I realized ISC WD doesn't seem to work well when IFO was locked with ALS_DARM.
Figure 1 shows the ISC WD signals when IFO went DOWN during the lock acquisition.
Even though IFO failed to lock, ISC WD signals (blue line in the top graph) doesn't touch the threshold (red line in the same graph).
It causes a large kick of the suspensions, which results in the long interval to the next lock trial to wait for the suspension calm down.
To improve the situation, it would be better to try to set ISC WD with proper threshold.