hcq/oceanbase
1
0
Fork 0
Code Issues Pull Requests Actions 2 Packages Projects Releases Wiki Activity

Upgrade compatibility of backup and restore

Browse Source
This commit is contained in:
hamstersox
2023-03-04 07:11:08 +00:00
committed by ob-robot
parent 5cd4dcc8b4
commit 59fe65cfe2
20 changed files with 418 additions and 54 deletions
Download Patch File Download Diff File Expand all files Collapse all files

122
src/rootserver/backup/ob_backup_data_scheduler.cpp
View File

@ -610,10 +610,15 @@ int ObBackupDataScheduler::start_tenant_backup_data_(const ObBackupJobAttr &job_
LOG_WARN("log archive is not doing, can't start backup", K(ret));
} else {
ObMySQLTransaction trans;
if (OB_FAIL(trans.start(sql_proxy_, gen_meta_tenant_id(new_job_attr.tenant_id_)))) {
uint64_t cluster_version = 0;
if (OB_FAIL(ObShareUtil::fetch_current_cluster_version(*sql_proxy_, cluster_version))) {
LOG_WARN("failed to get cluster version", K(ret));
} else if (OB_FAIL(trans.start(sql_proxy_, gen_meta_tenant_id(new_job_attr.tenant_id_)))) {
LOG_WARN("[DATA_BACKUP]failed to start trans", K(ret), K(job_attr));
} else {
if (OB_FAIL(get_next_job_id(trans, new_job_attr.tenant_id_, new_job_attr.job_id_))) {
if (OB_FAIL(persist_backup_version_(trans, new_job_attr.tenant_id_, cluster_version))) {
LOG_WARN("failed to persist backup version", K(ret));
} else if (OB_FAIL(get_next_job_id(trans, new_job_attr.tenant_id_, new_job_attr.job_id_))) {
LOG_WARN("[DATA_BACKUP]failed to get next job id", K(ret));
} else if (OB_FAIL(get_next_backup_set_id(trans, new_job_attr.tenant_id_, new_job_attr.backup_set_id_))) {
LOG_WARN("[DATA_BACKUP]failed to get next backup set id", K(ret));
@ -647,29 +652,81 @@ int ObBackupDataScheduler::start_tenant_backup_data_(const ObBackupJobAttr &job_
return ret;
}
int ObBackupDataScheduler::get_scn(common::ObISQLClient &sql_proxy, const uint64_t tenant_id, SCN &scn)
int ObBackupDataScheduler::persist_backup_version_(common::ObISQLClient &sql_proxy, const uint64_t tenant_id, const uint64_t &cluster_version)
{
int ret = OB_SUCCESS;
uint64_t data_version = 0;
uint64_t exec_tenant_id = gen_meta_tenant_id(tenant_id);
if (GCONF.enable_upgrade_mode) {
ret = OB_BACKUP_CAN_NOT_START;
LOG_USER_ERROR(OB_BACKUP_CAN_NOT_START, "cluster upgrading");
LOG_WARN("cluster upgrade, can't start backup", K(ret), K(tenant_id));
} else if (OB_FAIL(ObShareUtil::fetch_current_data_version(sql_proxy, exec_tenant_id, data_version))) {
LOG_WARN("failed to get data version", K(ret), K(exec_tenant_id));
} else if (OB_FAIL(ObLSBackupInfoOperator::set_backup_version(sql_proxy, exec_tenant_id, data_version))) {
LOG_WARN("failed to set backup version", K(ret), K(exec_tenant_id), K(data_version));
} else if (OB_FAIL(ObLSBackupInfoOperator::set_cluster_version(sql_proxy, exec_tenant_id, cluster_version))) {
LOG_WARN("failed to set set cluster version", K(ret), K(exec_tenant_id));
}
return ret;
}
int ObBackupDataScheduler::get_backup_scn(
common::ObISQLClient &sql_proxy, const uint64_t tenant_id, const bool is_start, SCN &scn)
{
int ret = OB_SUCCESS;
ObAllTenantInfo tenant_info;
const bool for_update = false;
if (OB_FAIL(ObAllTenantInfoProxy::load_tenant_info(tenant_id, &sql_proxy, for_update, tenant_info))) {
ObMySQLTransaction trans;
const uint64_t exec_tenant_id = gen_meta_tenant_id(tenant_id);
if (OB_FAIL(trans.start(&sql_proxy, exec_tenant_id))) {
LOG_WARN("failed to start trans", K(ret), K(exec_tenant_id));
} else if (OB_FAIL(ObAllTenantInfoProxy::load_tenant_info(tenant_id, &trans, true/*for update*/, tenant_info))) {
LOG_WARN("failed to get tenant info", K(ret), K(tenant_id));
} else {
// TODO when provide the scn_to_str, remove this .
