[HDFS-14043] Tolerate corrupted seen_txid file

• Type: Bug
• Status: Resolved
• Priority: Major
• Resolution: Fixed
• Affects Version/s: 2.9.2, 3.1.2, 2.9.3
• Fix Version/s: 2.10.0, 3.0.4, 3.1.2, 3.3.0, 3.2.1, 2.9.3
• Component/s: hdfs, namenode
• Hadoop Flags: Reviewed
• Assignee: Lukas Majercak
• Reporter: Lukas Majercak
• Created: 01/Nov/18 18:09
• Updated: 06/Nov/18 01:15
• Resolved: 06/Nov/18 00:58

Description

We already tolerate IOExceptions when reading seen_txid file from namenode's dirs. So we take the maximum txid of all the readable namenode dirs. We should extend this to when the file is corrupted. Currently, PersistentLongFile.readFile throws NumberFormatException in this case and the whole NN crashes.