Added two new properties for configuration to a network bridge,
advisoryPrefetchSize and advisoryAckPercentage. By default
advisoryPrefetchSize is set to 0, which is disabled, and will use the
prefetchSize value unless otherwise set. Also added validation to
prefetchSize to make sure it is greater than 0 as 0 is not allowed.
Moving beforeMarshall back to the store implementations because we don't
want all store implementations to marshall (such as memory store)
This reverts commit b9b98a45ce.
Reverting the change to move clearMarshalledState into the callback
listener as beforeMarshall is now called before the async message add
Revert 75990ef14a
Update for some added thread safety. Adds method healthStatus that will
regenrate the status from the healthList data which is more intuitive
than the getCurrentStatus which doesn't update state and requires
periodic calls to healthList to capture current metrics.
Refactor the iterator implementation in the PrioritizedPendingList to
not copy elements and instead use the level iterators. Add some
additional tests.
Disabling async queue message store when persistJMSRedelivered is turned
on for a destination. That flag will cause a sync update later on
dispatch which can cause a race condition if the original message add is
processed after the update. This can cause a duplicate message to be
stored.
Moving clearedMarshalledState execution to the async listener on an
async add to the message store. This is necessary to make sure this
logic doens't execute until after the message is marshalled for the
store.
Fixing the removal logic on virtual destination remove inside of
Advisory Broker to clean up virtual destination maps properly. Added a
test to verify. Also added new debug logging to help track down any
future issues.
Fixing an issue that caused statistics for a transactional temporary
queue to be updated twice by mistake. Moved the statistics update into
tryOrderedCursorAdd so that the update is called immediately only if
not inside a transaction.
Slight refactoring to make the WildcardFinder class use a static method
so that we don't have to allocate throw away objects that will
immediately be garbage collected.
- recoverSubscription()
-- map is defined as LinkedHashMap<MessageId, Message>
-- msg is defined as <map> entry.getValue() so must be a Message
-- condition if (msg.getClass() == MessageId.class) could never be true
-- no need to cast at all when using generics
- recoverNextMessages()
-- basically same code copy/pasted so same fix
Removed 2 conditions from ServerSessionPoolImpl that would result in impossible casts. Conditions removed were trying to cast ActiveMQQueueSession and ActiveMQTopicSession to ActiveMQSession which is illegal.
Since it isn't obvious what to do if you get an ActiveMQQueueSession or ActiveMQTopicSession from getServerSession() I make it fall back to the else condition which raises an async exception. This is better than getting a ClassCastException at runtime.
Remove impossible cast in MemoryMessageStore
Ensure that a message sent to the store for the DLQ is rewritten so that
its updated values are written to prevent exirpation loops and loss of
reollback cause etc.
AMQ-6122
Prevent messages on DLQ for looping back onto the same DLQ. Prevents
expired messages from bouncing back and duplicate messages from the
store from causing a deadlock.
The JavaRuntimeConfigurationBroker can now apply a subset of policy
properties retrospectively to existing destinations versus applying
all properties of the policy update.
Fixed contains method in PrioritizedPendinList which was not returning
correctly. This was causing messages to not be removed from the
dispatchPendingList when purge was called inside a Queue leading to an
eventual OOM error if enough messages were purged. This fix also
improves performance of the contains method.
Reverting commit because allowing 0 doesn't actually set the usage to
0 size. It ends up disabling the check all together which was the
opposite intention of the commit.
This reverts commit 3b658f8e1a.
Switching addMessageLast to tryAddMessageLast when messages are added
to a Queue pending cursor to allow a potential deadlock to be
avoided. There is more work to be done here but this will at least
prevent a deadlock from occurring.
Fix and test based off of a patch created by Timothy Bish.
Adding support for consumers on virtual destinations to create network
demand. This behavior is turned off by default but can be enabled.
