Gotcha #2: Session Life
Closely related to the issue of session persistence is session life. In many cases, sessions stored in memory stay in memory for long periods of time, which makes users happy because if they are interrupted for some reason they can return to the application and pick up where they left off.
This is a potential issue for applications when deployed in the cloud because it's not guaranteed that the server instance on which the session was created will still be on-line when a user returns to the application. It is possible that sessions can effectively be "lost" during relatively short periods of time due to a server instance becoming idle and subsequently being de-provisioned.
This gotcha is highly dependent on the cloud computing provider's definition of "idle" and how it is determined that an application instance should be de-provisioned. If it is based on current connections and idle time, losing sessions is likely. If the provider's provisioning services have visibility into the application server such that it can determine that sessions are still active and determining idle time is based on this information, at least in part, then this scenario is not likely to occur.
- As with session persistence, you can use session mirroring or database backed sessions to avoid losing a session when an instance goes idle for too long.
- Work with the cloud computing provider to determine whether you can set the parameters for what is considered "idle" and what is not. You may be able to idle the instance for the duration of a session, essentially using session-based quiescence rather than connection-based quiescence to determine when an application instance can be taken off-line. Be aware of the costs of doing so, however, as any instance, even idle, consumes some resources and will therefore incur charges.