Counter intuitive move 2

I was analysing a game which concluded one hour ago when I saw an unusual response by Stockfish. The computer (I used Stockfish on Chess.com for basic accounts) actually suggests 19...Rhb8 in the following position. I did not even consider it at all.

I can roughly come up with a reason why the computer would prefer 19...Rhb8, but it just looks so weird, because it locks in the a8 rook temporarily and it feels a little uncomfortable to keep that a8 rook enclosed by regular game principles.

Does anyone feel that 19...Rhb8 is the best move here, or was I right to instinctively protect the pawn via 19...Rab8?

G4 trying to win the bishop for I guess 3 pawns looks suspect to me, so I can readily believe its inferior, but the rooks are harder to parse.  Maybe, since black is winning, you wouldn't want to answer qa4 with rba8, and what else is there?  A6 might be weakening.  Maybe, in terms of how you actually win, you want to make a queenside passed pawn, in which case the rooks on b8 and a8 make sense?  I'm not sure how else black wins, and making a pushable past pawn would be a normal plan.  Black's pieces are not active enough and his king is not safe enough to pursue another plan as near as I can see.  

Somehow I when I look at 19...Rhb8, I am scared of a "ganging up" on my d6 pawn. That would mean I would have to move my b8 rook to d8 again, which kind of loses a tempo. The a8 and b8 pawns are not ready to be pushed yet due to the "laser power" along the c- and d- files, so based on these grounds I did not consider 19...Rhb8 at all.

But yes, as you pointed out, it does make sense that White's queenside is weak and there is scope for a pawn promotion on the queenside, and it is somewhat true that my h8 rook is not really that useful there, which could be the basis on why 19...Rhb8 is preferred. I am not sure what a stronger Stockfish program would think about the position.

If I were to play this game again, I would probably still not be able to see 19...Rhb8 though.