// The conversion accuracy of SCN to time_stamp is inconsistent under MySQL mode and Oracle mode.
// The conversion accuracy in ORALCE mode is nanosecond, but it is microsecond in mysql
// for backup and restore, we keep the scn round up to microseconds that keep the conversion accuracy is consistent.
// meanwhile, in order to solve that boundary is not included in the restore, scn + 1;
// 1658475549197665190 --> 1658475549197666000
SCN tmp_scn;
int64_t ts;
ts = tenant_info.get_standby_scn().convert_to_ts();
if (OB_FAIL(tmp_scn.convert_from_ts(ts))) {
LOG_WARN("fail to convert from ts", K(ret), K(ts));
} else if (tenant_info.get_standby_scn() != tmp_scn && OB_FAIL(tmp_scn.convert_from_ts(ts + 1))) {
LOG_WARN("fail to convert from ts", K(ret), K(ts));
if (tenant_info.is_primary()) {
// for parmary tenant, the backup start scn get from gts as the same as end scn.
if (OB_FAIL(ObBackupUtils::get_backup_scn(tenant_id, tmp_scn))) {
LOG_WARN("failed to get gts", K(ret), K(tenant_id));
}
} else if (tenant_info.is_standby()) {
// for standby tenant, the backup start scn must be the readable scn, so it get from sts.
// but end scn must be replayable scn, so it get from the tenant_info.
if (is_start) {
if (OB_FAIL(ObBackupUtils::get_backup_scn(tenant_id, tmp_scn))) {
LOG_WARN("failed to get gts", K(ret), K(tenant_id));
}
} else {
tmp_scn = tenant_info.get_replayable_scn();
}
} else {
scn = tmp_scn;
ret = OB_STATE_NOT_MATCH;
LOG_WARN("tenant role not match", K(ret), K(tenant_info));
}
if (OB_SUCC(ret)) {
if (OB_FAIL(trans.end(true))) {
LOG_WARN("failed to commit", K(ret));
} else {
// TODO when provide the scn_to_str, remove this .
// The conversion accuracy of SCN to time_stamp is inconsistent under MySQL mode and Oracle mode.
// The conversion accuracy in ORALCE mode is nanosecond, but it is microsecond in mysql
// for backup and restore, we keep the scn round up to microseconds that keep the conversion accuracy is consistent.
// meanwhile, in order to solve that boundary is not included in the restore, scn + 1;
// 1658475549197665190 --> 1658475549197666000
int64_t ts = 0;
ts = tmp_scn.convert_to_ts();
if (OB_FAIL(scn.convert_from_ts(ts))) {
LOG_WARN("fail to convert from ts", K(ret), K(ts));
} else if (tmp_scn != scn && OB_FAIL(scn.convert_from_ts(ts + 1))) {
LOG_WARN("fail to convert from ts", K(ret), K(ts));
}
}
} else {
int tmp_ret = OB_SUCCESS;
if (OB_SUCCESS != (tmp_ret = trans.end(false))) {
LOG_WARN("failed to roll back", K(ret), K(tmp_ret));
}
}
}
return ret;
@ -1227,8 +1284,23 @@ int ObUserTenantBackupJobMgr::report_failed_to_initiator_()
int ObUserTenantBackupJobMgr::check_can_backup_()
{
int ret = OB_SUCCESS;
uint64_t data_version = 0;
uint64_t cluster_version = 0;
uint64_t exec_tenant_id = gen_meta_tenant_id(job_attr_->tenant_id_);
if (share::ObBackupStatus::CANCELING == job_attr_->status_.status_) {
// backup job is canceling, no need to check log archive status
} else if (GCONF.enable_upgrade_mode) {
ret = OB_STATE_NOT_MATCH;
LOG_WARN("cluster is upgrade, backup can't continue", K(ret), KPC(job_attr_));
} else if (OB_FAIL(share::ObLSBackupInfoOperator::get_backup_version(*sql_proxy_, exec_tenant_id, data_version))) {
LOG_WARN("failed to get backup version", K(ret), K(exec_tenant_id));
