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

[BUG] fix unexpected error when server shutdown and tenant transfer out

Browse Source
This commit is contained in:
Handora
2023-01-11 04:42:04 +00:00
committed by ob-robot
parent b95869c8ed
commit 7e632f0b65
1 changed files with 25 additions and 11 deletions
Download Patch File Download Diff File Expand all files Collapse all files

32
src/storage/concurrency_control/ob_multi_version_garbage_collector.cpp
View File

@ -560,10 +560,17 @@ void ObMultiVersionGarbageCollector::decide_reserved_snapshot_version_(
} else if (reserved_snapshot < global_reserved_snapshot_) { } else if (reserved_snapshot < global_reserved_snapshot_) {
if ((global_reserved_snapshot_.get_val_for_tx() - if ((global_reserved_snapshot_.get_val_for_tx() -
reserved_snapshot.get_val_for_tx()) / 1000 > 30 * 1_min) { reserved_snapshot.get_val_for_tx()) / 1000 > 30 * 1_min) {
// We ignore the reserved snapshot with too late snapshot and report // We ignore the reserved snapshot with too late snapshot and report WARN
// error to maintain normal processes // because there may be servers offline and online suddenly and report a
MVCC_LOG(ERROR, "update a too smaller reserved snapshot", K(ret), KPC(this), // stale txn version. And we report error for a too too old snapshot.
if ((global_reserved_snapshot_.get_val_for_tx() -
reserved_snapshot.get_val_for_tx()) / 1000 > 100 * 1_min) {
MVCC_LOG(ERROR, "update a too too smaller reserved snapshot!!!", K(ret), KPC(this),
K(global_reserved_snapshot_), K(reserved_snapshot)); K(global_reserved_snapshot_), K(reserved_snapshot));
} else {
MVCC_LOG(WARN, "update a too too smaller reserved snapshot!", K(ret), KPC(this),
K(global_reserved_snapshot_), K(reserved_snapshot));
}
} else { } else {
MVCC_LOG(WARN, "update a smaller reserved snapshot", K(ret), KPC(this), MVCC_LOG(WARN, "update a smaller reserved snapshot", K(ret), KPC(this),
K(global_reserved_snapshot_), K(reserved_snapshot)); K(global_reserved_snapshot_), K(reserved_snapshot));
@ -879,9 +886,10 @@ int ObMultiVersionGarbageCollector::reclaim()
if (OB_TMP_FAIL(share::ObAllServerTracer::get_instance().check_server_alive(addr, is_alive))) { if (OB_TMP_FAIL(share::ObAllServerTracer::get_instance().check_server_alive(addr, is_alive))) {
MVCC_LOG(WARN, "check all server tracer failed", K(tmp_ret)); MVCC_LOG(WARN, "check all server tracer failed", K(tmp_ret));
} else if (is_alive) { } else if (is_alive) {
// Case 1.1: server is alive, we report the error for not renewing. // Case 1.1: server is alive, we report the WARN for not
// monitor should asserts the exception // renewing. because there may be tenant transfer out
MVCC_LOG(ERROR, "server alives while not renew for a long time", K(create_time), // which cause it will not be reclaimed forever
MVCC_LOG(WARN, "server alives while not renew for a long time", K(create_time),
K(current_timestamp), K(addr), K(snapshot_type), K(snapshot_version)); K(current_timestamp), K(addr), K(snapshot_type), K(snapshot_version));
need_reclaim = true; need_reclaim = true;
} else { } else {
@ -1160,10 +1168,16 @@ int ObMultiVersionGCSnapshotCalculator::operator()(const share::SCN snapshot_ver
current_ts - create_time > 2 * ObMultiVersionGarbageCollector::GARBAGE_COLLECT_RECLAIM_DURATION && current_ts - create_time > 2 * ObMultiVersionGarbageCollector::GARBAGE_COLLECT_RECLAIM_DURATION &&
// for mock or test that change GARBAGE_COLLECT_EXEC_INTERVAL to a small value // for mock or test that change GARBAGE_COLLECT_EXEC_INTERVAL to a small value
current_ts - create_time > 2 * 3 * 10_min) { current_ts - create_time > 2 * 3 * 10_min) {
// we report error here because we may relay on the reclaim strategy of if (REACH_COUNT_INTERVAL(10L)) {
// garbage collector to reclaim the data of died machine // we report error here for long time warning
MVCC_LOG(ERROR, "ignore too old version", K(snapshot_version), MVCC_LOG(ERROR, "ignore too old version too long", K(snapshot_version),
K(snapshot_type), K(current_ts), K(create_time), K(addr)); K(snapshot_type), K(current_ts), K(create_time), K(addr));
} else {
// we report WARN here because there may be servers offline and online
// suddenly and report a stale txn
MVCC_LOG(WARN, "ignore too old version", K(snapshot_version),
K(snapshot_type), K(current_ts), K(create_time), K(addr));
}
} else { } else {
reserved_snapshot_version_ = snapshot_version; reserved_snapshot_version_ = snapshot_version;
reserved_snapshot_type_ = snapshot_type; reserved_snapshot_type_ = snapshot_type;