We looked into the raw data file g2p_3885.dat.0 on the disk, the prestart event suggest that it is actually run 3886. This problem also happened yesterday.
Also we got a comment from Pengjia. We should keep the event amount for one run lower than 7M, which can avoid the coda generate more than 4 event file for one run. I don't see why it will help but I will let the 3rd person to try to do this to see if this can avoid the problem.