[HDFS-11709] StandbyCheckpointer should handle an non-existing legacyOivImageDir gracefully

• Type: Bug
• Status: Resolved
• Priority: Critical
• Resolution: Fixed
• Affects Version/s: 2.6.1
• Fix Version/s: 2.9.0, 2.7.4, 3.0.0-alpha4, 2.8.2
• Component/s: ha, namenode
• Target Version/s: 2.9.0, 2.7.4, 3.0.0-alpha4, 2.8.1
• Hadoop Flags: Reviewed
• Assignee: Erik Krogen
• Reporter: Zhe Zhang
• Created: 27/Apr/17 05:06
• Updated: 07/Jun/17 22:59
• Resolved: 27/Apr/17 17:11

Description

In StandbyCheckpointer, if the legacy OIV directory is not properly created, or was deleted for some reason (e.g. mis-operation), all checkpoint ops will fall. Not only the ANN won't receive new fsimages, the JNs will get full with edit log files, and cause NN to crash.

      // Save the legacy OIV image, if the output dir is defined.
      String outputDir = checkpointConf.getLegacyOivImageDir();
      if (outputDir != null && !outputDir.isEmpty()) {
        img.saveLegacyOIVImage(namesystem, outputDir, canceler);
      }

It doesn't make sense to let such an unimportant part (saving OIV) abort all checkpoints and cause NN crash (and possibly lose data).

Issue Links

• relates to: HDFS-6293 Issues with OIV processing PB-based fsimages
• relates to: HDFS-11717 Add unit test for HDFS-11709 StandbyCheckpointer should handle non-existing legacyOivImageDir gracefully