB_FAIL(add_sstables(in_slog_trans, tables_handle))) {
LOG_WARN("fail to add sstables", K(ret));
}
return ret;
}
int ObPGSSTableMgr::lock_multi_bucket(common::ObMultiBucketLockGuard& guard, common::ObIArray<ObITable*>& tables)
{
int ret = OB_SUCCESS;
ObArray<uint64_t> table_hash_array;
if (OB_UNLIKELY(!is_inited_)) {
ret = OB_NOT_INIT;
LOG_WARN("ObPGSSTableMgr has not been inited", K(ret));
} else if (0 == tables.count()) {
// do nothing
} else if (OB_FAIL(table_hash_array.reserve(tables.count()))) {
LOG_WARN("fail to reserve table hash array", K(ret));
} else {
for (int64_t i = 0; OB_SUCC(ret) && i < tables.count(); ++i) {
ObITable* table = tables.at(i);
if (OB_ISNULL(table)) {
ret = OB_INVALID_ARGUMENT;
LOG_WARN("invalid argument", K(ret));
} else if (OB_FAIL(table_hash_array.push_back(table->get_key().hash()))) {
LOG_WARN("fail to push back table hash array", K(ret));
}
}
if (OB_SUCC(ret)) {
if (OB_FAIL(guard.lock_multi_buckets(table_hash_array))) {
LOG_WARN("fail to lock multi buckets", K(ret));
}
}
}
return ret;
}
int ObPGSSTableMgr::recycle_unused_sstables(const int64_t max_recycle_cnt, int64_t& recycled_cnt)
{
int ret = OB_SUCCESS;
UnusedTableFunctor table_functor;
if (OB_UNLIKELY(!is_inited_)) {
ret = OB_NOT_INIT;
LOG_WARN("ObPGSSTableMgr has not been inited", K(ret));
} else if (!enable_write_log_) {
LOG_INFO("do not recycle unused sstable now");
} else if (OB_UNLIKELY(max_recycle_cnt < 0)) {
ret = OB_INVALID_ARGUMENT;
LOG_WARN("invalid argument", K(ret), K(max_recycle_cnt));
} else if (OB_FAIL(table_functor.init(max_recycle_cnt))) {
LOG_WARN("fail to init table functor", K(ret));
} else if (OB_FAIL(table_map_.foreach (table_functor))) {
LOG_WARN("fail to foreach table map", K(ret));
} else {
ObArray<ObITable*> unused_table_array;
const bool is_write_lock = true;
ObMultiBucketLockGuard lock_guard(bucket_lock_, is_write_lock);
if (OB_FAIL(lock_multi_bucket(lock_guard, table_functor.get_all_tables()))) {
LOG_WARN("fail to lock multi bucket", K(ret));
} else if (OB_FAIL(table_functor.get_tables(unused_table_array))) {
LOG_WARN("fail to get unused table array", K(ret));
} else if (0 == unused_table_array.count()) {
// no unused table, skip
} else if (OB_FAIL(SLOGGER.begin(OB_LOG_TABLE_MGR))) {
LOG_WARN("fail to begin transaction", K(ret));
} else {
recycled_cnt = unused_table_array.count();
for (int64_t i = 0; OB_SUCC(ret) && i < unused_table_array.count(); ++i) {
ObITable* table = unused_table_array.at(i);
if (OB_ISNULL(table)) {
ret = OB_ERR_SYS;
LOG_WARN("error sys, table must not be null", K(ret));
} else if (OB_FAIL(write_remove_sstable_log(table->get_key()))) {
LOG_WARN("fail to write remove sstable log", K(ret), K(table->get_key()));
}
}
if (OB_SUCC(ret)) {
int64_t lsn = 0;
if (OB_FAIL(SLOGGER.commit(lsn))) {
LOG_WARN("fail to commit slog", K(ret));
}
for (int64_t i = 0; OB_SUCC(ret) && i < unused_table_array.count(); ++i) {
ObITable* table = unused_table_array.at(i);
if (OB_FAIL(table_map_.erase(table->get_key()))) {
LOG_WARN("fail to erase from table map", K(ret), K(table->get_key()));
ob_abort();
} else {
free_sstable(static_cast<ObSSTable*>(table));
}
}
} else {
int tmp_ret = OB_SUCCESS;
if (OB_SUCCESS != (tmp_ret = SLOGGER.abort())) {
LOG_WARN("fail to abort slog", K(ret));
}
}
}
}
return ret;
}
int ObPGSSTableMgr::recycle_sstable(const ObITable::TableKey &table_key)
{
int ret = OB_SUCCESS;
if (OB_UNLIKELY(!is_inited_)) {
ret = OB_NOT_INIT;
LOG_WARN("ObPGSSTableMgr has not been inited", K(ret));
} else if (!enable_write_log_) {
LOG_INFO("do not recycle unused sstable now");
} else if (OB_UNLIKELY(!table_key.is_valid())) {
ret = OB_INVALID_ARGUMENT;
LOG_WARN("invalid argument", K(ret), K(table_key));
} else {
ObBucketHashWLockGuard lock_guard(bucket_lock_, table_key.hash());
ObITable *sstable = NULL;
if (OB_FAIL(table_map_.get(table_key, sstable))) {
if (OB_HASH_NOT_EXIST == ret) {
ret = OB_SUCCESS; // maybe this sstable has be removed.
