Improve error handling in background thread

[cmumford]

Original issue 174 created by dgrogan@chromium.org on 2013-05-29T19:14:11.000Z:

There's a report[1] on the chromium issue tracker of a background thread eating 10-20% of CPU. This is reportedly due to leveldb encountering an error during a compaction, sleeping for 1s, and retrying in an infinite loop. Instead of hardcoding 1 second, we could instead not retry at all for Corruption errors and do a limited exponential backoff for I/O errors.

1. https://code.google.com/p/chromium/issues/detail?id=244932

[cmumford]

Comment #1 originally posted by dgrogan@chromium.org on 2013-05-29T19:14:25.000Z:

<empty>

[cmumford]

Comment #2 originally posted by dgrogan@chromium.org on 2013-06-13T23:29:59.000Z:

LevelDB 1.11.0 does a very limited exponential backoff. The first error encountered by the compaction thread causes a 1 second wait. The second consecutive error causes a 2 second wait, etc, up to a max of 8 seconds.