Skip to content

Conversation

@DreamPearl
Copy link
Contributor

@DreamPearl DreamPearl commented Nov 20, 2024

PROTON-1442

AMQP Transaction Sequence:

  1. Declare transaction:
  • Client establishes link to Broker (Transaction resource) to target with transaction coordinator type (usual types are sender/receiver) (ATTACH frame)

  • Client (Transaction Controller) sends a special message to that link to create transaction (TRANSFER frame)

  • Broker returns a disposition with the transaction id (DISPOSITION frame)

  1. Send Message in the transaction:
  • Client sends a message to the broker with transaction id in the state field.
  1. Commit/Abort

@DreamPearl DreamPearl marked this pull request as draft November 20, 2024 08:57
@DreamPearl DreamPearl force-pushed the local-transactions branch 2 times, most recently from 74818fc to eb4514b Compare November 21, 2024 18:28
Copy link
Member

@astitcher astitcher left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I've taken a fairly thorough look. If you have any questions about these review items we should discuss further.

Comment on lines 101 to 62
class
PN_CPP_CLASS_EXTERN transaction_handler {
public:
PN_CPP_EXTERN virtual ~transaction_handler();

/// Called when a local transaction is declared.
PN_CPP_EXTERN virtual void on_transaction_declared(transaction);

/// Called when a local transaction is discharged successfully.
PN_CPP_EXTERN virtual void on_transaction_committed(transaction);

/// Called when a local transaction is discharged unsuccessfully (aborted).
PN_CPP_EXTERN virtual void on_transaction_aborted(transaction);

/// Called when a local transaction declare fails.
PN_CPP_EXTERN virtual void on_transaction_declare_failed(transaction);

/// Called when the commit of a local transaction fails.
PN_CPP_EXTERN virtual void on_transaction_commit_failed(transaction);
};

} // namespace proton

Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I think this should be in a different header file like message and messaging_handler

Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

So if the API makes transactions hidden in a session then these callbacks either go away or instead become session callbacks: on_session_transaction_committed(session&), etc. I think transaction_declared() goes away entirely and it's purpose is now another use for on_session_open(session&). on_.._declare_failed should be handled by the on_session_error(session&).

Copy link
Member

@astitcher astitcher left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I've not reviewed this comprehensively, but I think I'm now very much leaning towards not making the transaction class visible to the API user at all and having the transaction methods on the session. This is essentially the API in JMS and CMS.

Comment on lines 101 to 62
class
PN_CPP_CLASS_EXTERN transaction_handler {
public:
PN_CPP_EXTERN virtual ~transaction_handler();

/// Called when a local transaction is declared.
PN_CPP_EXTERN virtual void on_transaction_declared(transaction);

/// Called when a local transaction is discharged successfully.
PN_CPP_EXTERN virtual void on_transaction_committed(transaction);

/// Called when a local transaction is discharged unsuccessfully (aborted).
PN_CPP_EXTERN virtual void on_transaction_aborted(transaction);

/// Called when a local transaction declare fails.
PN_CPP_EXTERN virtual void on_transaction_declare_failed(transaction);

/// Called when the commit of a local transaction fails.
PN_CPP_EXTERN virtual void on_transaction_commit_failed(transaction);
};

} // namespace proton

Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

So if the API makes transactions hidden in a session then these callbacks either go away or instead become session callbacks: on_session_transaction_committed(session&), etc. I think transaction_declared() goes away entirely and it's purpose is now another use for on_session_open(session&). on_.._declare_failed should be handled by the on_session_error(session&).

transaction.accept(d);
current_batch += 1;
if(current_batch == batch_size) {
transaction = proton::transaction(); // null

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I believe we should do rather commit here, that way it works as expected and the receiver is closed after expected number of messages received (with the current implementation it's not).

Copy link

@pematous pematous Feb 5, 2025

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I believe we should do rather commit here, that way it works as expected and the receiver is closed after expected number of messages received (with the current implementation it's not).

above mentioned change:

-            transaction = proton::transaction(); // null
+           transaction.commit();

works as expected (against Artemis broker) when using multicast. However, when a pre-defined anycast queue is used, the commit() call ends with client segmentation fault.


void on_message(proton::delivery &d, proton::message &msg) override {
std::cout<<"# MESSAGE: " << msg.id() <<": " << msg.body() << std::endl;
transaction.accept(d);

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

btw. I'm thinking if transaction.accept() does actually have any effect? I mean, when I comment it out, the program behaves the same. btw. is this transaction accept any different from delivery accept? looks both settles the messages related to the delivery.

Does explicit delivery settling play role in transaction mode? or are they mutually exclusive and only commits and aborts applies? Can ie. single message in transaction be rejected while other messages accepted?

I tried to accept / reject the delivery and it seems to have no effect in transaction mode (while in python, doing so makes the messages to be threat outside of the transaction, just like a normal messages, https://issues.redhat.com/browse/ENTMQCL-513).

Copy link
Member

@astitcher astitcher left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Lopts of detailed comments here, but the main change I'd like to see is either make coordinator a subclass of target or fold it into a simple boolean option of target.

struct pn_session_t;

namespace proton {
class transaction_impl;
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Is this actually needed here?

Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

transaction_impl is pirely internal and should not be in an externally visible header file

Copy link
Member

@astitcher astitcher left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

First batch of comments

struct pn_session_t;

namespace proton {
class transaction_impl;
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

transaction_impl is pirely internal and should not be in an externally visible header file

@DreamPearl DreamPearl force-pushed the local-transactions branch 2 times, most recently from 4cfc58b to 6eb58f4 Compare October 4, 2025 05:18
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

3 participants