Victor Xu created FLINK-24460:
---------------------------------
Summary: Rocksdb Iterator Error Handling Improvement
Key: FLINK-24460
URL: https://issues.apache.org/jira/browse/FLINK-24460
Project: Flink
Issue Type: Improvement
Components: Runtime / State Backends
Affects Versions: 1.14.0
Reporter: Victor Xu
In FLINK-9373, we introduced RocksIteratorWrapper which was a wrapper around
RocksIterator to check the iterator status for all the methods. At that time,
it was required because the iterator may pass the blocks or files it had
difficulties in reading (because of IO errors, data corruptions, or other
issues) and continue with the next available keys. *The status flag may not be
OK, even if the iterator is valid.*
However, the above behaviour changed after
[3810|https://github.com/facebook/rocksdb/pull/3810] was merged on May 17, 2018:
*- If the iterator is valid, the status() is guaranteed to be OK;*
*- If the iterator is not valid, there are two possibilities:*
*1) We have reached the end of the data. And in this case, status() is OK;*
*2) There is an error. In this case, status() is not OK;*
More information can be found here:
https://github.com/facebook/rocksdb/wiki/Iterator#error-handling
Thus, it should be safe to proceed with other operations (e.g. seek, next,
seekToFirst, seekToLast, seekForPrev, and prev) without checking status(). And
we only need to check the status if the iterator is invalid. After the change,
there will be less status() native calls and could theoretically improve
performance.
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
