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

revert create table wait leader

Browse Source
This commit is contained in:
zs0 2021-07-28 13:51:14 +08:00 committed by wangzelin.wzl
parent 52a81d2861
commit 3209421d5c
3 changed files with 35 additions and 88 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

97
src/rootserver/ob_ddl_service.cpp
View File

@ -10834,7 +10834,6 @@ int ObDDLService::construct_create_partition_creator(const common::ObPartitionKe
// there is no persistent member list, standby_restore will have problems
} else {
int64_t restore = REPLICA_NOT_RESTORE;
common::ObRole role = FOLLOWER;
if (OB_CREATE_TABLE_MODE_RESTORE == create_mode &&
ObReplicaTypeCheck::is_replica_with_ssstore(a->replica_type_)) {
// logical restore
@ -10849,20 +10848,18 @@ int ObDDLService::construct_create_partition_creator(const common::ObPartitionKe
} else {
restore = REPLICA_NOT_RESTORE;
}
if (OB_FAIL(set_flag_role(a->initial_leader_, is_standby, restore, table_id, role))) {
LOG_WARN("fail to set flag role", KR(ret), K(is_standby), K(restore), K(table_id), K(role));
} else if (OB_FAIL(fill_create_partition_arg(table_id,
partition_cnt,
paxos_replica_num,
non_paxos_replica_num,
partition_id,
*a,
now,
is_bootstrap,
is_standby,
restore,
frozen_status,
arg))) {
if (OB_FAIL(fill_create_partition_arg(table_id,
partition_cnt,
paxos_replica_num,
non_paxos_replica_num,
partition_id,
*a,
now,
is_bootstrap,
is_standby,
restore,
frozen_status,
arg))) {
LOG_WARN("fail to fill ObCreatePartitionArg",
K(ret),
K(table_id),
@ -10874,7 +10871,7 @@ int ObDDLService::construct_create_partition_creator(const common::ObPartitionKe
*a,
"timestamp",
now);
} else if (OB_FAIL(fill_flag_replica(table_id, partition_cnt, partition_id, arg, *a, role, flag_replica))) {
} else if (OB_FAIL(fill_flag_replica(table_id, partition_cnt, partition_id, arg, *a, flag_replica))) {
LOG_WARN("fail to fill flag replica",
K(ret),
K(table_id),
@ -11030,27 +11027,8 @@ int ObDDLService::fill_create_partition_arg(const uint64_t table_id, const int64
return ret;
}
int ObDDLService::set_flag_role(const bool initial_leader, const bool is_standby, const int64_t restore,
const uint64_t table_id, common::ObRole& role)
{
int ret = OB_SUCCESS;
if (OB_FAIL(check_inner_stat())) {
LOG_WARN("variable is not init", KR(ret));
} else {
if (initial_leader && !is_standby && REPLICA_NOT_RESTORE == restore && !is_sys_table(table_id)) {
// When setting the flag role, only the non-restore leader of the primary cluster is considered,
// and the rest are reported normally
role = LEADER;
} else {
role = FOLLOWER;
}
}
return ret;
}
int ObDDLService::fill_flag_replica(const uint64_t table_id, const int64_t partition_cnt, const int64_t partition_id,
const ObCreatePartitionArg& arg, const ObReplicaAddr& replica_addr, const common::ObRole role,
share::ObPartitionReplica& flag_replica)
const ObCreatePartitionArg& arg, const ObReplicaAddr& replica_addr, share::ObPartitionReplica& flag_replica)
{
int ret = OB_SUCCESS;
if (OB_FAIL(check_inner_stat())) {
@ -11088,7 +11066,6 @@ int ObDDLService::fill_flag_replica(const uint64_t table_id, const int64_t parti
replica_addr.replica_type_);
} else {
flag_replica.is_restore_ = arg.restore_;
flag_replica.role_ = role;
if (arg.ignore_member_list_) {
// not fill member list
} else if (REPLICA_RESTORE_DATA == flag_replica.is_restore_ ||
@ -15121,7 +15098,6 @@ int ObDDLService::create_tablegroup_partitions_for_create(const share::schema::O
int64_t non_paxos_replica_num = OB_INVALID_COUNT;
const uint64_t tenant_id = tablegroup_schema.get_tenant_id();
const uint64_t tablegroup_id = tablegroup_schema.get_tablegroup_id();
bool is_standby = false;
if (OB_UNLIKELY(OB_INVALID_ID == tenant_id || !frozen_status.is_valid())) {
ret = OB_INVALID_ARGUMENT;
LOG_WARN("invalid argument", K(ret), K(tenant_id), K(frozen_status));
@ -15139,8 +15115,6 @@ int ObDDLService::create_tablegroup_partitions_for_create(const share::schema::O
ret = OB_ERR_UNEXPECTED;
LOG_WARN(
