[ws-daemon] Ensure add handler is called before update

[Furisto]

Description

Based on https://gitpod.slack.com/archives/C033MA45L5S/p1662729393858979 the following race condition can happen:

• New workspace is found which means it is added to the dictionary in the dispatcher
• We wait for the container to be ready and once that has happened we set the ContainerSeen state to true
• We call the add handler of the listeners to process the event
• An update event arrives and it is processed before the handler for the add event is called (the condition for the processing of the update event is that the workspace is in the dictionary and that we have seen the container, which is now fullfilled)
• 💥 The listener has not seen the workspace yet and returns an error
• The add handler of the listener is called and the workspace is added to the listener

Instead we wait for the add handlers to be processed before we allow update events. This means we have to hold the lock slightly longer but we call the handlers in separate go routines, so the hold duration is only increased by the amount of time it takes to spin up the go routines which should be negligible.

There is still a chance that the listeners themselves have not finished the processing of the add event when the update event handlers is called (due to being called in separate go routines), but due to the locking internal to the listeners we ensure that add and update events are processed atomically, so this does not matter.

Related Issue(s)

n.a.

How to test

Not sure, would be difficult to reproduce

Release Notes

None

Werft options:

• /werft with-preview