dstat_top_bio_adv.py

### Dstat all I/O process plugin
### Displays all processes' I/O read/write stats and CPU usage
###
### Authority: Guillermo Cantu Luna

class dstat_plugin(dstat):
    def __init__(self):
        self.name = 'most expensive block i/o process'
        self.vars = ('process              pid  read write cpu',)
        self.type = 's'
        self.width = 40
        self.scale = 0
        self.pidset1 = {}

    def check(self):
        if not os.access('/proc/self/io', os.R_OK):
            raise Exception, 'Kernel has no per-process I/O accounting [CONFIG_TASK_IO_ACCOUNTING], use at least 2.6.20'
        return True

    def extract(self):
        self.output = ''
        self.pidset2 = {}
        self.val['usage'] = 0.0
        for pid in proc_pidlist():
            try:
                ### Reset values
                if not self.pidset2.has_key(pid):
                    self.pidset2[pid] = {'read_bytes:': 0, 'write_bytes:': 0, 'cputime:': 0, 'cpuper:': 0}
                if not self.pidset1.has_key(pid):
                    self.pidset1[pid] = {'read_bytes:': 0, 'write_bytes:': 0, 'cputime:': 0, 'cpuper:': 0}

                ### Extract name
                name = proc_splitline('/proc/%s/stat' % pid)[1][1:-1]

                ### Extract counters
                for l in proc_splitlines('/proc/%s/io' % pid):
                    if len(l) != 2: continue
                    self.pidset2[pid][l[0]] = int(l[1])

                ### Get CPU usage
                l = proc_splitline('/proc/%s/stat' % pid)
                if len(l) < 15:
                    cpu_usage = 0
                else:
                    self.pidset2[pid]['cputime:'] = int(l[13]) + int(l[14])
                    cpu_usage = (self.pidset2[pid]['cputime:'] - self.pidset1[pid]['cputime:']) * 1.0 / elapsed / cpunr

            except ValueError:
                continue
            except IOError:
                continue
            except IndexError:
                continue

            read_usage = (self.pidset2[pid]['read_bytes:'] - self.pidset1[pid]['read_bytes:']) * 1.0 / elapsed
            write_usage = (self.pidset2[pid]['write_bytes:'] - self.pidset1[pid]['write_bytes:']) * 1.0 / elapsed
            usage = read_usage + write_usage

            ### Get the process that spends the most jiffies
            if usage > self.val['usage']:
                self.val['usage'] = usage
                self.val['read_usage'] = read_usage
                self.val['write_usage'] = write_usage
                self.val['pid'] = pid
                self.val['name'] = getnamebypid(pid, name)
                self.val['cpu_usage'] = cpu_usage

        if step == op.delay:
            self.pidset1 = self.pidset2

        if self.val['usage'] != 0.0:
            self.output = '%-*s%s%-5s%s%s%s%s%%' % (self.width-14-len(pid), self.val['name'][0:self.width-14-len(pid)], color['darkblue'], self.val['pid'], cprint(self.val['read_usage'], 'd', 5, 1024), cprint(self.val['write_usage'], 'd', 5, 1024), cprint(self.val['cpu_usage'], 'f', 3, 34), color['darkgray'])

    def showcsv(self):
        return 'Top: %s\t%s\t%s\t%s' % (self.val['name'][0:self.width-20], self.val['read_usage'], self.val['write_usage'], self.val['cpu_usage'])