This is an automated email from the ASF dual-hosted git repository.
dataroaring pushed a commit to branch master
in repository https://gitbox.apache.org/repos/asf/doris.git
The following commit(s) were added to refs/heads/master by this push:
new c2ca37dcfa0 [fix](restore) Fix atomic restore with exists replicas
(#40734)
c2ca37dcfa0 is described below
commit c2ca37dcfa0741d485415d8a0cbb37d145262ca6
Author: walter <w41te...@gmail.com>
AuthorDate: Fri Sep 13 11:49:48 2024 +0800
[fix](restore) Fix atomic restore with exists replicas (#40734)
1. create replicas with base tablet and schema hash
2. ignore storage medium when creating replicas with the base tablet
The atomic restore is introduced in #40353.
---
be/src/olap/tablet_manager.cpp | 13 +++++++---
.../java/org/apache/doris/backup/RestoreJob.java | 30 ++++++++++++----------
2 files changed, 26 insertions(+), 17 deletions(-)
diff --git a/be/src/olap/tablet_manager.cpp b/be/src/olap/tablet_manager.cpp
index bc883185465..801a288fde8 100644
--- a/be/src/olap/tablet_manager.cpp
+++ b/be/src/olap/tablet_manager.cpp
@@ -279,6 +279,7 @@ Status TabletManager::create_tablet(const TCreateTabletReq&
request, std::vector
// we need use write lock on shard-1 and then use read lock on shard-2
// if there have create rollup tablet C(assume on shard-2) from tablet
D(assume on shard-1) at the same time, we will meet deadlock
std::unique_lock two_tablet_lock(_two_tablet_mtx, std::defer_lock);
+ bool in_restore_mode = request.__isset.in_restore_mode &&
request.in_restore_mode;
bool is_schema_change_or_atomic_restore =
request.__isset.base_tablet_id && request.base_tablet_id > 0;
bool need_two_lock =
@@ -325,14 +326,20 @@ Status TabletManager::create_tablet(const
TCreateTabletReq& request, std::vector
if (base_tablet == nullptr) {
DorisMetrics::instance()->create_tablet_requests_failed->increment(1);
return Status::Error<TABLE_CREATE_META_ERROR>(
- "fail to create tablet(change schema), base tablet does
not exist. "
- "new_tablet_id={}, base_tablet_id={}",
+ "fail to create tablet(change schema/atomic restore), base
tablet does not "
+ "exist. new_tablet_id={}, base_tablet_id={}",
tablet_id, request.base_tablet_id);
}
// If we are doing schema-change or atomic-restore, we should use the
same data dir
// TODO(lingbin): A litter trick here, the directory should be
determined before
// entering this method
- if (request.storage_medium ==
base_tablet->data_dir()->storage_medium()) {
+ //
+ // ATTN: Since all restored replicas will be saved to HDD, so no
storage_medium check here.
+ if (in_restore_mode ||
+ request.storage_medium ==
base_tablet->data_dir()->storage_medium()) {
+ LOG(INFO) << "create tablet use the base tablet data dir.
tablet_id=" << tablet_id
+ << ", base tablet_id=" << request.base_tablet_id
+ << ", data dir=" << base_tablet->data_dir()->path();
stores.clear();
stores.push_back(base_tablet->data_dir());
}
diff --git a/fe/fe-core/src/main/java/org/apache/doris/backup/RestoreJob.java
b/fe/fe-core/src/main/java/org/apache/doris/backup/RestoreJob.java
index b499a9b525e..3f55e1b9088 100644
--- a/fe/fe-core/src/main/java/org/apache/doris/backup/RestoreJob.java
+++ b/fe/fe-core/src/main/java/org/apache/doris/backup/RestoreJob.java
@@ -643,7 +643,7 @@ public class RestoreJob extends AbstractJob implements
GsonPostProcessable {
}
// the new tablets -> { local tablet, schema hash }, used in atomic
restore.
- Map<Long, Pair<Long, Integer>> tabletBases = null;
+ Map<Long, Pair<Long, Integer>> tabletBases = new HashMap<>();
// Check and prepare meta objects.
AgentBatchTask batchTask = new AgentBatchTask();
@@ -802,7 +802,7 @@ public class RestoreJob extends AbstractJob implements
GsonPostProcessable {
if (isAtomicRestore && localTbl != null) {
// bind the backends and base tablets from local tbl.
