wait log replayed when server restarting

This commit is contained in:
obdev
2023-06-26 08:42:27 +00:00
committed by ob-robot
parent 5d3a888639
commit d7371fc109
8 changed files with 210 additions and 111 deletions

View File

@ -894,25 +894,6 @@ int ObServer::start()
}
FLOG_INFO("check if multi tenant synced", KR(ret), K(stop_), K(synced));
/*
* FIXME: skip partition service op first
if (OB_SUCC(ret)) {
do {
if (stop_) {
ret = OB_SERVER_IS_STOPPING;
// } else if (OB_FAIL(ObPartitionService::get_instance().wait_start_finish())) {
if (OB_EAGAIN == ret) {
ob_usleep(100 * 1000);
} else {
LOG_ERROR("wait scan inner table failed", KR(ret));
}
} else {
LOG_INFO("[NOTICE] wait scan inner table success");
}
} while (OB_EAGAIN == ret);
}
*/
bool schema_ready = false;
while (OB_SUCC(ret) && !stop_ && !schema_ready) {
schema_ready = schema_service_.is_sys_full_schema();
@ -941,14 +922,34 @@ int ObServer::start()
}
LOG_INFO("[NOTICE] check if sys srs usable", K(ret), K(stop_));
// check log replay and user tenant schema refresh status
if (OB_SUCC(ret)) {
if (stop_) {
ret = OB_SERVER_IS_STOPPING;
FLOG_WARN("server is in stopping status", KR(ret));
} else if (OB_FAIL(check_server_can_start_service())) {
LOG_ERROR("fail to check server can start service", KR(ret));
} else {
FLOG_INFO("success to check server can start service", KR(ret));
ObSEArray<uint64_t, 16> tenant_ids;
const int64_t MAX_CHECK_TIME = 15 * 60 * 1000 * 1000L; // 15min
const int64_t start_ts = ObTimeUtility::current_time();
int64_t schema_refreshed_ts = 0;
const int64_t expire_time = start_ts + MAX_CHECK_TIME;
if (OB_FAIL(multi_tenant_.get_mtl_tenant_ids(tenant_ids))) {
FLOG_ERROR("get mtl tenant ids fail", KR(ret));
} else if (tenant_ids.count() <= 0) {
// do nothing
} else {
// check user tenant schema refresh
check_user_tenant_schema_refreshed(tenant_ids, expire_time);
schema_refreshed_ts = ObTimeUtility::current_time();
// check log replay status
check_log_replay_over(tenant_ids, expire_time);
}
FLOG_INFO("[OBSERVER_NOTICE] check log replay and user tenant schema finished",
KR(ret),
K(tenant_ids),
"refresh_schema_cost_us", schema_refreshed_ts - start_ts,
"replay_log_cost_us", ObTimeUtility::current_time() - schema_refreshed_ts);
}
}
}
@ -2735,76 +2736,62 @@ int ObServer::reload_bandwidth_throttle_limit(int64_t network_speed)
return ret;
}
int ObServer::check_server_can_start_service()
void ObServer::check_user_tenant_schema_refreshed(const ObIArray<uint64_t> &tenant_ids, const int64_t expire_time)
{
int ret = OB_SUCCESS;
int64_t min_wrs = INT64_MAX;
// TODO: implement this function
return ret;
/*
for (int64_t i = 0; i < tenant_ids.count()
&& ObTimeUtility::current_time() < expire_time; ++i) {
uint64_t tenant_id = tenant_ids.at(i);
bool tenant_schema_refreshed = false;
while (!tenant_schema_refreshed
&& !stop_
&& ObTimeUtility::current_time() < expire_time) {
//On the standby database, it is very likely that the minimum standby machine-readable timestamp cannot be pushed because the main database does not exist.
//Do not stop the server from starting, otherwise you may not be able to create a connection
int64_t get_min_wrs_ts = ObTimeUtility::current_time();
do {
bool can_start_service = true;
if (stop_) {
ret = OB_SERVER_IS_STOPPING;
tenant_schema_refreshed = is_user_tenant(tenant_id) ?
gctx_.schema_service_->is_tenant_refreshed(tenant_id) : true;
if (!tenant_schema_refreshed) {
