From: Sam Wilson [mailto:Sam.Wilson@...]
Sent: 11 November 2014 4:46 PM
To: Glenn McClements; Tom Doust; openmama-dev@...
Subject: Re: [Openmama-dev] Asynchronous publisher events
Hey Glenn, Tom,
I have to agree with Glenn here. I don't think these kind of events belong alongside regular messages, for a couple of reasons.
For one, simply enough, they aren't messages. They don't originate from a publisher, they wouldn't travel along the wire, and they probably would have to be synthesized in the middleware bridge.
Secondly, Tom's solution only works for sendFromInbox. The callback mechanism is much more general, and would work for non-inbox messages as well, which is important for our use case.
Assuming we do go ahead with implementing callbacks for asynchronous publisher events, what needs to be thought about/discussed before moving on to writing/accepting a patch?
I have two ideas on how to implement this, but keep in mind that I'm rather new to OpenMAMA and market data in general.
We could use a mamaTransportTopic event, and report the errors with a callback on the transport.
The other option that I see, which Glenn hinted at, is that we create a structure similar to mamaMsgCallbacks, called mamaPublisherCallbacks; and a pair of new methods mamaPublisher_createWithCB and/or mamaPublisher_setCallbacks. That way we report the error on the specific publisher.
What are your thoughts?
On 14-11-07 10:57 AM, Glenn McClements wrote:
Hi Tom, Sam,
Before you responded I was considering the use of an inbox as a
solution to Sam’s query but I wasn’t overly keen on it. The reason
being that being rejected by the broker felt very much like a
middleware level event, not an application or market data event so it
feels better to me that the error comes from the middleware as a callback.
Also, responding with a MamaMsg to the sender means assuming the
connection is up and alive, whereas a middleware ACL policy might
actually prevent this from happening. It could be that you form the
message on the client side in the bridge, but this itself feels
incorrect as it’s translating from a middleware level even into a data
message. (Also a minor point is that the MAMA_MSG_TYPE_SEC_STATUS
relates to security status events on a exchange). This is not to say
that a message response is incorrect for RMDS posts however, it’s just a different type of event.
For Sam’s point I do prefer the idea of currently returning a
MAMA_TRANSPORT_PUBLISHER_DISCONNECT, and the transport is currently
the best/only place to do this.
Going forward we could add callbacks to the MamaPublisher object to
inform the client asynchronously of events like this, and
MAMA_TRANSPORT_TOPIC_PUBLISH_DENIED seems reasonable though we would
need to consider how this would be used/interact with the higher level
concept of market data entitlements.
Another related concept is guaranteed or acknowledged messaging, which
would generate other asynchronous event on a publishers. Like the
original issue, this *could* be done at a high level with MamaMsgs
being passed back to acknowledge delivery, but generally this is
implemented at the middleware level so again a publisher callback
would be better (with a handle to the original message as well for context).
On 07/11/2014 14:20, "Tom Doust" <tom.doust@...> wrote:
In the Tick42 RMDS bridge we address this problem by allowing
applications to use the sendFromInbox methods on a publisher and
where applicable converting the an asynchronous response from the
RMDS into a mama message which is delivered to the inbox.
Although this is intended to support the RMDS message "post"
publishing model that will ACK (or NAK) every message I think it will
work for any middleware that has an asynchronous response mechanism;
other publishing models on RMDS do not have a message reponse.
The message sent to the inbox is arbitrary but it would make sense if
other bridges that used this technique used the same message structure.
The message we use is formed as follows
mamaMsg_addI32(msg, "MdMsgType", 1, MAMA_MSG_TYPE_SEC_STATUS);
mamaMsgStatus status = MAMA_MSG_STATUS_OK;
if(nakCode != RSSL_NAKC_NONE)
status = RsslNakCode2MamaMsgStatus(nakCode);
mamaMsg_addI32(msg, "MdMsgStatus", 2, status);
mamaMsg_addString(msg, "wSymbol",470, symbol_.c_str());
As you can see we just convert an rssl NAK code to an equivalent mama
code, or use an OK status in the case of an ACK. These include states
like MAMA_MSG_STATUS_NOT_ENTITLED and MAMA_MSG_STATUS_BAD_SYMBOL
amongst a number of others. Obviously the set of code and the
mappings are platform specific
The one thing I think is missing here is a field containing a text
string which could carry more information relevant to the client application.
We would propose that everyone addressing the issue of processing
asynchronous publishing responses takes this approach and that as a
community we agree on a convention for the message content.
[mailto:openmama-dev-bounces@...] On Behalf Of Sam
Sent: 06 November 2014 7:45 PM
Subject: [Openmama-dev] Asynchronous publisher events
We're looking for a way to notify a mama application asynchronously
of errors while publishing, specifically when the published message
was rejected by the message broker's ACL.
Our API assumes the messages were accepted, and returns immediately.
After some time you might get a connection-level event callback
informing the application that its messages were rejected.
I have two questions.
First off, what is the best way to represent this situation in 2.3.1?
We've been tossing around the idea of just tearing down the
connection, giving the application a MAMA_TRANSPORT_DISCONNECT or a
MAMA_TRANSPORT_PUBLISHER_DISCONNECT, and logging the error. Since the
application or the environment is incorrectly configured, there's not
much that can be done by the application anyways.
Secondly, if openmama were to be extended in the future to support
these kind of events, what would such extensions look like? Perhaps
adding a MAMA_TRANSPORT_TOPIC_PUBLISH_DENIED to the
Openmama-dev mailing list
Openmama-dev mailing list