#!/usr/bin/env python # This script reports disk stat metrics to ganglia. # # Notes: # This script exposes values in /proc/diskstats and calculates # various statistics based on the Linux kernel 2.6. To find # more information on these values, look in the Linux kernel # documentation for "I/O statistics fields". # # By default, the script would monitor any entry listed under # /proc/diskstats that is not containing a number. Override it by passing # a list of devices in the 'devices' parameter. # # This script has the option of explicitly setting which devices # to check using the "devices" option in your configuration. If # you set this value, it will invalidate the MIN_DISK_SIZE and # IGNORE_DEV options described below. This enables you to # monitor specific partitions instead of the entire device. # Example value: "sda1 sda2". # Example value: "sda sdb sdc". # # This script also checks for a minimum disk size in order to # only measure interesting devices by default. # [Can be overriden if "devices" is set] # # This script looks for disks to check in /proc/partitions while # ignoring any devices present in IGNORE_DEV by default. # [Can be overriden if "devices" is set] # # Example Metrics # * diskstat_XXX_io_time # * diskstat_XXX_read_kbytes_per_sec # * diskstat_XXX_read_time # * diskstat_XXX_reads # * diskstat_XXX_reads_merged # * diskstat_XXX_weighted_io_time # * diskstat_XXX_write_kbytes_per_sec # * diskstat_XXX_write_time # * diskstat_XXX_writes # * diskstat_XXX_writes_merged # Changelog: # v1.0.1 - 2010-07-22 # * Initial version # # v1.0.2 - 2010-08-03 # * Modified reads_per_sec to not calculate per second delta. # This enables us to generate a better graph by stacking # reads/writes with reads/writes merged. # # Copyright Jamie Isaacs. 2010 # License to use, modify, and distribute under the GPL # http://www.gnu.org/licenses/gpl.txt import logging import os import re import stat import time descriptors = [] logging.basicConfig(level=logging.ERROR, format="%(asctime)s - %(name)s - %(levelname)s\t Thread-%(thread)d - %(message)s") logging.debug('starting up') last_update = 0 cur_time = 0 stats = {} last_val = {} MAX_UPDATE_TIME = 15 BYTES_PER_SECTOR = 512 # 5 GB MIN_DISK_SIZE = 5242880 # Set to None to trigger disk discovery under /proc/diskstats # Pass a 'devices' parameter to explicitly list disks to monitor DEVICES = None IGNORE_DEV = 'dm-|loop|drbd' PARTITIONS_FILE = '/proc/partitions' DISKSTATS_FILE = '/proc/diskstats' DMDIR = '/dev/mapper' device_mapper = '' PARTITIONS = [] dmnames = [] dm2pair = {} pair2dm = {} devnames = [] dev2pair = {} pair2dev = {} def build_dmblock_major_minor_tables(): """Returns 1) a table of filenames that are all device mapper block special files 2) a dict mapping each device mapper name to (major,minor) 3) a dict mapping each (major,minor) pair to a table of devce mapper names""" names = [] name2pair = {} pair2name = {} mapper_entries = [] mapper_entries = os.listdir(DMDIR) for n in mapper_entries: s = os.lstat(DMDIR + '/' + n) if stat.S_ISBLK(s[stat.ST_MODE]): names.append(n) maj = str(os.major(s.st_rdev)) min = str(os.minor(s.st_rdev)) name2pair[n] = (maj, min) pair2name[(maj, min)] = n logging.debug('grabbed dmsetup device info') logging.debug('dmsetup devices: ' + str(name2pair)) return (names, name2pair, pair2name) def build_block_major_minor_tables(): """Returns 1) a table of filenames that are all block special files 2) a dict mapping each dev name to (major,minor) 3) a dict mapping each (major,minor) pair to a table of dev names""" dnames = [] d2p = {} p2d = {} # Get values from diskstats file lines = tuple(open(PARTITIONS_FILE, 'r')) logging.debug('grabbed diskstat device info') logging.debug('diskstat devices: ' + '\n'.join(lines)) for line in lines: if line: # read the first three fields from each line (maj, min, name) = line.split()[:3] dnames.append(name) d2p[name] = (maj, min) p2d[(maj, min)] = name return (dnames, d2p, p2d) def get_devname(dev): """Returns device mapper name converted to dev name""" (maj, min) = dm2pair[dev] name = pair2dev[(maj, min)] return name def list_dmnames(): """Returns string of device names associated with device mapper names""" global dmnames global dm2pair global pair2dm global devnames global dev2pair global pair2dev devlist = '' dmnames, dm2pair, pair2dm = build_dmblock_major_minor_tables() logging.debug('dmnames: ' + str(dmnames)) devnames, dev2pair, pair2dev = build_block_major_minor_tables() logging.debug('devnames: ' + str(dmnames)) for d in dmnames: devlist = devlist + ' ' + str(d) logging.debug('devlist: ' + str(devlist)) return devlist def get_partitions(): logging.debug('getting partitions') global PARTITIONS # We need DEVICES != "" because on at least Centos 6 this evaluates to true if DEVICES is not None and DEVICES != "": # Explicit device list has been set logging.debug(' DEVICES has already been set') out = DEVICES else: # Load partitions devices = [] lines = tuple(open(PARTITIONS_FILE, 'r')) for line in lines: # There needs to be a major value or otherwise we skip that line. Skips headers etc. if not re.search("(\w+)?