ferret1to2.py

# Script to convert procwatch/ferret v1 to libcollector-based v2

import sys
import os

# rewritten files to a subdirectory to avoid name conflicts
#
outDir = './ferret2'

# v2 requires these fields
#
statFields = "pid comm ppid utime stime cutime cstime num_threads starttime processor"

if __name__ == "__main__":

    traceNames = sys.argv[1:]

    if not os.path.exists(outDir):
        os.mkdir(outDir)

    for traceName in traceNames:

        outFile = os.path.join(outDir, traceName)

        with open(traceName, 'r') as traceFile:

            with open(outFile, 'w') as newTraceFile:

                header = list()
                header.append("# Converted by ferret1to2.py")
                header.append("I _SC_CLK_TCK 100")
                header.append("I Status %s" % statFields)
                header.append("")

                header = "\n".join(header)
                newTraceFile.write(header)

                cpuList = None

                for row in traceFile:
                    records = row.split(',')

                    if len(records) < 3:
                        # v1 had a habit of writing partial sample rows
                        # when interrupted.
                        #
                        continue

                    assert len(records) > 1, row

                    # v2 timestamps are integer microseconds, not seconds with microsecond
                    # resolution
                    #
                    now = int(float(records[0]) * 1.0e6)
                    status = list()
                    cpuFreq = list()

                    for r in records[1:]:

                        if r[0] == 's':

                            state = r.split()[1:]
                            del state[2]  # delete "state" field from stat
                            status.append(" ".join(state))

                        elif r[0] == 'f':

                            raw = r.split()[1:]
                            cpus = raw[0:-1:2]
                            freqs = raw[1::2]

                            # cpus are labelled cpu[0-9]+ so just extract the digit(s)
                            #
                            cpus = [c[3:] for c in cpus]

                            if not cpuList:
                                # use the first CPU DVFS record to declare the CPU set being
                                # sampled
                                #
                                cpuList = cpus[:]
                                newTraceFile.write("I CPUList %s\n" % " ".join(cpus))

                            # and put it all back together again
                            #
                            cpuFreq.append(" ".join(sum(zip(cpus, freqs), ())))

                        else:
                            assert False, ("Unrecognised record", r[0], row)

                    assert len(cpuFreq) == 1, (row, cpuFreq)

                    newTraceFile.write("T %s\n" % now)
                    newTraceFile.write("F %s\n" % cpuFreq[0])
                    for s in status:
                        newTraceFile.write("S %s\n" % s)