fix(satp): shutdown of SATP gateway should assure there are no pending transfers before shutting down

[RafaelAPB]

Describe the bug

The shutdown code at https://github.com/hyperledger-cacti/cacti/blob/satp-dev/packages/cactus-plugin-satp-hermes/src/main/typescript/plugin-satp-hermes-gateway.ts#L585 does not clean up existing sessions. It should wait that all transfers are resolved prior to shutdown.

To Reproduce
NA

Expected behavior

A call to gateway.shutdown() should wait for all pending sessions to counterparty gateways to be resolved.

NA

[LordKubaya]

@RafaelAPB should we wait forever for the sessions to finish? Should we have a function that forces the gateway to shutdown and triggering rollbacks of the unfinished sessions?

[RafaelAPB]

@LordKubaya I'd say we should wait indefinitely because if the crash recovery timeout is triggered then the bad session is handled, there will be no more pending sessions and the gateway can shutdown gracefully. In other words, the gateway should have to wait at most the delta for crash recovery + any extra time the rollback steps take. If this is not respected we can emit warnings/errors but IMO we should wait.

Otherwise we could be unintentionally be causing our gateway to crash from the perspective of other gateways, consequently triggering the (expensive) rollback protocol

[LordKubaya]

@LordKubaya I'd say we should wait indefinitely because if the crash recovery timeout is triggered then the bad session is handled, there will be no more pending sessions and the gateway can shutdown gracefully. In other words, the gateway should have to wait at most the delta for crash recovery + any extra time the rollback steps take. If this is not respected we can emit warnings/errors but IMO we should wait.

Otherwise we could be unintentionally be causing our gateway to crash from the perspective of other gateways, consequently triggering the (expensive) rollback protocol

@RafaelAPB

Ok, that makes sense.
Imagine that we need to shutdown the gateway fast (maintenance, etc), we can have a mecanism to transfer the transfer to other gateway that have support for the transfer we are doing?

[RafaelAPB]

I'd say no, if the gateway does not answer it is consider crashed.