\d+", line): continue device_name = line.split()[3] device_ends_with_number = re.search('\d$', device_name) if 'md' in device_name or not device_ends_with_number: # only include md devices and base block devices devices.append(device_name) out = ' '.join(devices) logging.debug(' result: ' + out) for dev in out.split(): if DEVICES is not None: # Explicit device list has been set PARTITIONS.append(dev) else: # Load disk block size c = open('/sys/block/' + dev + '/size', 'r').read(256) # Make sure device is large enough to collect stats if (int(c) * BYTES_PER_SECTOR / 1024) > MIN_DISK_SIZE: PARTITIONS.append(dev) else: logging.debug(' ignoring ' + dev + ' due to size constraints') return 0 ########################################################################### # This is the order of metrics in /proc/diskstats # 0 major Major number # 1 minor Minor number # 2 blocks Blocks # 3 name Name # 4 reads This is the total number of reads completed successfully. # 5 merge_read Reads and writes which are adjacent to each other may be merged for # efficiency. Thus two 4K reads may become one 8K read before it is # ultimately handed to the disk, and so it will be counted (and queued) # as only one I/O. This field lets you know how often this was done. # 6 s_read This is the total number of sectors read successfully. # 7 ms_read This is the total number of milliseconds spent by all reads. # 8 writes This is the total number of writes completed successfully. # 9 merge_write Reads and writes which are adjacent to each other may be merged for # efficiency. Thus two 4K reads may become one 8K read before it is # ultimately handed to the disk, and so it will be counted (and queued) # as only one I/O. This field lets you know how often this was done. # 10 s_write This is the total number of sectors written successfully. # 11 ms_write This is the total number of milliseconds spent by all writes. # 12 ios The only field that should go to zero. Incremented as requests are # given to appropriate request_queue_t and decremented as they finish. # 13 ms_io This field is increases so long as field 9 is nonzero. # 14 ms_weighted This field is incremented at each I/O start, I/O completion, I/O ########################################################################### def update_stats(): logging.debug('updating stats') global last_update, stats, last_val, cur_time global MAX_UPDATE_TIME cur_time = time.time() if cur_time - last_update < MAX_UPDATE_TIME: logging.debug(' wait ' + str(int(MAX_UPDATE_TIME - (cur_time - last_update))) + ' seconds') return True # Update diskstats stats = {} if not PARTITIONS: part = get_partitions() if part: # Fail if return is non-zero logging.warning('error getting partitions') return False # Get values for each disk device for dev in PARTITIONS: dev = dev.replace('/', '-') logging.debug(" dev: " + dev) # Setup storage lists if dev not in stats: stats[dev] = {} if dev not in last_val: last_val[dev] = {} # Convert from dmname to devname if device_mapper == 'true': olddev = dev dev = get_devname(dev) # Get values from diskstats file lines = tuple(open(DISKSTATS_FILE, 'r')) for line in lines: if dev in line.split(): vals = line.split() logging.debug(' vals: ' + str(vals)) # Reset back to orignal dev name if device_mapper == 'true': dev = olddev get_diff(dev, 'reads', int(vals[3])) get_diff(dev, 'writes', int(vals[7])) get_diff(dev, 'reads_merged', int(vals[4])) get_diff(dev, 'writes_merged', int(vals[8])) get_delta(dev, 'read_bytes_per_sec', int(vals[5]), float(BYTES_PER_SECTOR)) get_delta(dev, 'write_bytes_per_sec', int(vals[9]), float(BYTES_PER_SECTOR)) get_delta(dev, 'read_time', float(vals[6]), 0.001) get_delta(dev, 'write_time', float(vals[10]), 0.001) get_diff(dev, 'io_time', float(vals[12]), 0.001) get_percent_time(dev, 'percent_io_time', float(stats[dev]['io_time'])) get_delta(dev, 'weighted_io_time', float(vals[13]), 0.001) logging.debug('success refreshing stats') logging.debug('stats: ' + str(stats)) logging.debug('last_val: ' + str(last_val)) last_update = cur_time return True def get_delta(dev, key, val, convert=1): logging.debug(' get_delta for ' + dev + '_' + key) global stats, last_val if convert == 0: logging.warning(' convert is zero!') interval = cur_time - last_update if key in last_val[dev] and interval > 0: if val < last_val[dev][key]: logging.debug(' fixing int32 wrap') val += 4294967296 stats[dev][key] = (val - last_val[dev][key]) * float(convert) / float(interval) else: stats[dev][key] = 0 last_val[dev][key] = int(val) def get_percent_time(dev, key, val): logging.debug(' get_percent_time for ' + dev + '_' + key) global stats, last_val interval = cur_time - last_update if interval > 0: stats[dev][key] = (val / interval) * 100 else: stats[dev][key] = 