The gravitational wave event from the binary neutron star (BNS) merger GW170817 and the following multi-messenger observations present strong evidence for i) merger ejecta expanding with substantial velocities and ii) a relativistic jet which had to propagate through the merger ejecta. The ejecta's expansion velocity is not negligible for the jet head motion, which is a fundamental difference from the other systems like collapsars and active galactic nuclei. Here we present an analytic model of the jet propagation in an expanding medium. In particular, we notice a new term in the expression of the breakout time and velocity. In parallel, we perform a series of over a hundred 2D numerical simulations of jet propagation. The BNS merger ejecta is prepared based on numerical relativity simulations of a BNS merger with the highest-resolution to date. We show that our analytic results agree with numerical simulations over a wide parameter space. Then we apply our analytic model to GW170817, and obtain two solid constraints on: i) the central engine luminosity as $L_{iso,0} \sim 3\times10^{49}-2.5\times10^{52}$ erg s$^{-1}$, and on ii) the delay time between the merger and engine activation $t_0-t_m < 1.3$ s. The engine power implies that the apparently-faint \textit{short} gamma-ray burst (\textit{s}GRB) \textit{s}GRB 170817A is similar to typical \textit{s}GRBs if observed on-axis.

Aiming to study GRB engine duration, we present numerical simulations to investigate collapsar jets. We consider typical explosion energy ($10^{52}$ erg) but different engine durations, in the widest domain to date from 0.1 to 100 s. We employ an AMR 2D hydrodynamical code. Our results show that engine duration strongly influences jet nature. We show that the efficiency of launching and collimating relativistic outflow increases with engine duration, until the intermediate engine range where it is the highest, past this point to long engine range, the trend is slightly reversed; we call this point where acceleration and collimation are the highest "sweet spot" ($\sim$10 -- 30 s). Moreover, jet energy flux shows that variability is also high in this duration domain. We argue that not all engine durations can produce the collimated, relativistic and variable long GRB-jets. Considering a typical progenitor and engine energy, we conclude that the ideal engine duration to reproduce a long GRB is $\sim$10 -- 30 s, where the launch of relativistic, collimated and variable jets is favored. We note that this duration domain makes a good link with Lazzati et al. (2013), which suggested that the bulk of BATSE's long GRBs is powered by $\sim$10 -- 20 s collapsar engines.