diff --git a/docs/en/docs/admin-manual/config/fe-config.md b/docs/en/docs/admin-manual/config/fe-config.md index c92eeab136ff53..bc3acae538cedc 100644 --- a/docs/en/docs/admin-manual/config/fe-config.md +++ b/docs/en/docs/admin-manual/config/fe-config.md @@ -1785,6 +1785,20 @@ In some very special circumstances, such as code bugs, or human misoperation, et Set to true so that Doris will automatically use blank replicas to fill tablets which all replicas have been damaged or missing +#### `recover_with_skip_missing_version` + +Default：disable + +IsMutable：true + +MasterOnly：true + +In some scenarios, there is an unrecoverable metadata problem in the cluster, and the visibleVersion of the data does not match be. In this case, it is still necessary to restore the remaining data (which may cause problems with the correctness of the data). This configuration is the same as` recover_with_empty_tablet` should only be used in emergency situations +This configuration has three values: +* disable : If an exception occurs, an error will be reported normally. +* ignore_version: ignore the visibleVersion information recorded in fe partition, use replica version +* ignore_all: In addition to ignore_version, when encountering no queryable replica, skip it directly instead of throwing an exception + #### `min_clone_task_timeout_sec` `And max_clone_task_timeout_sec` Default：Minimum 3 minutes, maximum two hours diff --git a/docs/zh-CN/docs/admin-manual/config/fe-config.md b/docs/zh-CN/docs/admin-manual/config/fe-config.md index c506c3fa229d4f..99b97544bd53a4 100644 --- a/docs/zh-CN/docs/admin-manual/config/fe-config.md +++ b/docs/zh-CN/docs/admin-manual/config/fe-config.md @@ -1785,6 +1785,24 @@ show data （其他用法：HELP SHOW DATA） 在这种情况下，您可以将此配置设置为 true。 系统会将损坏的 tablet 替换为空 tablet，以确保查询可以执行。 （但此时数据已经丢失，所以查询结果可能不准确） +#### `recover_with_skip_missing_version` + +默认值：disable + +是否可以动态配置：true + +是否为 Master FE 节点独有的配置项：true + +有些场景下集群出现了不可恢复的元数据问题，数据已的visibleversion 已经和be 不匹配， + +这种情况下仍然需要恢复剩余的数据（可能能会导致数据的正确性有问题），这个配置同`recover_with_empty_tablet` 一样只能在紧急情况下使用 + +这个配置有三个值： + + * disable ：出现异常会正常报错。 + * ignore_version: 忽略 fe partition 中记录的visibleVersion 信息， 使用replica version + * ignore_all: 除了ignore_version， 在遇到找不到可查询的replica 时，直接跳过而不是抛出异常 + #### `min_clone_task_timeout_sec` 和 `max_clone_task_timeout_sec` 默认值：最小3分钟，最大两小时 diff --git a/fe/fe-common/src/main/java/org/apache/doris/common/Config.java b/fe/fe-common/src/main/java/org/apache/doris/common/Config.java index 2b40215064a9a2..0322385c9bb0d4 100644 --- a/fe/fe-common/src/main/java/org/apache/doris/common/Config.java +++ b/fe/fe-common/src/main/java/org/apache/doris/common/Config.java @@ -1439,6 +1439,20 @@ public class Config extends ConfigBase { @ConfField(mutable = true, masterOnly = true) public static boolean recover_with_empty_tablet = false; + /** + * In some scenarios, there is an unrecoverable metadata problem in the cluster, + * and the visibleVersion of the data does not match be. In this case, it is still + * necessary to restore the remaining data (which may cause problems with the correctness of the data). + * This configuration is the same as` recover_with_empty_tablet` should only be used in emergency situations + * This configuration has three values: + * disable : If an exception occurs, an error will be reported normally. + * ignore_version: ignore the visibleVersion information recorded in fe partition, use replica version + * ignore_all: In addition to ignore_version, when encountering no queryable replica, + * skip it directly instead of throwing an exception + */ + @ConfField(mutable = true, masterOnly = true) + public static String recover_with_skip_missing_version = "disable"; + /** * Whether to add a delete sign column when create unique table */ diff --git a/fe/fe-core/src/main/java/org/apache/doris/planner/OlapScanNode.java b/fe/fe-core/src/main/java/org/apache/doris/planner/OlapScanNode.java index cee6ff90d05f37..4f776e55e47d27 100644 --- a/fe/fe-core/src/main/java/org/apache/doris/planner/OlapScanNode.java +++ b/fe/fe-core/src/main/java/org/apache/doris/planner/OlapScanNode.java @@ -53,6 +53,7 @@ import org.apache.doris.catalog.Replica; import org.apache.doris.catalog.Tablet; import org.apache.doris.common.AnalysisException; +import org.apache.doris.common.Config; import org.apache.doris.common.ErrorCode; import org.apache.doris.common.ErrorReport; import org.apache.doris.common.UserException; @@ -708,6 +709,20 @@ private void addScanRangeLocations(Partition partition, } for (Tablet tablet : tablets) { long tabletId = tablet.getId(); + if (!Config.recover_with_skip_missing_version.equalsIgnoreCase("disable")) { + long tabletVersion = -1L; + for (Replica replica : tablet.getReplicas()) { + if (replica.getVersion() > tabletVersion) { + tabletVersion = replica.getVersion(); + } + } + if (tabletVersion != visibleVersion) { + LOG.warn("tablet {} version {} is not equal to partition {} version {}", + tabletId, tabletVersion, partition.getId(), visibleVersion); + visibleVersion = tabletVersion; + visibleVersionStr = String.valueOf(visibleVersion); + } + } TScanRangeLocations scanRangeLocations = new TScanRangeLocations(); TPaloScanRange paloRange = new TPaloScanRange(); paloRange.setDbName(""); @@ -783,7 +798,12 @@ private void addScanRangeLocations(Partition partition, scanBackendIds.add(backend.getId()); } if (tabletIsNull) { - throw new UserException(tabletId + " have no queryable replicas. err: " + Joiner.on(", ").join(errs)); + if (Config.recover_with_skip_missing_version.equalsIgnoreCase("ignore_all")) { + continue; + } else { + throw new UserException(tabletId + " have no queryable replicas. err: " + + Joiner.on(", ").join(errs)); + } } TScanRange scanRange = new TScanRange(); scanRange.setPaloScanRange(paloRange);