Information networks, such as social networks and that extracted from bibliographic data, are changing dynamically over time. It is crucial to discover time-evolving communities in dynamic networks. In this paper, we study the problem of finding time-evolving communities such that each community freely forms, evolves, and dissolves for any time period. Although the previous t-partite graph based methods are quite effective for discovering such communities from large-scale dynamic networks, they have some weak points such as finding only stable clusters of single path type and not being scalable w.r.t. the time period. We propose CHRONICLE, an efficient clustering algorithm that discovers not only clusters of single path type but also clusters of path group type. In order to find clusters of both types and also control the dynamicity of clusters, CHRONICLE performs the two-stage density-based clustering, which performs the 2nd-stage density-based clustering for the t-partite graph constructed from the 1st-stage density-based clustering result for each timestamp network. For a given data set, CHRONICLE finds all clusters in a fixed time by using a fixed amount of memory, regardless of the number of clusters and the length of clusters. Experimental results using real data sets show that CHRONICLE finds a wider range of clusters in a shorter time with a much smaller amount of memory than the previous method.