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

fixed switch to follower gracefully failed because of palf has changed leader.

Browse Source
This commit is contained in:
obdev 2022-10-26 11:54:16 +00:00 committed by wangzelin.wzl
parent 06264f4e7b
commit d441da9118
2 changed files with 6 additions and 47 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

49
src/logservice/rcservice/ob_role_change_service.cpp
View File

@ -541,18 +541,18 @@ int ObRoleChangeService::switch_leader_to_follower_gracefully_(
// we must ensure that the 'end_lsn' provid by 'apply_service_' is correctly.
// just switch_role to follower firstly, avoid sync log failed because palf has changed leader.
} else if (FALSE_IT(log_handler->switch_role(new_role, curr_proposal_id))) {
} else if (FALSE_IT(time_guard.click("wait_apply_service_apply_done_when_change_leader_"))
|| OB_FAIL(wait_apply_service_apply_done_when_change_leader_(log_handler, curr_proposal_id, ls_id, end_lsn))) {
CLOG_LOG(WARN, "wait_apply_service_apply_done_when_change_leader_ failed", K(ret),
} else if (FALSE_IT(time_guard.click("wait_apply_service_apply_done_"))
|| OB_FAIL(wait_apply_service_apply_done_(ls_id, end_lsn))) {
CLOG_LOG(WARN, "wait_apply_service_apply_done_ failed", K(ret),
K(new_role), K(new_proposal_id), K(dst_addr));
// apply service will not update end_lsn after switch_to_follower, so wait apply done first here
} else if (FALSE_IT(time_guard.click("apply_service->switch_to_follower"))
|| OB_FAIL(apply_service_->switch_to_follower(ls_id))) {
CLOG_LOG(WARN, "apply_service_ switch_to_follower failed", K(ret), K(new_role), K(new_proposal_id), K(dst_addr));
CLOG_LOG(ERROR, "apply_service_ switch_to_follower failed", K(ret), K(new_role), K(new_proposal_id), K(dst_addr));
} else if (FALSE_IT(time_guard.click("replay_service->switch_to_follower"))
|| OB_FAIL(replay_service_->switch_to_follower(ls_id, end_lsn))) {
CLOG_LOG(WARN, "replay_service_ switch_to_follower failed", K(ret), KPC(ls), K(new_role), K(new_proposal_id));
// NB: execute 'change_leader_to' lastly, can make 'wait_apply_service_apply_done_when_change_leader_' finish quickly.
CLOG_LOG(ERROR, "replay_service_ switch_to_follower failed", K(ret), KPC(ls), K(new_role), K(new_proposal_id));
// NB: execute 'change_leader_to' lastly, can make 'wait_apply_service_apply_done_' finish quickly.
} else if (OB_FAIL(log_handler->change_leader_to(dst_addr))) {
CLOG_LOG(WARN, "ObLogHandler change_leader failed", K(ret), K(new_role), K(new_proposal_id), K(dst_addr));
} else {
@ -720,43 +720,6 @@ int ObRoleChangeService::wait_apply_service_apply_done_(
return ret;
}
int ObRoleChangeService::wait_apply_service_apply_done_when_change_leader_(
const ObLogHandler *log_handler,
const int64_t proposal_id,
const share::ObLSID &ls_id,
palf::LSN &end_lsn)
{
int ret = OB_SUCCESS;
bool is_done = false;
palf::LSN max_lsn;
common::ObRole unused_curr_role;
int64_t unused_curr_proposal_id;
common::ObRole new_role;
int64_t new_proposal_id;
bool is_pending_state = false;
while (OB_SUCC(ret) && (false == is_done || end_lsn != max_lsn)) {
if (OB_FAIL(apply_service_->is_apply_done(ls_id, is_done, end_lsn))) {
CLOG_LOG(WARN, "apply_service_ is_apply_done failed", K(ret), K(is_done), K(end_lsn));
} else if (OB_FAIL(log_handler->get_max_lsn(max_lsn))) {
CLOG_LOG(WARN, "get_end_lsn from palf failed", K(ret), K(ls_id), K(end_lsn));
} else if (OB_FAIL(log_handler->prepare_switch_role(unused_curr_role, unused_curr_proposal_id,
new_role, new_proposal_id, is_pending_state))) {
CLOG_LOG(WARN, "failed prepare_switch_role", K(ret), K(new_role), K(proposal_id), K(ls_id));
// if palf has changed role, return OB_STATE_NOT_MATCH, change leader failed.
} else if (LEADER != new_role || proposal_id != new_proposal_id) {
ret = OB_STATE_NOT_MATCH;
CLOG_LOG(WARN, "palf has changed leader, wait_apply_service_apply_done_when_change_leader_ failed", K(ret), K(proposal_id),
K(new_proposal_id));
} else if (false == is_done || end_lsn != max_lsn) {
CLOG_LOG(INFO, "wait apply done return false, need retry", K(ls_id), K(is_done),
K(end_lsn), K(max_lsn));
ob_usleep(5*1000);
} else {
}
}
return ret;
}
ObRoleChangeService::RoleChangeOptType ObRoleChangeService::get_role_change_opt_type_(
const ObRole &old_role,
const ObRole &new_role,

4
src/logservice/rcservice/ob_role_change_service.h
View File

@ -107,10 +107,6 @@ private:
const palf::LSN &end_lsn);
int wait_apply_service_apply_done_(const share::ObLSID &ls_id,
palf::LSN &end_lsn);
int wait_apply_service_apply_done_when_change_leader_(const ObLogHandler *log_handler,
const int64_t proposal_id,
const share::ObLSID &ls_id,
palf::LSN &end_lsn);
bool check_need_execute_role_change_(const int64_t curr_proposal_id, const common::ObRole &curr_role,
const int64_t new_proposal_id, const common::ObRole &new_role) const;
private: