Log parsing has been long studied by researchers, given its high importance in the software engineering community: the process identifies dynamic variables and constructs log templates with static components. Prior work has proposed many statistic-based log parsers (e.g., Drain), which are highly efficient; they, unfortunately, met the bottleneck of parsing performance in comparison to semantic-based log parsers, which require labeling and more computational resources. In the meanwhile, we noticed that previous works on log parsing mainly focused on the parsing stage and usually used an ad hoc preprocessing step (e.g., masking numbers or IP addresses). However, we argue that both preprocessing and parsing are essential for log parsers to identify dynamic variables. The lack of understanding of log preprocessing may prevent the optimal use of log parsers and hinder future research in developing parsing algorithms and configuring their preprocessing step. Therefore, our work first studied the existing approaches for log preprocessing, in particular, by analyzing the existing preprocessing steps used for different log datasets provided in Loghub, a popular log parsing benchmark. We then developed preprocessing framework based on our findings and evaluated its impact on log parsing. According to our experiment, our preprocessing framework can significantly boost the overall performance of the four state-of- the-art statistic-based parsers examined in the study. The best statistic-based log parser, Drain, obtained improvement on all four parsing metrics (e.g., the F1 score of template accuracy, FTA, increases by 108.9%). Moreover, in comparison to the optimal semantic-based log parsers, it obtained a 28.3% improvement in grouping accuracy (GA), 38.1% enhancement on the F1 score of grouping accuracy (FGA), and an 18.6% increment on the FTA. Our work pioneered studying the process of log preprocessing and provided a generalizable framework to enhance the state-of- the-art of log parsing.