This reverts commit 8fcfe91a24.
Revert "Preparing hbase release 2.5.5RC0; tagging and updates to CHANGES.md and RELEASENOTES.md"
This reverts commit 429b34a969.
Preparing to restage 2.5.5RC0.
We need to add a native library load check with a helpful error message
if xerial snappy fails to initialize due to a too old glibc or similar
reason, and disable the unit test if the native library fails to load.
Signed-off-by: Duo Zhang <zhangduo@apache.org>
Signed-off-by: Viraj Jasani <vjasani@apache.org>
When we evict a block from L1 and move it to L2 there is a brief window of
time where we have removed the block from the L1 map and yet the victim
handler has not completed execution. Some read-your-write use cases can be
significantly impacted even though the window is small.
Victim handling can be made atomic with respect to the unmapping operation.
The upside is there will be no L1+L2 misses during the transition. The
downside is if the victim handler takes a long time to execute – currently
they are all very fast, so only a theoretical risk – then other removals or
insertions in L1 can block until it completes.
Signed-off-by: Duo Zhang <zhangduo@apache.org>
Signed-off-by: Viraj Jasani <vjasani@apache.org>
When HFiles are placed on a filesystem other than HDFS a rename operation can be
a non-atomic file copy operation. It can take a long time to copy a large hfile
and if interrupted there may be a partially copied file present at the destination.
If we fail to “rename” the files into the archive we will continue to fail
indefinitely. Before larger changes are considered, perhaps to StoreFileTracker, we
should mitigate this problem.
Signed-off-by: Duo Zhang <zhangduo@apache.org>
Signed-off-by: Viraj Jasani <vjasani@apache.org>
Signed-off-by: Wellington Chevreuil <wchevreuil@apache.org>
Signed-off-by: Xiaolin Ha <haxiaolin@apache.org>
From 50^3 to 30^3, 125000 -> 27000
Signed-off-by: Liangjun He <heliangjun@apache.org>
Signed-off-by: Xin Sun <ddupgs@gmail.com>
(cherry picked from commit 8b8c684bcc)
Surefire version updated from 3.0.0-M6 -> 3.1.0.
SurefireForkNodeFactory is a new strategy to control how the forked
nodes communicate with the main maven process. It uses a tcp channel
instead of pipes and fixes some corrupted messages seen in the s390x
build.
Signed-off-by: Duo Zhang <zhangduo@apache.org>
(cherry picked from commit 67612cc264)
HBaseClassTestRule applies a timeout and a system exit rule to tests.
The timeout rule throws an exception if it hits the timeout threshold.
Since the timeout rule is applied after the system exit rule, the
system exit rule does not see the exception and does not re-enable
the system exit behavior which can cause maven to hang on some
tests.
This change applies the timeout rule before the system exit rule so
that normal system exit can be restored before the surefire forked
node is shutdown.
Signed-off-by: Duo Zhang <zhangduo@apache.org>
(cherry picked from commit 220eacf7c5)
Note: This version includes a fix that prevented zstd-jni from
running on s390x.
Signed-off-by: Duo Zhang <zhangduo@apache.org>
(cherry picked from commit 5d2b6956e5)
Also done some cleanup around the MockMasterServices related classes and tests
Signed-off-by: Liangjun He <heliangjun@apache.org>
(cherry picked from commit 89e80da57f)
Also modified the tests so we will only start the cluster once instead
of start it in setUpBeforeClass and then restart it when running the
actual test method
Signed-off-by: Liangjun He <heliangjun@apache.org>
(cherry picked from commit 21d61cfa1c)
Track and log better stats.
Fix use of memcached API such that we are not accidentally immediately expiring keys.
Use a default retention period of 30 days, the max per memcached spec.
Use set instead of add to store keys.
Gracefully handle failures to cache and read timeouts.
Add unit tests using jmemcached as a test dependency.
Signed-off-by: Viraj Jasani <vjasani@apache.org>