// check wait and retry
usleep(1000 * 1000);
if (REACH_TIME_INTERVAL(10 * 1000 * 1000)) {
FLOG_INFO("[OBSERVER_NOTICE] Refreshing user tenant schema, need to wait ", K(tenant_id));
}
// check success
} else if (i == tenant_ids.count() - 1) {
FLOG_INFO("[OBSERVER_NOTICE] Refresh all user tenant schema successfully ", K(tenant_ids));
// check timeout
} else if (ObTimeUtility::current_time() > expire_time) {
FLOG_INFO("[OBSERVER_NOTICE] Refresh user tenant schema timeout ", K(tenant_id));
} else {
//Check whether the lagging amount of all partitions of the machine is greater than max_stale_time_for_weak_consistency. If true, it cannot be restarted
int64_t tmp_min_wrs = INT64_MAX;
weak_read_service_.check_server_can_start_service(can_start_service, tmp_min_wrs);
FLOG_INFO("[OBSERVER_NOTICE] Refresh user tenant schema successfully ", K(tenant_id));
}
}
}
}
void ObServer::check_log_replay_over(const ObIArray<uint64_t> &tenant_ids, const int64_t expire_time)
{
for (int64_t i = 0; i < tenant_ids.count()
&& ObTimeUtility::current_time() < expire_time; ++i) {
SCN min_version;
uint64_t tenant_id = tenant_ids.at(i);
bool can_start_service = false;
while (!can_start_service
&& !stop_
&& ObTimeUtility::current_time() < expire_time) {
weak_read_service_.check_tenant_can_start_service(tenant_id, can_start_service, min_version);
// check wait and retry
if (!can_start_service) {
const int64_t STANDBY_WAIT_WRS_DURATION = 60 * 1000 * 1000;
const int64_t current_time = ObTimeUtility::current_time();
if (min_wrs != tmp_min_wrs) {
get_min_wrs_ts = current_time;
min_wrs = tmp_min_wrs;
}
if (GCTX.is_standby_cluster() && STANDBY_WAIT_WRS_DURATION < current_time - get_min_wrs_ts
// && ObPartitionService::get_instance().is_scan_disk_finished()) {
//If it is in the standby database, and the minimum standby machine readable timestamp has been one minute, there is no way to continue advancing, and the clog playback has ended,
//You need to let go of this processing
ret = OB_SUCCESS;
LOG_INFO("[NOTICE] in standby cluster, no need to wait weak read timestamp", K(min_wrs),
"hanging time", current_time - get_min_wrs_ts);
usleep(1000 * 1000);
// check success
} else if (i == tenant_ids.count() -1) {
FLOG_INFO("[OBSERVER_NOTICE] all tenant replay log finished, start to service ", K(tenant_ids));
// check timeout
} else if (ObTimeUtility::current_time() > expire_time) {
FLOG_INFO("[OBSERVER_NOTICE] replay log timeout and force to start service ", K(tenant_id));
} else {
ret = OB_EAGAIN;
ob_usleep(1000 * 1000);
if (REACH_TIME_INTERVAL(10 * 1000 * 1000)) {
LOG_INFO("[NOTICE] clog is behind, service starting need to wait !");
}
}
} else {
ret = OB_SUCCESS;
}
}
} while (OB_EAGAIN == ret);
//Wait for OBS to load information related to cluster_info
if (OB_SUCC(ret)) {
ObClusterRole cluster_role = INVALID_CLUSTER_ROLE;
share::ServerServiceStatus server_status = OBSERVER_INVALID_STATUS;
while (OB_SUCC(ret)) {
gctx_.get_cluster_role_and_status(cluster_role, server_status);
if (OBSERVER_INVALID_STATUS == server_status
|| INVALID_CLUSTER_ROLE == cluster_role) {
ob_usleep(1000 * 1000);
if (REACH_TIME_INTERVAL(10 * 1000 * 1000)) {
LOG_INFO("[NOTICE] not load cluster info, service starting need to wait !");
}
if (stop_) {
ret = OB_SERVER_IS_STOPPING;
break;
}
} else {
break;
// do nothing
}
}
}
return ret;
*/
}
ObServer::ObCTASCleanUpTask::ObCTASCleanUpTask()