For example, if a consumer comes online for a queue that is part of a
VirtualTopic, this will cause a network of brokers to forward messages
because a demand subscription will be created. Same for if a consumer
comes online for a forwarded destination from a composite
destination.
There is also an option to enable flow based on the existence of a
virtual destination if the virtual destination is forwarding to a
Queue.
Full configuration instructions for this feature will be on the wiki page.
Improving the warning message for a percent disk usage check so it is
more clear what the original limit was set to and what the current
limit is being set to.
When possible check for noLocl changes on durable subscription
reactivation and recreate the durable sub if it changes. For both
selector change and noLocal change also update the AbstractSubscription
selectorExpression so it matches with what was requested.
Adding metrics to track the pending message size for a queue and for
subscribers. This is useful so that not only the pending count is
known but also the total message size left to consume. Also improving
the message size store tests as well.
-Cleaned up missing license headers and refactored packages.
-Added configuration options for a protocol detection timeout and
for the max number of connections accepted at the same time.
-Fixed a regression with connection counts
-Also added some more tests
Adding support for auto detection of wire protocols over a transport.
OpenWire, AMQP, STOMP, and MQTT can all be detected and the broker
will properly handle each one over a given Transport. Currently
auto TCP, NIO, SSL, and NIO+SSL transports can handle auto-detection
of the wire format and client but support could be added in the
future for other transports like websockets.
Updating MemoryTopicMessageStore to decrement store statistics on cache
eviction. Updating KahaDBMessageStoreSizeStatTest to account for the
fact that a LRU cache is used so the last 100 messages are kept in
memroy.
Switching to a List to track dispatched messages in a TopicSubscription
to be consistent with a PrefetchSubscription and to prevent an error
in case acks come back out of order.
Apply patch from Ganesh Murthy to resolve issue of creation of durable
subscriber fails via broker AdminView due to lack of SecurityContext on
ConnectionContext
Reworking DurableConduitBridge to look up subscriptions from the
TopicRegion instead of the RegionBroker since inactive subscriptions
need to be looked at as well when dynamicOnly is false for a
network bridge.
resolves#123
This commit resolves an issue where unmatched acknowledgement
messages could be received when running a MDB consumer and
sending messages with a short TTL. The expiration logic when
receiveing an expired Message Ack will now only expire messages
in dispatch relating to the received ack, not all expired messages
in the dispatch list.
Use the latest openwire version marshallers in the KahaDB store when
starting from a clean install, drop back to the version used in the
existing store if one is found.
This commit adds an Inflight message size statistic to SubscriptionStatistics
so we can know the size of all the messages that are inflight, besides just the count.
Added a getMessageSize method to MessageStore to support retrieving the
total message size of all stored messages for a destination. Added a
new storeMessageSize statistic to DestinationStatistics.
Remove of a durable subscription when the keepDurableSubActive flag is
true (default) did not actually stop and cleanup the still active cursor
due to the fact that the cursor checks the flag of its subscription to
see if it should stay active when its stop method is called. The
subscription needs to disable that flag when the subscription is being
removed so that the cursor will correctly shutdown.
A QueueBrowser no longer checks expiry on messages to prevent a
browser from hanging in between the hasMoreElements check and actually
getting a message. This means that if messages were in the queue when
the browser started it will receive messages even if they are now
expired. Even though the browser will get the expired message, the
broker will still expire it to prevent future access to it.
Thanks to Henno Vermeulen for providing a test case.
This commit is based off a patch submitted by admin@int3solutions.com.
It fixes several locations in the code where there are potential resource leaks.
fix for incorrect logic in reconnect attempt limits and also fix a case
where an NPE could occur (saw in the tests). Based on patch provided
by Benoit Wiart
Timothy Bish
gtully
Quinn Stevenson
Christopher L. Shannon (cshannon)
Dejan Bosanac
Brian D. Johnson
Jeff Genender
Patrik Dudits
Marc Breslow
Daniel Kulp
Altaflux
Chad Zobrisky
Christian Schneider