0 def get_diff(dev, key, val, convert=1): logging.debug(' get_diff for ' + dev + '_' + key) global stats, last_val if key in last_val[dev]: stats[dev][key] = (val - last_val[dev][key]) * float(convert) else: stats[dev][key] = 0 # If for some reason we have a negative diff we should assume counters reset # and should set it back to 0 if stats[dev][key] < 0: stats[dev][key] = 0 last_val[dev][key] = val def get_stat(name): logging.debug(' getting stat: ' + name) global stats ret = update_stats() if ret: if name.startswith('diskstat_'): fir = name.find('_') sec = name.find('_', fir + 1) dev = name[fir + 1:sec] label = name[sec + 1:] try: return stats[dev][label] except: logging.warning('failed to fetch [' + dev + '] ' + name) return 0 else: label = name try: return stats[label] except: logging.warning('failed to fetch ' + name) return 0 else: return 0 def metric_init(params): global descriptors, device_mapper global MIN_DISK_SIZE, DEVICES, IGNORE_DEV # Use params.get here to assure function via gmond if params.get('device-mapper') == 'true': devices = list_dmnames() DEVICES = devices IGNORE_DEV = 'loop|drbd' device_mapper = 'true' logging.debug('dm block devices: ' + str(devices)) else: DEVICES = params.get('devices') logging.debug('init: ' + str(params)) time_max = 60 descriptions = dict( reads={ 'units': 'reads', 'description': 'The number of reads completed'}, reads_merged={ 'units': 'reads', 'description': 'The number of reads merged. Reads which are adjacent to each other may be merged for efficiency. Multiple reads may become one before it is handed to the disk, and it will be counted (and queued) as only one I/O.'}, read_bytes_per_sec={ 'units': 'bytes/sec', 'description': 'The number of bytes read per second'}, read_time={ 'units': 's', 'description': 'The time in seconds spent reading'}, writes={ 'units': 'writes', 'description': 'The number of writes completed'}, writes_merged={ 'units': 'writes', 'description': 'The number of writes merged. Writes which are adjacent to each other may be merged for efficiency. Multiple writes may become one before it is handed to the disk, and it will be counted (and queued) as only one I/O.'}, write_bytes_per_sec={ 'units': 'bytes/sec', 'description': 'The number of bytes written per second'}, write_time={ 'units': 's', 'description': 'The time in seconds spent writing'}, io_time={ 'units': 's', 'description': 'The time in seconds spent in I/O operations'}, percent_io_time={ 'units': 'percent', 'value_type': 'float', 'format': '%f', 'description': 'The percent of disk time spent on I/O operations'}, weighted_io_time={ 'units': 's', 'description': 'The weighted time in seconds spend in I/O operations. This measures each I/O start, I/O completion, I/O merge, or read of these stats by the number of I/O operations in progress times the number of seconds spent doing I/O.'} ) update_stats() for label in descriptions: for dev in PARTITIONS: if label in stats[dev]: d = { 'name': 'diskstat_' + dev + '_' + label, 'call_back': get_stat, 'time_max': time_max, 'value_type': 'float', 'units': '', 'slope': 'both', 'format': '%f', 'description': label, 'groups': 'diskstat' } # Apply metric customizations from descriptions d.update(descriptions[label]) descriptors.append(d) else: logging.error("skipped " + label) # For command line testing # time.sleep(MAX_UPDATE_TIME) # update_stats() return descriptors def metric_cleanup(): logging.shutdown() # pass if __name__ == '__main__': from optparse import OptionParser import os logging.debug('running from cmd line') parser = OptionParser() parser.add_option('-d', '--devices', dest='devices', default=None, help='devices to explicitly check') parser.add_option('-b', '--gmetric-bin', dest='gmetric_bin', default='/usr/bin/gmetric', help='path to gmetric binary') parser.add_option('-c', '--gmond-conf', dest='gmond_conf', default='/etc/ganglia/gmond.conf', help='path to gmond.conf') parser.add_option('-g', '--gmetric', dest='gmetric', action='store_true', default=False, help='submit via gmetric') parser.add_option('-m', '--device-mapper', dest='device_mapper', action='store_true', default=False, help='utilize all device mapper devices if set') parser.add_option('-q', '--quiet', dest='quiet', action='store_true', default=False) (options, args) = parser.parse_args() if options.device_mapper: metric_init({ 'device-mapper': 'true', }) else: metric_init({ 'devices': options.devices, }) while True: for d in descriptors: v = d['call_back'](d['name']) if not options.quiet: print ' %s: %s %s [%s]' % (d['name'], v, d['units'], d['description']) if options.gmetric: if d['value_type'] == 'uint': value_type = 'uint32' else: value_type = d['value_type'] cmd = "%s --conf=%s --value='%s' --units='%s' --type='%s' --name='%s' --slope='%s'" % \ (options.gmetric_bin, options.gmond_conf, v, d['units'], value_type, d['name'], d['slope']) os.system(cmd) print 'Sleeping 15 seconds' time.sleep(15)