Connectivity matrix problem using in AgglomerativeClustering on rs-fmri data

I have tried to clustering rs-fmri data by using ward method of nilearn, but i got warning message like:
“the number of connected components of the connectivity matrix is 7 >1”, but it still could finish the parcellation step. However, It is weird that i chose one roi of aal 90 atlas, and could be sure that it is a complete roi checked by afni. It just shown “7 connected components > 1”. I am confused. Is any way to change ‘connect’ strategy just like ‘corner’、‘face’ standard used in fmri cluster method?

Thanks for noticing