database/doris
2
0
Fork 0
Code Issues Pull Requests Actions 3 Packages Projects Releases Wiki Activity

Fix old tablet inserting bug (#5113)

Browse Source 
#4996
When BE is restarting and the older tablet have been added to the garbage collection queue but not deleted yet.
In this case, since the data_dirs are parallel loaded, a later loaded tablet may be older than previously loaded one, which should not be acknowledged as a failure.

It should be noted that the _add_tablet_unlocked() method will also be called when creating a new tablet. In that case, the changes in this pull request will not be accessed so there is no affect on the tablet creating process.
This commit is contained in:
Skysheepwang
2020-12-24 15:20:54 +08:00
committed by GitHub
parent 80209ef1b6
commit 86e40dd3e5
3 changed files with 11 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files

10
be/src/olap/data_dir.cpp
View File

@ -687,7 +687,8 @@ OLAPStatus DataDir::load() {
const std::string& value) -> bool {
OLAPStatus status = _tablet_manager->load_tablet_from_meta(this, tablet_id, schema_hash,
value, false, false);
if (status != OLAP_SUCCESS && status != OLAP_ERR_TABLE_ALREADY_DELETED_ERROR) {
if (status != OLAP_SUCCESS && status != OLAP_ERR_TABLE_ALREADY_DELETED_ERROR
&& status != OLAP_ERR_ENGINE_INSERT_OLD_TABLET) {
// load_tablet_from_meta() may return OLAP_ERR_TABLE_ALREADY_DELETED_ERROR
// which means the tablet status is DELETED
// This may happen when the tablet was just deleted before the BE restarted,
@ -695,6 +696,13 @@ OLAPStatus DataDir::load() {
// will read the tablet in the DELETE state from rocksdb. These tablets have been
// added to the garbage collection queue and will be automatically deleted afterwards.
// Therefore, we believe that this situation is not a failure.
// Besides, load_tablet_from_meta() may return OLAP_ERR_ENGINE_INSERT_OLD_TABLET
// when BE is restarting and the older tablet have been added to the
// garbage collection queue but not deleted yet.
// In this case, since the data_dirs are parallel loaded, a later loaded tablet
// may be older than previously loaded one, which should not be acknowledged as a
// failure.
LOG(WARNING) << "load tablet from header failed. status:" << status
<< ", tablet=" << tablet_id << "." << schema_hash;
failed_tablet_ids.insert(tablet_id);

1
be/src/olap/olap_define.h
View File

@ -194,6 +194,7 @@ enum OLAPStatus {
OLAP_ERR_TABLE_INSERT_DUPLICATION_ERROR = -503,
OLAP_ERR_DELETE_VERSION_ERROR = -504,
OLAP_ERR_GC_SCAN_PATH_ERROR = -505,
OLAP_ERR_ENGINE_INSERT_OLD_TABLET = -506,
// FetchHandler
// [-600, -700)

2
be/src/olap/tablet_manager.cpp
View File

@ -147,7 +147,7 @@ OLAPStatus TabletManager::_add_tablet_unlocked(TTabletId tablet_id, SchemaHash s
res = _add_tablet_to_map_unlocked(tablet_id, schema_hash, tablet, update_meta, keep_files,
true /*drop_old*/);
} else {
res = OLAP_ERR_ENGINE_INSERT_EXISTS_TABLE;
res = OLAP_ERR_ENGINE_INSERT_OLD_TABLET;
}
LOG(WARNING) << "add duplicated tablet. force=" << force << ", res=" << res
<< ", tablet_id=" << tablet_id << ", schema_hash=" << schema_hash