View File

@ -17,6 +17,8 @@
#include "lib/signal/ob_signal_worker.h"
#include "lib/net/ob_net_util.h"
#include "lib/random/ob_mysql_random.h"
#include "lib/container/ob_iarray.h"
#include "share/stat/ob_opt_stat_service.h"
#include "share/ratelimit/ob_rl_mgr.h"
@ -290,7 +292,8 @@ private:
int init_refresh_network_speed_task();
int init_refresh_cpu_frequency();
int set_running_mode();
int check_server_can_start_service();
void check_user_tenant_schema_refreshed(const common::ObIArray<uint64_t> &tenant_ids, const int64_t expire_time);
void check_log_replay_over(const common::ObIArray<uint64_t> &tenant_ids, const int64_t expire_time);
int try_create_hidden_sys();
int parse_mode();

View File

@ -77,9 +77,10 @@ public:
/// @param [out] version wrs version
virtual int get_cluster_version(const uint64_t tenant_id, share::SCN &version) = 0;
/// check server can start service or not
virtual void check_server_can_start_service(bool &can_start_service,
int64_t &min_wrs) const = 0;
/// check tenant can start service or not
virtual int check_tenant_can_start_service(const uint64_t tenant_id,
bool &can_start_service,
share::SCN &version) const = 0;
/// get RPC instance
virtual ObIWrsRpc &get_wrs_rpc() = 0;

View File

@ -44,6 +44,7 @@ public:
share::SCN &wrs_version,
const int64_t max_stale_time);
share::SCN get_ls_weak_read_ts() const { return ls_weak_read_ts_; }
bool can_skip_ls() const { return !is_enabled_; }
private:
int generate_weak_read_timestamp_(oceanbase::storage::ObLS &ls, const int64_t max_stale_time, share::SCN &timestamp);

View File

@ -711,6 +711,81 @@ void ObTenantWeakReadService::run1()
ISTAT("thread end", K_(tenant_id));
}
int ObTenantWeakReadService::check_can_start_service(const SCN &current_gts,
bool &can_start_service,
SCN &min_version,
share::ObLSID &ls_id)
{
int ret = OB_SUCCESS;
storage::ObLSService *ls_svr = MTL(storage::ObLSService*);
common::ObSharedGuard<storage::ObLSIterator> iter;
ObLSID tmp_ls_id;
SCN tmp_min_version;
int64_t total_ls_cnt = 0;
can_start_service = true;
const int64_t MAX_STALE_TIME = 30 * 1000 * 1000;
if (OB_ISNULL(ls_svr)) {
ret = OB_ERR_UNEXPECTED;
FLOG_ERROR("unexpected ls service", K(ret), KP(ls_svr));
} else if (OB_FAIL(ls_svr->get_ls_iter(iter, ObLSGetMod::TRANS_MOD))) {
if (OB_NOT_RUNNING != ret) {
FLOG_WARN("fail to alloc ls iter", KR(ret));
}
} else {
int64_t start_time = ObTimeUtility::current_time();
while (OB_SUCCESS == ret) {
ObLS *ls = NULL;
if (OB_FAIL(iter->get_next(ls))) {
if (OB_ITER_END == ret) {
// do nothing
} else {
FLOG_WARN("iterate next ls fail", KR(ret));
}
} else if (OB_ISNULL(ls)) {
ret = OB_PARTITION_NOT_EXIST;
FLOG_WARN("iterate ls fail", KP(ls));
} else if (ls->get_ls_wrs_handler()->can_skip_ls()) {
// do nothing
} else {
++total_ls_cnt;
if (tmp_min_version.is_valid()) {
if (ls->get_ls_wrs_handler()->get_ls_weak_read_ts() < tmp_min_version) {
tmp_min_version = ls->get_ls_wrs_handler()->get_ls_weak_read_ts();
tmp_ls_id = ls->get_ls_id();
}
} else {
tmp_min_version = ls->get_ls_wrs_handler()->get_ls_weak_read_ts();
tmp_ls_id = ls->get_ls_id();
}
}
} // while
if (OB_ITER_END == ret) {
ret = OB_SUCCESS;
}
if (total_ls_cnt == 0) {
can_start_service = true;
FLOG_INFO("empty ls, no need to wait replaying log", K(current_gts), K(tmp_min_version), K(ls_id));
} else if (OB_SUCC(ret)) {
min_version = tmp_min_version;
ls_id = tmp_ls_id;
can_start_service = (tmp_min_version.is_valid() &&
current_gts.convert_to_ts() - MAX_STALE_TIME < tmp_min_version.convert_to_ts());
if (!can_start_service) {
FLOG_WARN("current ls can not start service, waiting for replaying log",
"target_ts", current_gts.convert_to_ts(),
"min_ts", tmp_min_version.convert_to_ts(),
"delta_us", (current_gts.convert_to_ts() - tmp_min_version.convert_to_ts()),
K(ls_id));
}
} else {
// do nothing
}
}
return ret;
}
void ObTenantWeakReadService::set_cluster_service_master_(const ObAddr &addr)
{

View File

@ -108,6 +108,10 @@ public:
// get weak read info stat
void get_weak_read_stat(ObTenantWeakReadStat &wrs_stat) const;
int check_can_start_service(const SCN &current_gts,
bool &can_start_service,
SCN &min_version,
share::ObLSID &ls_id);
public:
// tenant level variables init and destroy function
static int mtl_init(ObTenantWeakReadService* &twrs);

