Currently, the HProfFilter is invoked K times to partition K of the N dumps into K files. Instead the filter should support a single pass fan-out of each dump. The simplest way is to accept a sparse map of index to file. However, this will require rearchitecting most of the parser either to read all the state in and then multiplex it out or to have multiple writers open at the same time. This should be considered in conjunction with any work from #4.
Currently, the HProfFilter is invoked K times to partition K of the N dumps into K files. Instead the filter should support a single pass fan-out of each dump. The simplest way is to accept a sparse map of index to file. However, this will require rearchitecting most of the parser either to read all the state in and then multiplex it out or to have multiple writers open at the same time. This should be considered in conjunction with any work from #4.