I'm not absolutely positive but based on Signal's notification service extension code found here, what you're saying is correct. The way they solve it is as such:
// The lifecycle of the NSE looks something like the following:
// 1) App receives notification
// 2) System creates an instance of the extension class
// and calls this method in the background
// 3) Extension processes messages / displays whatever
// notifications it needs to
// 4) Extension notifies its work is complete by calling
// the contentHandler
// 5) If the extension takes too long to perform its work
// (more than 30s), it will be notified and immediately
// terminated
//
// Note that the NSE does *not* always spawn a new process to
// handle a new notification and will also try and process notifications
// in parallel. `didReceive` could be called twice for the same process,
// but will always be called on different threads. To deal with this we
// ensure that we only do setup *once* per process and we dispatch to
// the main queue to make sure the calls to the message fetcher job
// run serially.
Make sure you see the rest of their code