View File

@ -23,6 +23,7 @@
#include "ob_weak_read_util.h" //ObWeakReadUtil
#include "storage/tx_storage/ob_ls_map.h"
#include "storage/tx_storage/ob_ls_service.h"
#include "storage/tx/ob_ts_mgr.h"
#include "logservice/ob_log_service.h"
namespace oceanbase
@ -134,29 +135,56 @@ int ObWeakReadService::get_cluster_version(const uint64_t tenant_id, SCN &versio
return ret;
}
// TODO shanyan.g
void ObWeakReadService::check_server_can_start_service(bool &can_start_service,
int64_t &min_wrs) const
int ObWeakReadService::check_tenant_can_start_service(const uint64_t tenant_id,
bool &can_start_service,
SCN &version) const
{
UNUSEDx(can_start_service, min_wrs);
// int ret = OB_SUCCESS;
// int64_t safe_weak_read_snapshot = INT64_MAX;
// ObPartitionService &ps = ObPartitionService::get_instance();
// if (OB_FAIL(ps.check_can_start_service(can_start_service, safe_weak_read_snapshot, pkey))) {
// LOG_WARN("partition service check can start service error", KR(ret));
// } else if (!can_start_service) {
// if (REACH_TIME_INTERVAL(5 * 1000 * 1000)) {
// LOG_INFO("[WRS] [NOTICE] server can not start service", K(can_start_service),
// K(safe_weak_read_snapshot),
// "delta", ObTimeUtility::current_time() - safe_weak_read_snapshot, K(pkey));
// }
// } else {
// LOG_INFO("[WRS] [NOTICE] server can start service", K(can_start_service), K(safe_weak_read_snapshot),
// "delta", ObTimeUtility::current_time() - safe_weak_read_snapshot, K(pkey));
// }
// if (OB_SUCC(ret)) {
// min_wrs = safe_weak_read_snapshot;
// }
int ret = OB_SUCCESS;
SCN gts_scn;
ObLSID ls_id;
SCN min_version;
MTL_SWITCH(tenant_id) {
ObTenantWeakReadService *twrs = MTL(ObTenantWeakReadService *);
if (OB_ISNULL(twrs)) {
ret = OB_ERR_UNEXPECTED;
FLOG_ERROR("MTL ObTenantWeakReadService object is NULL", K(twrs), K(tenant_id), KR(ret));
} else if (OB_FAIL(OB_TS_MGR.get_gts(tenant_id, NULL, gts_scn))) {
FLOG_WARN("[WRS] [OBSERVER_NOTICE] get gts scn error", K(ret), K(tenant_id));
} else if (OB_FAIL(twrs->check_can_start_service(gts_scn,
can_start_service,
min_version,
ls_id))) {
FLOG_WARN("get tenant weak read service cluster version fail", KR(ret), K(tenant_id));
} else if (!can_start_service) {
version = min_version;
} else {
// success
}
} else {
FLOG_WARN("change tenant context fail when get weak read service cluster version",
KR(ret), K(tenant_id));
}
if (can_start_service) {
FLOG_INFO("[WRS] [OBSERVER_NOTICE] current tenant start service successfully",
K(tenant_id),
"target_ts", gts_scn.convert_to_ts(),
"min_ts", (min_version.is_valid() ? min_version.convert_to_ts() : 0),
"delta_us", (min_version.is_valid() ? (gts_scn.convert_to_ts() - min_version.convert_to_ts()) : 0));
} else {
if (REACH_TIME_INTERVAL(5 * 1000 * 1000)) {
int64_t tmp_version = min_version.is_valid() ? min_version.convert_to_ts() : 0;
FLOG_INFO("[WRS] [OBSERVER_NOTICE] waiting log replay... ",
K(ret),
K(tenant_id),
K(can_start_service),
"min_version", tmp_version,
"delta", ObTimeUtility::current_time() - tmp_version,
"slowest_ls_id", ls_id);
}
}
return ret;
}
void ObWeakReadService::process_get_cluster_version_rpc(const uint64_t tenant_id,

View File

@ -50,7 +50,7 @@ public:
/// get CLUSTER level weak read version
int get_cluster_version(const uint64_t tenant_id, share::SCN &version);
void check_server_can_start_service(bool &can_start_service, int64_t &min_wrs) const;
int check_tenant_can_start_service(const uint64_t tenant_id, bool &can_start_service, SCN &version) const;
///////////// RPC process functions /////////////////
void process_get_cluster_version_rpc(const uint64_t tenant_id,