I have the impression that this is related to your frame, like if there was some aerodynamic effect that creates a pitch up torque: the controller produces a downward pitch control signal that goes through the allocator and sets the two front motors that ramp down to their minimum value while the rear ones ramp up until saturation but the drone still pitches up. At the same time you have a roll error (roll-pitch coupling?)
Are you really flying a quad? Do you have aerodynamic surfaces or heavy spinning rotors (gyroscopic effect) or an asymmetric drone (large inertia cross-coupling) ? I also see that you have extremely large integral values, which shows that it isn’t a standard setup (or that you changed the PID architecture).
Furthermore, during that time the drone doesn’t climb nor descend (stays at 121m ?? wrong measurement ?)
Note on the side: I can also tell you is that you have a constant negative yaw offset. A yaw offset with a nonlinear PWM-to-thrust curve produces roll-pitch coupling. Some ESCs try to linearize that curve, others don’t.
Hope that helped, maybe I can give you more help if you tell us a bit more about the drone.