Hernan Gelaf-Romer created HBASE-29776:
------------------------------------------
Summary: Log filtering in IncrementalBackupManager can lead to
data loss
Key: HBASE-29776
URL: https://issues.apache.org/jira/browse/HBASE-29776
Project: HBase
Issue Type: Bug
Components: backup&restore
Reporter: Hernan Gelaf-Romer
At the moment, incremental backups will filter out old wals that belong to RS
which do not have any active WAL files.
The code
{code:java}
/*
* It is possible that there is no old timestamp in backup system table for
this host. At the
* time of our last backup operation, this rs did not exist. The reason can be
one of the two:
* 1. The rs already left/crashed. Its logs were moved to .oldlogs. 2. The rs
was added after
* our last backup.
*/
if (oldTimeStamp == null) {
if (currentLogTS < Long.parseLong(savedStartCode)) {
// This log file is really old, its region server was before our last backup.
continue;
} else {
resultLogFiles.add(currentLogFile);
}
} else if (currentLogTS > oldTimeStamp) {
resultLogFiles.add(currentLogFile);
} {code}
Is doing the opposite of what the comment (correctly) says.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
