This is an automated email from the ASF dual-hosted git repository.
joewitt pushed a change to tag HDF-3.4.1.127-2-tag
in repository https://gitbox.apache.org/repos/asf/nifi.git
*** WARNING: tag HDF-3.4.1.127-2-tag was deleted! ***
was f08a736105 HOTFIX-3949 NIFI-7999: Do not call
NioAsyncLoadBalanceClient.nodeDisconnected() if node was already in a
disconnected state. Doing so was resulting in that method being called
constantly on startup, and with the synchronization in place that can result in
a huge performance hit on startup. Also updated RemoteQueuePartition to move a
small predicate into its own method. This was done because the predicate was
previously defined within a synchronized method, which meant [...]
This change permanently discards the following revisions:
discard f08a736105 HOTFIX-3949 NIFI-7999: Do not call
NioAsyncLoadBalanceClient.nodeDisconnected() if node was already in a
disconnected state. Doing so was resulting in that method being called
constantly on startup, and with the synchronization in place that can result in
a huge performance hit on startup. Also updated RemoteQueuePartition to move a
small predicate into its own method. This was done because the predicate was
previously defined within a synchronized method, which meant [...]
discard 45d611100f HOTFIX-3949: Fix build issue and test problems caused by
cherry-picks
discard 7e4a927a41 HOTFIX-3949 NIFI-7117: When SocketLoadBalancedFlowFileQueue
creates its array of Queue Partitions in the constructor, it added the local
partition as the first element in that list. This list should be ordered the
same across all nodes in the cluster. By making the local partition the first
in the array, each node had a different ordering of these partitions. As a
result, Partition by Attribute strategy would constantly rebalance flowfiles
that it received to other nod [...]
discard a64423d84b HOTFIX-3949: NIFI-7076: This closes #4024. Revert
"NIFI-6787 - Before: When checking if a load balanced connection queue is full,
we compare the totalSize.get() and getMaxQueueSize()."
discard 9d6d7ee8bf HOTFIX-3949: NIFI-6787: Added a comment to the UI tooltip,
user guide, and javadocs to indicate that round robin may skip a node if that
node is not receiving the data as fast as other nodes in the cluster, in order
to maximum throughput.
discard f863a63301 HOTFIX-3949: NIFI-6787 - Before: When checking if a load
balanced connection queue is full, we compare the totalSize.get() and
getMaxQueueSize(). After: Go over all partitions and see if all of them are
full. Wrapping RoundRobinPartitioner in a (new) AvailableSeekingPartitioner
which selects a new partition if the currently selected one is full.
discard d9805ecdd1 HOTFIX-3949: NIFI-6760: When writing/reading the length of
a DataFrame, do so usin… (#3801)
discard a8518895a0 HOTFIX-3949: NIFI-6759: When encountering Exception during
load-balanced connection communications, after closing socket, return from
method rather than continuing loop
discard 3a96ec3153 HOTFIX-3949: NIFI-6736: Create thread on demand to handle
incoming request from client for load balancing. This allows us to avoid
situations where we don't have enough threads and we block on the server side,
waiting for data, when clients are trying to send data in another connection
discard d920593ea9 HOTFIX-3949: NIFI-6696: Ensured that callback to
RemoteQueuePartition do not attempt to obtain the Partition Read Lock. The Read
Lock is not necessary as long as the 'partitioner' is volatile, because it
doesn't matter whether or not the actual partitions themselves change, since
the only partition that would be touched is the Rebalancing Partition,on, which
is fixed. Obtaining the partition read lock can lead to a deadlock as outlined
in the Jira description.
discard 3f5221b3e9 HOTFIX-3949: NIFI-6517: Ensure that if IOException is
thrown from LoadBalanceSession that we properly catch the Exception, mark
session as complete, and then re-throw. There was one condition in which this
was not occurring. This commit addresses that situation.
discard 2d700959f4 HOTFIX-3949: NIFI-6353: If LoadBalanceSession determines
remote node's queue is full, mark session complete and penalize the Partition,
rather than just penalizing the session. This allows the client to move on to
another session, which avoids an issue where backpressure on one queue can
prevent data from being transferred between nodes in another queue
discard 74da2814e3 HOTFIX-3949: NIFI-6285: Addressed issue that resulted in
swapped data not being swapped back in if load balancing strategy changed while
data was swapped out; added integration tests for swapping. In testing, also
encountered an issue with data being swapped out while swap files were being
recovered causing the queue size to be wrong and causing errors about not being
able to swap data in, because it attempted to swap the data in twice.
discard 40ef5eb015 HOTFIX-3949: NIFI-6183: Implemented base class for
Framework-level integration tests that is responsible for instantiating the key
components for integration tests, including FlowController,
WriteAheadFlowFileRepository, WriteAheadProvenanceRepository,
FileSystemRepository, FileSystemSwapManager, StandardFlowFileQueue,
StandardProcessScheduler, etc. Implemented several integration tests to span
different parts of the framework
discard 2ec4423d4d HOTFIX-3949: Contains the following changes: * NIFI-7079
security fix * NIFI-6085 - need for sec fix * NIFI-6280 - need for sec fix *
additional changes which needed for successful build
discard 2d96f44654 HOTFIX-3949 NIFI-7023 This closes #3991. Removed SLF4J and
Log4J transitive dependencies from Zookeeper so tests log correctly. Changed
template handling. Added unit tests.