} else {
LOG_WARN("fail to get sstable", K(ret), K(table_key));
}
} else if (OB_ISNULL(sstable)) {
ret = OB_ERR_SYS;
LOG_WARN("error sys, table must not be null", K(ret), KP(sstable));
} else if (OB_UNLIKELY(sstable->get_ref() > 0)) {
// do nothing.
} else {
int64_t lsn = 0;
if (OB_FAIL(SLOGGER.begin(OB_LOG_TABLE_MGR))) {
LOG_WARN("fail to begin transaction", K(ret));
} else if (OB_FAIL(write_remove_sstable_log(sstable->get_key()))) {
LOG_WARN("fail to write remove sstable log", K(ret), K(sstable->get_key()));
} else if (OB_FAIL(SLOGGER.commit(lsn))) {
LOG_WARN("fail to commit slog", K(ret));
} else if (OB_FAIL(table_map_.erase(sstable->get_key()))) {
LOG_WARN("fail to erase from table map", K(ret), K(sstable->get_key()));
abort();
} else {
free_sstable(static_cast<ObSSTable *>(sstable));
}
if (OB_FAIL(ret)) {
int tmp_ret = OB_SUCCESS;
if (OB_SUCCESS != (tmp_ret = SLOGGER.abort())) {
LOG_WARN("fail to abort slog", K(ret));
}
}
}
}
return ret;
}
int64_t ObPGSSTableMgr::get_serialize_size()
{
int ret = OB_SUCCESS;
int64_t serialize_size = 0;
if (OB_UNLIKELY(!is_inited_)) {
ret = OB_NOT_INIT;
LOG_WARN("ObPGSSTableMgr has not been inited", K(ret));
} else {
TableMap::Iterator iter(table_map_);
const int64_t table_cnt = table_map_.get_count();
serialize_size += serialization::encoded_length_i64(table_cnt);
ObITable* table = nullptr;
while (OB_SUCC(ret)) {
if (OB_FAIL(iter.get_next(table))) {
if (OB_ITER_END == ret) {
ret = OB_SUCCESS;
break;
} else {
LOG_WARN("fail to get next table", K(ret));
}
} else if (OB_ISNULL(table)) {
ret = OB_ERR_SYS;
LOG_WARN("error sys, table must not be null", K(ret));
} else {
ObSSTable* sstable = static_cast<ObSSTable*>(table);
serialize_size += sstable->get_serialize_size();
}
}
}
if (OB_FAIL(ret)) {
serialize_size = 0;
}
return serialize_size;
}
int ObPGSSTableMgr::serialize(
common::ObIAllocator& allocator, char*& buf, int64_t& serialize_size, ObTablesHandle* tables_handle)
{
int ret = OB_SUCCESS;
ObBucketTryRLockAllGuard guard(bucket_lock_);
int64_t request_size = 0;
if (OB_UNLIKELY(!is_inited_)) {
ret = OB_NOT_INIT;
LOG_WARN("ObPGSSTableMgr has not been inited", K(ret));
} else if (OB_FAIL(guard.get_ret())) {
LOG_WARN("fail to get read lock", K(ret));
} else if (FALSE_IT(request_size = get_serialize_size())) {
} else if (request_size <= 0) {
ret = OB_ERR_SYS;
LOG_WARN("error sys, serialize size must not be zero", K(request_size));
} else if (OB_ISNULL(buf = static_cast<char*>(allocator.alloc(request_size)))) {
ret = OB_ALLOCATE_MEMORY_FAILED;
LOG_WARN("fail to alloc memory", K(ret), K(request_size));
} else if (OB_FAIL(serialization::encode_i64(buf, request_size, serialize_size, table_map_.get_count()))) {
LOG_WARN("fail to encode table map size", K(ret));
} else {
TableMap::Iterator iter(table_map_);
ObITable* table = nullptr;
ObSSTable* sstable = nullptr;
while (OB_SUCC(ret)) {
if (OB_FAIL(iter.get_next(table))) {
if (OB_ITER_END == ret) {
ret = OB_SUCCESS;
break;
} else {
LOG_WARN("fail to get next table", K(ret));
}
} else if (OB_ISNULL(table)) {
ret = OB_ERR_SYS;
LOG_WARN("error sys, table must not be null", K(ret));
} else {
sstable = static_cast<ObSSTable*>(table);
if (OB_FAIL(sstable->serialize(buf, request_size, serialize_size))) {
LOG_WARN("fail to serialize sstable", K(ret));
} else if (nullptr != tables_handle) {
if (OB_FAIL(tables_handle->add_table(sstable))) {
