[Date Prev] | [Thread Prev] | [Thread Next] | [Date Next] -- [Date Index] | [Thread Index] | [Elist Home]
Subject: [ebxml-msg] reliable messaging
It has been pointed out to me that ebXML reliable messaging is not reliable under system failure. At least one person who mentioned it considers ebXML messaging to be broken as a result. Here is a scenario: Party A send a message reliably to Party B. Party B's MSH receives and persists the message. Party B's MSH attempts to send the reliable-messaging acknowledgment but Party B's system goes down before the acknowledgment gets on the wire. Party A exhausts its retries and concludes that the message was not delivered. Party B eventually comes up and the destination application processes the persisted message as prescribed in the MSG specification. Parties A and B are now out of sync with respect to that transaction and do not know they are out of sync. Party A believes that the transaction failed. Party B has in fact processed the message that it received from Party A. Reliable messaging has failed to deliver on its promise. The solution to this problem is not trivial and the MSG team needs to give it a lot of thought. At a minimum, the following are needed in the spec: 1. Both parties to the message exchange MUST persist enough state to allow recovery and getting back in sync. Specific state variables must be prescribed. They are at least those variables needed to restore the state of the transaction and conversation after system recovery, such as the conversation ID, CPA Id, service, action, and perhaps other parts of the message header. 2. Timeouts and retries, as prescribed in the MSG spec, are not sufficient to cover system failures since the failure could last a very long time. Instead, if the party that sent the message doesn't receive a reply in a reasonable time, it must be able to send a status query to the other party and keep requesting status periodically until it receives a response. The status query protocol must be defined in the MSG specification. If the appropriate state information is persisted at both ends, when party B comes up, it will receive and respond properly to the status query. The timeouts could be retained in the spec but their main use would be to signal the "attached human" to make a phone call. The MSG team should consider this a work item for version 3. Should the team not wish to solve this problem, at the very least, a caveat should be added to the MSG specification that messaging reliability under conditions of system failure is outside the scope of the MSG team. Regards, Marty ************************************************************************************* Martin W. Sachs IBM T. J. Watson Research Center P. O. B. 704 Yorktown Hts, NY 10598 914-784-7287; IBM tie line 863-7287 Notes address: Martin W Sachs/Watson/IBM Internet address: mwsachs @ us.ibm.com *************************************************************************************
[Date Prev] | [Thread Prev] | [Thread Next] | [Date Next] -- [Date Index] | [Thread Index] | [Elist Home]
Powered by eList eXpress LLC