spike: service redundancy

[rayjlinden]

This is more of a question I suppose than a bug report. But it is about the current architecture of paygate that appears like it could be a problem for us.

Currently, paygate "holds" generated ACH files in a directory in the container. It would then write that out to FTP site closer to the given cutoff time.

For this design - how would things work if I want to run multiple paygate services for redundancy and (eventually!) volume/scale reasons? One service would not know about the staged files of the other? And related to that - what if a service unexpectedly dies?

Obviously, we could use a mount a volume to store the "interim" files that can be remounted to deal with a service getting restarted. However, sharing that volume amongst multiple services becomes a problem.

Maybe you already have a plan to address this issue. Can you share that plan if so? I want to make sure it is going to work with our current infrastructure capabilities. If not, can we talk about some options? :)

[adamdecaf]

I've thought about how we could accomplish this. An initial thought would be using a store like etcd to hold a lock over a filename and store the file contents there.

As a separate issue I know improving paygate's flexibility to upload would help. I'm not sure if an admin endpoint to upload or just doing so more frequently would help.