- tabletBases =
bindLocalAndRemoteOlapTableReplicas((OlapTable) localTbl, remoteOlapTbl);
+ status =
bindLocalAndRemoteOlapTableReplicas((OlapTable) localTbl, remoteOlapTbl,
tabletBases);
if (!status.ok()) {
return;
}
@@ -886,7 +886,7 @@ public class RestoreJob extends AbstractJob implements
GsonPostProcessable {
if (restoreTbl.getType() == TableType.OLAP) {
OlapTable restoreOlapTable = (OlapTable) restoreTbl;
for (Partition restorePart :
restoreOlapTable.getPartitions()) {
- createReplicas(db, batchTask, restoreOlapTable,
restorePart);
+ createReplicas(db, batchTask, restoreOlapTable,
restorePart, tabletBases);
BackupOlapTableInfo backupOlapTableInfo =
jobInfo.getOlapTableInfo(restoreOlapTable.getName());
genFileMapping(restoreOlapTable, restorePart,
backupOlapTableInfo.id,
backupOlapTableInfo.getPartInfo(restorePart.getName()),
@@ -1025,10 +1025,9 @@ public class RestoreJob extends AbstractJob implements
GsonPostProcessable {
// No log here, PENDING state restore job will redo this method
}
- private Map<Long, Pair<Long, Integer>> bindLocalAndRemoteOlapTableReplicas(
- OlapTable localOlapTbl, OlapTable remoteOlapTbl) {
- Map<Long, Pair<Long, Integer>> tabletBases = new HashMap<>();
-
+ private Status bindLocalAndRemoteOlapTableReplicas(
+ OlapTable localOlapTbl, OlapTable remoteOlapTbl,
+ Map<Long, Pair<Long, Integer>> tabletBases) {
localOlapTbl.readLock();
try {
for (Partition partition : remoteOlapTbl.getPartitions()) {
@@ -1045,22 +1044,20 @@ public class RestoreJob extends AbstractJob implements
GsonPostProcessable {
}
int schemaHash =
localOlapTbl.getSchemaHashByIndexId(localIndexId);
if (schemaHash == -1) {
- status = new Status(ErrCode.COMMON_ERROR,
String.format(
+ return new Status(ErrCode.COMMON_ERROR, String.format(
"schema hash of local index %d is not found,
remote table=%d, remote index=%d, "
+ "local table=%d, local index=%d",
localIndexId, remoteOlapTbl.getId(), index.getId(),
localOlapTbl.getId(), localIndexId));
- return null;
}
List<Tablet> localTablets = localIndex.getTablets();
List<Tablet> remoteTablets = index.getTablets();
if (localTablets.size() != remoteTablets.size()) {
- status = new Status(ErrCode.COMMON_ERROR,
String.format(
+ return new Status(ErrCode.COMMON_ERROR, String.format(
"the size of local tablet %s is not equals to
the remote %s, "
+ "is_atomic_restore=true, remote table=%d,
remote index=%d, "
+ "local table=%d, local index=%d",
localTablets.size(), remoteTablets.size(),
remoteOlapTbl.getId(), index.getId(),
localOlapTbl.getId(), localIndexId));
- return null;
}
for (int i = 0; i < remoteTablets.size(); i++) {
Tablet localTablet = localTablets.get(i);
@@ -1068,14 +1065,13 @@ public class RestoreJob extends AbstractJob implements
GsonPostProcessable {
List<Replica> localReplicas =
localTablet.getReplicas();
List<Replica> remoteReplicas =
remoteTablet.getReplicas();
if (localReplicas.size() != remoteReplicas.size()) {
- status = new Status(ErrCode.COMMON_ERROR,
String.format(
+ return new Status(ErrCode.COMMON_ERROR,
String.format(
"the size of local replicas %s is not
equals to the remote %s, "
+ "is_atomic_restore=true, remote
table=%d, remote index=%d, "
+ "local table=%d, local index=%d, local
replicas=%d, remote replicas=%d",
localTablets.size(), remoteTablets.size(),
remoteOlapTbl.getId(),
index.getId(), localOlapTbl.getId(),
localIndexId, localReplicas.size(),
remoteReplicas.size()));
- return null;
}
for (int j = 0; j < remoteReplicas.size(); j++) {
long backendId =
localReplicas.get(j).getBackendId();
@@ -1093,7 +1089,7 @@ public class RestoreJob extends AbstractJob implements
GsonPostProcessable {
} finally {
localOlapTbl.readUnlock();
}
- return tabletBases;
+ return Status.OK;
}
private void prepareAndSendSnapshotTaskForOlapTable(Database db) {
@@ -1274,6 +1270,8 @@ public class RestoreJob extends AbstractJob implements
GsonPostProcessable {
// ensure this replica is bound to the same backend
disk as the origin table's replica.
Pair<Long, Integer> baseTablet =
tabletBases.get(restoreTablet.getId());
task.setBaseTablet(baseTablet.first,
baseTablet.second);
+ LOG.info("set base tablet {} for replica {} in restore
job {}, tablet id={}",
+ baseTablet.first, restoreReplica.getId(),
jobId, restoreTablet.getId());
}
batchTask.addTask(task);
}
@@ -1379,6 +1377,10 @@ public class RestoreJob extends AbstractJob implements
GsonPostProcessable {
long refTabletId = -1L;
if (tabletBases != null &&
tabletBases.containsKey(localTablet.getId())) {
refTabletId =
tabletBases.get(localTablet.getId()).first;
+ if (LOG.isDebugEnabled()) {
+ LOG.debug("restored tablet {} is based on exists
tablet {}",
+ localTablet.getId(), refTabletId);
+ }
}
long noReplicaId = -1L;
---------------------------------------------------------------------
To unsubscribe, e-mail: commits-unsubscr...@doris.apache.org
For additional commands, e-mail: commits-h...@doris.apache.org
