Windows Azure中的大型会话状态对象

Question

I am building an online document editor (think google docs) with Windows Azure.

I need access to my document object every time a user performs an operation on it (keystroke = word completion, edit text = save, etc.)

The best setup I have been able to come up with so far is to have one or more dedicated cache role(s) in Azure, and retrieve the documents from there from my web role(s). However, this means that a document that could take up as much as 10 MB when serialized as JSON would have to be transmitted between the cache role and web role, potentially for each keystroke. It seems like an incredible amount of load to put on the system for a trivial task.

What I would rather do is to have the document loaded in memory on the web role that handles the request, but that conflicts with the stateless nature of Azure; it is difficult to do sticky load balancing, the logic for handling a server that crashes is complicated, and the latency benefit of Azure goes out the window if a specific document can only exist on one specific web role (= geographical location).

Any good ideas for a solution to this?

Answer 1

Try redesigning your design. You're right in that what you describe doesn't sound like a scalable approach.

Do you really need the document for work completion? I would think that you could make a call (to a separate REST server) that holds a dictionary to do the word completion. You could build a dictionary for each document if you wanted to autocomplete names and misspelled words too, but even then your dictionary probably would be a lot smaller.

as for saves, how often do they happen? automatically or only when the user presses the save button. You could fetch the document from the database, apply the save and store the result.

that said, it isn't easy to implement your own google docs.