A sub-class of FlowLogic implements a flow using direct, straight line blocking code. Thus you can write complex flow logic in an ordinary fashion, without having to think about callbacks, restarting after a node crash, how many instances of your flow there are running and so on.
Invoking the network will cause the call stack to be suspended onto the heap and then serialized to a database using the Quasar fibers framework. Because of this, if you need access to data that might change over time, you should request it just-in-time via the serviceHub property which is provided. Don't try and keep data you got from a service across calls to send/receive/sendAndReceive because the world might change in arbitrary ways out from underneath you, for instance, if the node is restarted or reconfigured!
Additionally, be aware of what data you pin either via the stack or in your FlowLogic implementation. Very large objects or datasets will hurt performance by increasing the amount of data stored in each checkpoint.
If you'd like to use another FlowLogic class as a component of your own, construct it on the fly and then pass it to the subFlow method. It will return the result of that flow when it completes.
If your flow (whether it's a top-level flow or a subflow) is supposed to initiate a session with the counterparty and request they start their counterpart flow, then make sure it's annotated with InitiatingFlow. This annotation also has a version property to allow you to version your flow and enables a node to restrict support for the flow to that particular version.
Functions that suspend the flow (including all functions on FlowSession) accept a maySkipCheckpoint parameter defaulting to false, false meaning a checkpoint should always be created on suspend. This parameter may be set to true which allows the implementation to potentially optimise away the checkpoint, saving a roundtrip to the database.
This option however comes with a big warning sign: Setting the parameter to true requires the flow's code to be replayable from the previous checkpoint (or start of flow) up until the next checkpoint (or end of flow) in order to prepare for hard failures. As suspending functions always commit the flow's database transaction regardless of this parameter the flow must be prepared for scenarios where a previous running of the flow already committed itsrelevant database transactions. Only set this option to true if you know what you're doing.