} else if (OB_FAIL(share::ObLSBackupInfoOperator::get_cluster_version(*sql_proxy_, exec_tenant_id, cluster_version))) {
LOG_WARN("failed to get cluster version", K(ret), K(exec_tenant_id));
} else if (cluster_version != GET_MIN_CLUSTER_VERSION()) {
ret = OB_VERSION_NOT_MATCH;
LOG_WARN("cluster version not match, backup can't continue", K(ret), K(cluster_version));
} else if (OB_FAIL(ObBackupUtils::check_tenant_data_version_match(exec_tenant_id, data_version))) {
LOG_WARN("failed to check tenant data version", K(ret), K(exec_tenant_id), K(data_version));
} else {
ObTenantArchiveRoundAttr round_attr;
if (OB_FAIL(ObTenantArchiveMgr::get_tenant_current_round(job_attr_->tenant_id_, job_attr_->incarnation_id_, round_attr))) {
@ -1403,7 +1475,8 @@ int ObUserTenantBackupJobMgr::insert_backup_set_task_(common::ObISQLClient &sql_
LOG_WARN("[DATA_BACKUP]failed to assign backup dest", K(ret), KPC(job_attr_));
} else if (OB_FAIL(get_next_task_id_(sql_proxy, backup_set_task.task_id_))) {
LOG_WARN("[DATA_BACKUP]failed to get next task id", K(ret));
} else if (OB_FAIL(ObBackupDataScheduler::get_scn(sql_proxy, job_attr_->tenant_id_, backup_set_task.start_scn_))) {
} else if (OB_FAIL(ObBackupDataScheduler::get_backup_scn(
sql_proxy, job_attr_->tenant_id_, true/*start scn*/, backup_set_task.start_scn_))) {
LOG_WARN("fail t get start scn", K(ret));
} else {
backup_set_task.tenant_id_ = job_attr_->tenant_id_;
@ -1461,6 +1534,8 @@ int ObUserTenantBackupJobMgr::fill_backup_set_desc_(
ObBackupSetFileDesc &backup_set_desc)
{
int ret = OB_SUCCESS;
uint64_t data_version = 0;
uint64_t cluster_version = 0;
if (!job_attr.is_valid() || prev_inc_backup_set_id < 0 || prev_full_backup_set_id < 0) {
ret = OB_INVALID_ARGUMENT;
LOG_WARN("[DATA_BACKUP]invalid argument", K(ret), K(job_attr), K(prev_inc_backup_set_id), K(prev_full_backup_set_id));
@ -1470,6 +1545,10 @@ int ObUserTenantBackupJobMgr::fill_backup_set_desc_(
LOG_WARN("[DATA_BACKUP]failed to assign passwd", K(ret));
} else if (OB_FAIL(backup_set_desc.backup_path_.assign(job_attr.backup_path_))) {
LOG_WARN("[DATA_BACKUP]failed to assign backup dest", K(ret));
} else if (OB_FAIL(ObShareUtil::fetch_current_data_version(*sql_proxy_, job_attr.tenant_id_, data_version))) {
LOG_WARN("failed to get min data version", K(ret));
} else if (OB_FAIL(ObShareUtil::fetch_current_cluster_version(*sql_proxy_, cluster_version))) {
LOG_WARN("failed to get cluster version", K(ret));
} else {
backup_set_desc.backup_set_id_ = job_attr.backup_set_id_;
backup_set_desc.tenant_id_ = job_attr.tenant_id_;
@ -1487,8 +1566,9 @@ int ObUserTenantBackupJobMgr::fill_backup_set_desc_(
backup_set_desc.encryption_mode_ = job_attr.encryption_mode_;
backup_set_desc.start_replay_scn_ = SCN::min_scn();
backup_set_desc.min_restore_scn_ = SCN::min_scn();
backup_set_desc.backup_compatible_ = ObBackupSetFileDesc::Compatible::COMPATIBLE_VERSION_1;
backup_set_desc.tenant_compatible_ = GET_MIN_CLUSTER_VERSION();
backup_set_desc.backup_compatible_ = ObBackupSetFileDesc::Compatible::COMPATIBLE_VERSION_2;
backup_set_desc.tenant_compatible_ = data_version;
backup_set_desc.cluster_version_ = cluster_version;
backup_set_desc.plus_archivelog_ = job_attr.plus_archivelog_;
}
return ret;