Log-based anomaly detection is an essential aspect of maintaining software reliability. However, existing log-based anomaly detection approaches rely on a huge amount of historical labeled data which is inaccessible in many real-world systems. To mitigate this problem, we leverage the features of the abundant historical labeled logs of mature systems to help build anomaly detection models of a new system with very few labels, that is, to generalize the model ability trained from labeled logs of a mature system to achieve anomaly detection on new systems with insufficient data labels. Specifically, we propose MetaLog, a generalizable cross-system anomaly detection approach. MetaLog first incorporates a globally consistent semantic embedding module to obtain log event semantic embedding vectors in a shared global space. Then it leverages meta-learning paradigm to improve the model’s generalization ability. We evaluate MetaLog’s performance on four open log datasets (HDFS, BGL, OpenStack and Thunderbird) from four totally different systems. Results show that MetaLog reaches over 80% F1 score when using only 1% labeled logs of the target system, showing similar performance with state-of-art supervised anomaly detection models trained with 100% labeled data. Besides, it outperforms state-of-art transfer-learning based cross-system anomaly detection models by 20% in the same settings of 1% labeled training logs of the target system.