"array count unexpected", K(ret), "tg_addr_count", tablegroup_addr.count(), "part_ids_count", part_ids.count());
} else if (OB_FAIL(get_is_standby_cluster(is_standby))) {
LOG_WARN("fail to get cluster", KR(ret), K(is_standby), K(tablegroup_schema));
} else {
obrpc::ObCreatePartitionArg arg;
share::ObSplitPartition split_info;
@ -15165,7 +15139,6 @@ int ObDDLService::create_tablegroup_partitions_for_create(const share::schema::O
FOREACH_CNT_X(a, part_addr, OB_SUCC(ret))
{
ObPartitionReplica flag_replica;
common::ObRole role = FOLLOWER;
if (OB_UNLIKELY(nullptr == a || nullptr == rpc_proxy_)) {
ret = OB_ERR_UNEXPECTED;
LOG_WARN("addr or rpc_proxy is null", K(ret));
@ -15181,27 +15154,24 @@ int ObDDLService::create_tablegroup_partitions_for_create(const share::schema::O
} else {
restore = REPLICA_NOT_RESTORE;
}
if (OB_FAIL(set_flag_role(a->initial_leader_, is_standby, restore, tablegroup_id, role))) {
LOG_WARN("fail to set flag role", KR(ret));
} else if (OB_FAIL(fill_create_partition_arg(tablegroup_id,
tablegroup_schema.get_partition_cnt(),
paxos_replica_num,
non_paxos_replica_num,
partition_id,
*a,
now,
false /*is_bootstrap*/,
false /*is_standby*/,
restore,
frozen_status,
arg))) {
if (OB_FAIL(fill_create_partition_arg(tablegroup_id,
tablegroup_schema.get_partition_cnt(),
paxos_replica_num,
non_paxos_replica_num,
partition_id,
*a,
now,
false /*is_bootstrap*/,
false /*is_standby*/,
restore,
frozen_status,
arg))) {
LOG_WARN("fail to fill ObCreatePartitionArg", K(ret), K(tablegroup_id));
} else if (OB_FAIL(fill_flag_replica(tablegroup_id,
tablegroup_schema.get_partition_cnt(),
partition_id,
arg,
*a,
role,
flag_replica))) {
LOG_WARN("fail to fill flag replica", K(ret), K(tablegroup_id));
} else if (OB_FAIL(creator.add_flag_replica(flag_replica))) {
@ -15266,7 +15236,6 @@ int ObDDLService::create_partitions_for_physical_restore(ObSchemaGetterGuard& sc
int ret = OB_SUCCESS;
const uint64_t schema_id = restore_arg.schema_id_;
const ObCreateTableMode create_mode = ObCreateTableMode::OB_CREATE_TABLE_MODE_PHYSICAL_RESTORE;
bool is_standby = false;
if (OB_FAIL(check_inner_stat())) {
LOG_WARN("variable is not init", K(ret));
} else if (!restore_arg.is_valid() || addrs.count() <= 0 || last_schema_version <= OB_CORE_SCHEMA_VERSION) {
@ -15275,8 +15244,6 @@ int ObDDLService::create_partitions_for_physical_restore(ObSchemaGetterGuard& sc
} else if (!partition_schema.has_self_partition()) {
ret = OB_NOT_SUPPORTED;
LOG_WARN("table/tablegroup should has self partition", K(ret), K(schema_id));
} else if (OB_FAIL(get_is_standby_cluster(is_standby))) {
LOG_WARN("fail to get standby cluster", KR(ret), K(partition_schema));
} else {
ObPartIdsGenerator gen(partition_schema);
ObArray<int64_t> part_ids;
@ -15328,12 +15295,9 @@ int ObDDLService::create_partitions_for_physical_restore(ObSchemaGetterGuard& sc
FOREACH_CNT_X(a, part_addr, OB_SUCC(ret))
{
ObPartitionReplica flag_replica;
common::ObRole role = FOLLOWER;
if (OB_UNLIKELY(nullptr == a || nullptr == rpc_proxy_)) {
ret = OB_ERR_UNEXPECTED;
LOG_WARN("addr or rpc_proxy is null", K(ret));
} else if (OB_FAIL(set_flag_role(a->initial_leader_, is_standby, restore, partition_id, role))) {
LOG_WARN("fail to set flag role", KR(ret));
} else if (OB_FAIL(fill_create_partition_arg(schema_id,
partition_schema.get_partition_cnt(),
paxos_replica_num,
@ -15347,13 +15311,8 @@ int ObDDLService::create_partitions_for_physical_restore(ObSchemaGetterGuard& sc
frozen_status,
arg))) {
LOG_WARN("fail to fill ObCreatePartitionArg", K(ret), K(schema_id));
} else if (OB_FAIL(fill_flag_replica(schema_id,
partition_schema.get_partition_cnt(),
partition_id,
arg,
*a,
role,
flag_replica))) {
} else if (OB_FAIL(fill_flag_replica(
schema_id, partition_schema.get_partition_cnt(), partition_id, arg, *a, flag_replica))) {
LOG_WARN("fail to fill flag replica", K(ret), K(schema_id));
} else if (OB_FAIL(creator.add_flag_replica(flag_replica))) {
LOG_WARN("fail to add flag replica to partition creator", K(ret), K(flag_replica));