LOG_WARN("fail to add sstable", K(ret));
}
}
}
}
}
if (OB_SUCC(ret)) {
LOG_INFO("succeed to serialize sstable mgr", K(pg_key_));
}
return ret;
}
int ObPGSSTableMgr::deserialize(const char* buf, const int64_t buf_len, int64_t& pos)
{
int ret = OB_SUCCESS;
int64_t table_cnt = 0;
if (OB_UNLIKELY(!is_inited_)) {
ret = OB_NOT_INIT;
LOG_WARN("ObPGSSTableMgr has not been inited", K(ret));
} else if (OB_FAIL(serialization::decode_i64(buf, buf_len, pos, &table_cnt))) {
LOG_WARN("fail to decode table cnt", K(ret));
} else if (OB_UNLIKELY(!file_handle_.is_valid())) {
ret = OB_ERR_UNEXPECTED;
LOG_ERROR("storage file handle is invalid", K(ret), K(file_handle_));
} else {
for (int64_t i = 0; OB_SUCC(ret) && i < table_cnt; ++i) {
ObSSTable sstable;
if (OB_FAIL(sstable.set_storage_file_handle(file_handle_))) {
LOG_WARN("fail to set storage file handle", K(ret), K(file_handle_));
} else if (OB_FAIL(sstable.deserialize(buf, buf_len, pos))) {
LOG_WARN("fail to deserialize sstable", K(ret));
} else if (OB_FAIL(add_sstable_to_map(sstable, false /*overwrite*/))) {
LOG_WARN("fail to add replay sstable", K(ret));
} else {
LOG_INFO("replay add sstable to map", K(pg_key_), K(sstable));
}
}
}
FLOG_INFO("deserialize pg sstable mgr", K(ret), K(table_cnt), K(pg_key_));
return ret;
}
int ObPGSSTableMgr::add_sstable_to_map(ObSSTable& sstable, const bool overwrite)
{
int ret = OB_SUCCESS;
ObSSTable* new_sstable = nullptr;
TableNode* table_node = nullptr;
if (OB_UNLIKELY(!is_inited_)) {
ret = OB_NOT_INIT;
LOG_WARN("ObPGSSTableMgr has not been inited", K(ret));
} else if (OB_UNLIKELY(!sstable.is_valid())) {
ret = OB_INVALID_ARGUMENT;
LOG_WARN("invalid argument", K(ret), K(sstable));
} else if (OB_FAIL(alloc_sstable(sstable.get_key().get_tenant_id(), new_sstable))) {
LOG_WARN("fail to alloc sstable", K(ret));
} else if (OB_FAIL(new_sstable->set_sstable(sstable))) {
LOG_WARN("fail to set sstable", K(ret));
} else if (OB_UNLIKELY(!new_sstable->is_valid())) {
ret = OB_ERR_SYS;
LOG_WARN("error sys, sstable must not be null", K(ret));
} else if (OB_ISNULL(table_node = table_map_.alloc_node(*new_sstable))) {
ret = OB_ALLOCATE_MEMORY_FAILED;
LOG_WARN("fail to alloc memory", K(ret));
} else if (OB_FAIL(table_map_.put(*table_node))) {
if (OB_HASH_EXIST != ret) {
LOG_WARN("fail to put node", K(ret), K(overwrite));
} else if (overwrite) {
const ObITable::TableKey& delete_key = table_node->get_key();
ObITable* delete_table = nullptr;
if (OB_FAIL(table_map_.erase(delete_key, delete_table))) {
LOG_WARN("erase table fail", K(ret), K(delete_key));
} else if (OB_FAIL(table_map_.put(*table_node))) {
LOG_WARN("put table fail", K(ret), K(delete_key), K(overwrite));
} else {
free_sstable(static_cast<ObSSTable*>(delete_table));
delete_table = nullptr;
table_node = nullptr;
new_sstable = nullptr;
}
} else {
ret = OB_SUCCESS;
}
} else {
table_node = nullptr;
new_sstable = nullptr;
}
if (nullptr != table_node) {
table_map_.free_node(table_node);
table_node = nullptr;
}
if (nullptr != new_sstable) {
free_sstable(new_sstable);
new_sstable = nullptr;
}
return ret;
}
int ObPGSSTableMgr::replay_add_sstable(ObSSTable& sstable)
{
int ret = OB_SUCCESS;
if (OB_UNLIKELY(!is_inited_)) {
ret = OB_NOT_INIT;
LOG_WARN("ObPGSSTableMgr has not been inited", K(ret));
} else if (OB_UNLIKELY(!sstable.is_valid())) {
ret = OB_INVALID_ARGUMENT;
LOG_WARN("invalid arguments", K(ret), K(sstable));
} else if (OB_FAIL(add_sstable_to_map(sstable, true /*overwrite*/))) {
LOG_WARN("fail to add sstable to map", K(ret));
}
return ret;
}
int ObPGSSTableMgr::replay_remove_sstable(const ObITable::TableKey& table_key)
{
int ret = OB_SUCCESS;
ObITable* table = nullptr;
ObSSTable* sstable = nullptr;
if (OB_UNLIKELY(!is_inited_)) {
ret = OB_NOT_INIT;
LOG_WARN("ObPGSSTableMgr has not been inited", K(ret));
} else if (OB_UNLIKELY(!table_key.is_valid())) {
ret = OB_INVALID_ARGUMENT;
LOG_WARN("invalid arguments", K(ret), K(table_key));
} else if (OB_FAIL(table_map_.erase(table_key, table))) {
if (OB_HASH_NOT_EXIST == ret) {
FLOG_INFO("erase not exist sstable when replay remove sstable", K(table_key));
ret = OB_SUCCESS;
} else {
LOG_WARN("fail to erase from map", K(ret), K(table_key));
}
} else {
sstable = static_cast<ObSSTable*>(table);
free_sstable(sstable);
sstable = nullptr;
}
return ret;
}
int ObPGSSTableMgr::alloc_sstable(const int64_t tenant_id, ObSSTable*& sstable)
{
int ret = OB_SUCCESS;
sstable = NULL;
ObMemAttr memattr(tenant_id);
memattr.label_ = ObModIds::OB_SSTABLE;
memattr.ctx_id_ = ObCtxIds::STORAGE_SHORT_TERM_META_CTX_ID;
if (OB_ISNULL(sstable = OB_NEW_ALIGN32(ObSSTable, memattr))) {
ret = OB_ALLOCATE_MEMORY_FAILED;
LOG_WARN("failed to alloc table", K(ret), K(memattr), K(tenant_id));
}
return ret;
}
int ObPGSSTableMgr::check_all_sstable_unused(bool& all_unused)
{
int ret = OB_SUCCESS;
CheckAllTableUnusedFunctor functor;
if (OB_UNLIKELY(!is_inited_)) {
ret = OB_NOT_INIT;
LOG_WARN("ObPGSSTableMgr has not been inited", K(ret));
} else if (OB_FAIL(table_map_.foreach (functor))) {
LOG_WARN("fail to foreach table map", K(ret));
} else {
all_unused = functor.is_all_unused();
}
return ret;
}
int ObPGSSTableMgr::enable_write_log()
{
int ret = OB_SUCCESS;
ObBucketWLockAllGuard guard(bucket_lock_);
LOG_INFO("enable write pg sstable table mgr log");
enable_write_log_ = true;
return ret;
}
int ObPGSSTableMgr::GetCleanOutLogIdFunctor::operator()(ObITable& table, bool& is_full)
{
int ret = OB_SUCCESS;
is_full = false;
if (table.get_ref() > 0 && table.is_multi_version_minor_sstable()) {
if (table.is_complement_minor_sstable()) {
min_complement_log_ts_ = std::min(min_complement_log_ts_, table.get_end_log_ts());
} else if (OB_FAIL(clean_out_log_ts_.push_back(table.get_end_log_ts()))) {
LOG_WARN("push into the clean out log ts array failed", K(ret));
} else {
// do nothing
}
}
return ret;
}
int ObPGSSTableMgr::get_clean_out_log_ts(ObIArray<int64_t> &clean_out_log_ts)
{
int ret = OB_SUCCESS;
if (OB_UNLIKELY(!is_inited_)) {
ret = OB_NOT_INIT;
LOG_WARN("ObPGSSTableMgr has not been inited", K(ret));
} else {
GetCleanOutLogIdFunctor functor;
if (OB_FAIL(table_map_.foreach (functor))) {
LOG_WARN("fail to foreach table map", K(ret));
} else {
functor.sort();
const ObIArray<int64_t> &log_ts_array = functor.get_clean_out_log_ts();
int64_t last_log_ts = -1;
// duplicate log ts may exist, need to remove duplicate
for (int64_t i = 0; OB_SUCC(ret) && i < log_ts_array.count(); i++) {
if (log_ts_array.at(i) <= functor.get_min_complement_log_ts()) {
if (last_log_ts != log_ts_array.at(i)) {
if (OB_FAIL(clean_out_log_ts.push_back(log_ts_array.at(i)))) {
LOG_WARN("failed to push back log ts", K(ret));
} else {
last_log_ts = log_ts_array.at(i);
}
}
}
}
}
}
return ret;
}
Reference in New Issue View Git Blame Copy Permalink