Re: [Quickfixj-users] Managing .seqnum files across different hosts?

[Colin D. <co...@ma...>]

We do something a little different to handle this problem. We do an 
in-memory cache of sequence numbers backed by a shared database with the 
in-memory cache being shared between machines by means of a cluster 
mechanism. One of the implementations is Hazelcast, but others would do 
just as well.

However, we don't worry too much about the seq nums being completely 
synced by as we do async instead of sync. Our theory is that the failure 
case is in the great minority, statistically, so we don't want to punish 
successful operation. On node failure, i.e. we lost the primary 
machine/process, when the backup takes over, it will replay a certain 
number of messages to make sure that nothing is lost. It does this by 
calculating the last message it can guarantee was received and 
completely processed and resetting the seq num to there before starting 
the session, which forces a seq num replay from the counter.

In order to make this work, the business logic has to be able to test an 
incoming message to see if it's already been processed, and skip it if 
it has.

On 10/31/18 4:23 PM, Robert Nicholson wrote:
> QuickFIX/J Documentation: http://www.quickfixj.org/documentation/
> QuickFIX/J Support: http://www.quickfixj.org/support/
>
>
> How do folks handle seqnum management in the context of secondary machines needing to know where the primary session finished when
> the secondary is activated?
>
> I’ve seen a couple of approaches such as asynchronous writes to a shared database.
>
> But what you don’t want is just a write to the database that’s synchronous as that would kill performance.
>
> Has anybody successfully done this with rsync seqnum files between servers?
>
>
>
> _______________________________________________
> Quickfixj-users mailing list
> Qui...@li...
> https://lists.sourceforge.net/lists/listinfo/quickfixj-users