I seem to be greatly in the minority here (a familiar position!) but IMO if someone sets off the powderkeg without warning it is - or should be - that act of setting it off that starts proceedings. This means that either the setter-off should act first if most or all others are surprised*, or should get a whacking big bonus on initiative if they are not.
Don't get me wrong, I support the 5e initiative system because it works most of the time while being really simple, which is an immense quality - and for me one of the reasons of success of 5e, compared to games who try to envisage all edge cases in the main rules and become unreadable and almost unplayable.
That being said, the second strength of 5e is to have fully reinstated the DM's prerogative, and in particular rulings "over rules", which means that I certainly don't forbid myself to create specific rulings for specific situations, like lighting the powerkeg meaning that:
- People were really not expecting this, so the guy doing the lighting is surprising everyone else, and all the others, being surprised, don't act in the first round.
- He was in such a good position (physically, mentally, etc.) that I give him advantage on initiative.
- It was a readied action to light the powerkeg if someone did something particular, so he will light the powerkeg right at the combat start (this is our house rule and one of the reasons for it).
- etc.
As usual, the problem is some people (not you) arguing from a position of bad faith and pretending the whole system does not work because it''s not catering to very specific edge cases, completely forgetting the principles that made a success of 5e and the fact that no system has ever been written which properly took into account all edge cases anyway.
I've had this argument with my DM numerous times, when I've in theory been the setter-off but by the time my initiative comes around my intended action would be nearly pointless.
See above, for me the great principles of 5e apply at our table and clearly allow for circumstances to create local rulings anyway.
* - don't get me started on 5e's garbage surprise rules as those are a whole other issue.
It's not perfect, but once more it works in most cases, and, as with stealth, there are so many edge cases that no system could account for that properly anyway. If you have an edge case, do a local ruling and that's the end of the story. This has worked for us every time, and we find it infinitely superior to having to read 10 times more rules that we will never apply anyway and which are in any case not complete enough to cover our specific edge case.