This HBase release is a development release in the 0.89 series.

The development release series is meant to provide the HBase user community
with early access to the new features in HBase trunk, so they can begin
testing their workloads and providing feedback to the developers prior
to the official release of 0.90.0.

As a development release, there are many known bugs that have not been
addressed. This list is not comprehensive, but just serves as a short overview
of some known issues that may be encountered.

If you are encountering one of these issues, or encounter an issue that does
not appear to be previously known, please email user@hbase.apache.org with
a description, relevant logs, etc.

- HBASE-2774 Performance may drop to unacceptibly low levels with threads
  hung in completeMemstoreInsert

Reported on some clusters, write performance drops inexplicably to very low
throughput, then is restored upon flush.

- HBASE-2670 Gets and scans may see inconsistent views of rows being written

Under heavy load to a single row, a reader may see a view of the row that
contains a partially-applied edit made by a concurrent writer.

- HBASE-2238 Not all transitions are crash-proof

If a region server crashes while in the middle of certain transition actions
(eg splits, log rolling, etc) there are some situations in which the region
may become inconsistent, doubly assigned, or unassigned.

- HBASE-2727 Crashes that occur during region recovery may lose edits

If a region server crashes, its logs will be split and other region servers
will start to recover the data. If the recovering region servers crash
while performing recovery, some edits may be lost.