Rolling mean partinioned by different collumns in R

I'm working in R and I need to do a rolling mean of the last 3 values of a data set, partinioned by 3 different collumns. Each time one of the values of the collumn changes, the rolling mean should restart the rolling. The following example portrays my objective:

Material   Channel   client   value   mean   
    1         2        2        5      NA
    1         2        2        10     NA
    1         2        2        7      7,3
    1         2        2        10     9
    1         2        5        5      NA
    1         2        5        3      NA
    1         2        5        5      13

I've tried to use the rollmean function, but it doesn't partinion the dataset by collumns.

input_mes_cluster[, peso_medio:=rollmean(peso, k = 12) , by= c("cluster","material","canal", "cliente", "regiao")]

How can I get the expected result?

Thanks!