clear black servers when reload backup tasks

This commit is contained in:
hamstersox
2023-08-18 08:44:13 +00:00
committed by ob-robot
parent 3b2ab28340
commit 576d0ab5fe
3 changed files with 8 additions and 4 deletions

View File

@ -311,7 +311,7 @@ int ObBackupDataLSTaskMgr::handle_execute_over(
} else if (OB_FALSE_IT(new_ls_attr.end_ts_ = ObTimeUtility::current_time())) { } else if (OB_FALSE_IT(new_ls_attr.end_ts_ = ObTimeUtility::current_time())) {
} else if (OB_FAIL(ObBackupUtils::get_full_replica_num(ls_attr.tenant_id_, full_replica_num))) { } else if (OB_FAIL(ObBackupUtils::get_full_replica_num(ls_attr.tenant_id_, full_replica_num))) {
LOG_WARN("failed to get full replica num", K(ret)); LOG_WARN("failed to get full replica num", K(ret));
} else if (new_ls_attr.black_servers_.count() + 1 == full_replica_num && OB_FALSE_IT(new_ls_attr.black_servers_.reset())) { } else if (new_ls_attr.black_servers_.count() + 1 >= full_replica_num && OB_FALSE_IT(new_ls_attr.black_servers_.reset())) {
} else if (OB_SUCCESS != result_info.result_ && OB_FAIL(new_ls_attr.black_servers_.push_back(result_info.addr_))) { } else if (OB_SUCCESS != result_info.result_ && OB_FAIL(new_ls_attr.black_servers_.push_back(result_info.addr_))) {
LOG_WARN("failed to push back black server", K(ret), K(result_info)); LOG_WARN("failed to push back black server", K(ret), K(result_info));
} else if (OB_FALSE_IT(new_ls_attr.result_ = result_info.result_)) { } else if (OB_FALSE_IT(new_ls_attr.result_ = result_info.result_)) {

View File

@ -127,17 +127,18 @@ int ObBackupDataScheduler::get_need_reload_task(
int ObBackupDataScheduler::do_get_need_reload_task_( int ObBackupDataScheduler::do_get_need_reload_task_(
const ObBackupJobAttr &job, const ObBackupJobAttr &job,
const ObBackupSetTaskAttr &set_task_attr, const ObBackupSetTaskAttr &set_task_attr,
const ObIArray<ObBackupLSTaskAttr> &ls_tasks, ObIArray<ObBackupLSTaskAttr> &ls_tasks,
common::ObIAllocator &allocator, common::ObIAllocator &allocator,
ObIArray<ObBackupScheduleTask *> &tasks) ObIArray<ObBackupScheduleTask *> &tasks)
{ {
int ret = OB_SUCCESS; int ret = OB_SUCCESS;
int64_t full_replica_num = 0;
if (!job.is_valid() || ls_tasks.empty()) { if (!job.is_valid() || ls_tasks.empty()) {
ret = OB_INVALID_ARGUMENT; ret = OB_INVALID_ARGUMENT;
LOG_WARN("[DATA_BACKUP]invalid argument", K(ret), K(job), K(ls_tasks)); LOG_WARN("[DATA_BACKUP]invalid argument", K(ret), K(job), K(ls_tasks));
} else { } else {
for (int64_t i = 0; OB_SUCC(ret) && i < ls_tasks.count(); ++i) { for (int64_t i = 0; OB_SUCC(ret) && i < ls_tasks.count(); ++i) {
const ObBackupLSTaskAttr &ls_task = ls_tasks.at(i); ObBackupLSTaskAttr &ls_task = ls_tasks.at(i);
ObBackupScheduleTask *task = nullptr; ObBackupScheduleTask *task = nullptr;
bool is_dropped = false; bool is_dropped = false;
if (OB_FAIL(ObBackupDataLSTaskMgr::check_ls_is_dropped(ls_task, *sql_proxy_, is_dropped))) { if (OB_FAIL(ObBackupDataLSTaskMgr::check_ls_is_dropped(ls_task, *sql_proxy_, is_dropped))) {
@ -152,6 +153,9 @@ int ObBackupDataScheduler::do_get_need_reload_task_(
*backup_service_, *sql_proxy_, ls_task, result_info))) { *backup_service_, *sql_proxy_, ls_task, result_info))) {
LOG_WARN("failed to handle execute over", K(ret), K(ls_task)); LOG_WARN("failed to handle execute over", K(ret), K(ls_task));
} }
} else if (OB_FAIL(ObBackupUtils::get_full_replica_num(ls_task.tenant_id_, full_replica_num))) {
LOG_WARN("failed to get full replica num", K(ret));
} else if (ls_task.black_servers_.count() >= full_replica_num && OB_FALSE_IT(ls_task.black_servers_.reset())) {
} else if (OB_FAIL(build_task_(job, set_task_attr, ls_task, allocator, task))) { } else if (OB_FAIL(build_task_(job, set_task_attr, ls_task, allocator, task))) {
LOG_WARN("[DATA_BACKUP]failed to build task", K(ret), K(job), K(ls_task)); LOG_WARN("[DATA_BACKUP]failed to build task", K(ret), K(job), K(ls_task));
} else if (OB_ISNULL(task)) { } else if (OB_ISNULL(task)) {

View File

@ -82,7 +82,7 @@ private:
int get_need_cancel_tenants_(const uint64_t tenant_id, const common::ObIArray<uint64_t> &backup_tenant_ids, int get_need_cancel_tenants_(const uint64_t tenant_id, const common::ObIArray<uint64_t> &backup_tenant_ids,
common::ObIArray<uint64_t> &need_cancel_backup_tenants); common::ObIArray<uint64_t> &need_cancel_backup_tenants);
int do_get_need_reload_task_(const share::ObBackupJobAttr &job, const share::ObBackupSetTaskAttr &set_task_attr, int do_get_need_reload_task_(const share::ObBackupJobAttr &job, const share::ObBackupSetTaskAttr &set_task_attr,
const ObIArray<share::ObBackupLSTaskAttr> &ls_tasks, ObIArray<share::ObBackupLSTaskAttr> &ls_tasks,
common::ObIAllocator &allocator, ObIArray<ObBackupScheduleTask *> &tasks); common::ObIAllocator &allocator, ObIArray<ObBackupScheduleTask *> &tasks);
int build_task_(const share::ObBackupJobAttr &job, const share::ObBackupSetTaskAttr &set_task_attr, int build_task_(const share::ObBackupJobAttr &job, const share::ObBackupSetTaskAttr &set_task_attr,
const share::ObBackupLSTaskAttr &ls_task, const share::ObBackupLSTaskAttr &ls_task,