4
src/rootserver/ob_ddl_service.h
View File

@ -802,10 +802,8 @@ private:
const int64_t non_paxos_replica_num, const int64_t partition_id, const ObReplicaAddr& replica_addr,
const int64_t lease_start_ts, const bool is_bootstrap, const bool is_standby, const int64_t restore,
const share::ObSimpleFrozenStatus& frozen_status, obrpc::ObCreatePartitionArg& arg);
int set_flag_role(const bool initial_leader, const bool is_standby, const int64_t restore, const uint64_t table_id,
common::ObRole& role);
int fill_flag_replica(const uint64_t table_id, const int64_t partition_cnt, const int64_t partition_id,
const obrpc::ObCreatePartitionArg& arg, const ObReplicaAddr& replica_addr, const common::ObRole role,
const obrpc::ObCreatePartitionArg& arg, const ObReplicaAddr& replica_addr,
share::ObPartitionReplica& flag_replica);
int try_modify_tenant_primary_zone_entity_count(common::ObMySQLTransaction& trans,

22
src/share/partition_table/ob_persistent_partition_table.cpp
View File

@ -568,22 +568,12 @@ int ObPersistentPartitionTable::update(const ObPartitionReplica& replica)
if (OB_SUCC(ret)) {
if (new_replica.is_leader_like()) {
if (!new_replica.is_flag_replica()) {
if (new_replica.data_version_ <= 0) {
ret = OB_INVALID_ARGUMENT;
LOG_WARN("data_version must > 0 for leader replica", K(ret), K(new_replica));
} else if (new_replica.to_leader_time_ <= 0) {
ret = OB_INVALID_ARGUMENT;
LOG_WARN("change to leader time must > 0 for leader replica", K(ret), K(new_replica));
}
} else {
// flag replica with only the role of leader, to_leader_time_ and data_version_ are not set
// flag is set only by rs,observer shall not set the flag replica
// observer will update the partition table asynchronously by invoking
// ObPGPartitionMTUpdateOperator::batch_update
}
if (OB_FAIL(ret)) {
// nothing todo
if (new_replica.data_version_ <= 0) {
ret = OB_INVALID_ARGUMENT;
LOG_WARN("data_version must > 0 for leader replica", K(ret), K(new_replica));
} else if (new_replica.to_leader_time_ <= 0) {
ret = OB_INVALID_ARGUMENT;
LOG_WARN("change to leader time must > 0 for leader replica", K(ret), K(new_replica));
} else if (OB_FAIL(update_leader_replica(*proxy, partition, new_replica))) {
LOG_WARN("update leader replica failed", K(ret), K(new_replica));
}