[HADOOP-6986] SequenceFile.Reader should distinguish between Network IOE and Parsing IOE

• Type: Bug
• Status: Open
• Priority: Minor
• Resolution: Unresolved
• Affects Version/s: 0.20-append, 0.21.1, 0.22.0
• Component/s: io
• Assignee: Unassigned
• Reporter: Nicolas Spiegelberg
• Created: 04/Oct/10 21:36
• Updated: 11/Mar/15 23:34

Description

The SequenceFile.Reader api should give the user an easy way to distinguish between a Network/Low-level IOE and a Parsing IOE. The use case appeared recently in the HBase project:

Originally, if a RegionServer got an IOE from HDFS while opening a region file, it would abort the open and let the HMaster reassign the region. The assumption being that this is a network failure that will likely disappear at a later time or different partition of the network. However, if HBase gets parsing exceptions, we want to log the problem and continue opening the region anyways, because parsing is an idempotent problem and retries won't fix this issue.

Although this problem was found in HBase, it seems to be a generic problem of being able to more easily identify idempotent vs transient errors.

Issue Links

• is depended upon by: HBASE-2933 Skip EOF Errors during Log Recovery
• is required by: HBASE-3081 Log Splitting & Replay: Distinguish between Network IOE and Parsing IOE