mirror_zfs/cmd/arc_summary

1073 lines
38 KiB
Plaintext
Raw Permalink Normal View History

#!/usr/bin/env python3
#
# Copyright (c) 2008 Ben Rockwood <benr@cuddletech.com>,
# Copyright (c) 2010 Martin Matuska <mm@FreeBSD.org>,
# Copyright (c) 2010-2011 Jason J. Hellenthal <jhell@DataIX.net>,
# Copyright (c) 2017 Scot W. Stevenson <scot.stevenson@gmail.com>
# All rights reserved.
#
# Redistribution and use in source and binary forms, with or without
# modification, are permitted provided that the following conditions
# are met:
#
# 1. Redistributions of source code must retain the above copyright
# notice, this list of conditions and the following disclaimer.
# 2. Redistributions in binary form must reproduce the above copyright
# notice, this list of conditions and the following disclaimer in the
# documentation and/or other materials provided with the distribution.
#
# THIS SOFTWARE IS PROVIDED BY AUTHOR AND CONTRIBUTORS ``AS IS'' AND
# ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
# IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
# ARE DISCLAIMED. IN NO EVENT SHALL AUTHOR OR CONTRIBUTORS BE LIABLE
# FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
# DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
# OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
# HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
# LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
# OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
# SUCH DAMAGE.
"""Print statistics on the ZFS ARC Cache and other information
Provides basic information on the ARC, its efficiency, the L2ARC (if present),
the Data Management Unit (DMU), Virtual Devices (VDEVs), and tunables. See
the in-source documentation and code at
https://github.com/openzfs/zfs/blob/master/module/zfs/arc.c for details.
The original introduction to arc_summary can be found at
http://cuddletech.com/?p=454
"""
import argparse
import os
import subprocess
import sys
import time
import errno
# We can't use env -S portably, and we need python3 -u to handle pipes in
# the shell abruptly closing the way we want to, so...
import io
if isinstance(sys.__stderr__.buffer, io.BufferedWriter):
os.execv(sys.executable, [sys.executable, "-u"] + sys.argv)
DESCRIPTION = 'Print ARC and other statistics for OpenZFS'
INDENT = ' '*8
LINE_LENGTH = 72
DATE_FORMAT = '%a %b %d %H:%M:%S %Y'
TITLE = 'ZFS Subsystem Report'
SECTIONS = 'arc archits dmu l2arc spl tunables vdev zil'.split()
SECTION_HELP = 'print info from one section ('+' '.join(SECTIONS)+')'
# Tunables and SPL are handled separately because they come from
# different sources
SECTION_PATHS = {'arc': 'arcstats',
'dmu': 'dmu_tx',
'l2arc': 'arcstats', # L2ARC stuff lives in arcstats
'zfetch': 'zfetchstats',
'zil': 'zil'}
parser = argparse.ArgumentParser(description=DESCRIPTION)
parser.add_argument('-a', '--alternate', action='store_true', default=False,
help='use alternate formatting for tunables and SPL',
dest='alt')
parser.add_argument('-d', '--description', action='store_true', default=False,
help='print descriptions with tunables and SPL',
dest='desc')
parser.add_argument('-g', '--graph', action='store_true', default=False,
help='print graph on ARC use and exit', dest='graph')
parser.add_argument('-p', '--page', type=int, dest='page',
help='print page by number (DEPRECATED, use "-s")')
parser.add_argument('-r', '--raw', action='store_true', default=False,
help='dump all available data with minimal formatting',
dest='raw')
parser.add_argument('-s', '--section', dest='section', help=SECTION_HELP)
ARGS = parser.parse_args()
if sys.platform.startswith('freebsd'):
# Requires py36-sysctl on FreeBSD
import sysctl
def is_value(ctl):
return ctl.type != sysctl.CTLTYPE_NODE
def namefmt(ctl, base='vfs.zfs.'):
# base is removed from the name
cut = len(base)
return ctl.name[cut:]
def load_kstats(section):
base = 'kstat.zfs.misc.{section}.'.format(section=section)
fmt = lambda kstat: '{name} : {value}'.format(name=namefmt(kstat, base),
value=kstat.value)
kstats = sysctl.filter(base)
return [fmt(kstat) for kstat in kstats if is_value(kstat)]
def get_params(base):
ctls = sysctl.filter(base)
return {namefmt(ctl): str(ctl.value) for ctl in ctls if is_value(ctl)}
def get_tunable_params():
return get_params('vfs.zfs')
def get_vdev_params():
return get_params('vfs.zfs.vdev')
def get_version_impl(request):
# FreeBSD reports versions for zpl and spa instead of zfs and spl.
name = {'zfs': 'zpl',
'spl': 'spa'}[request]
mib = 'vfs.zfs.version.{}'.format(name)
version = sysctl.filter(mib)[0].value
return '{} version {}'.format(name, version)
def get_descriptions(_request):
ctls = sysctl.filter('vfs.zfs')
return {namefmt(ctl): ctl.description for ctl in ctls if is_value(ctl)}
elif sys.platform.startswith('linux'):
KSTAT_PATH = '/proc/spl/kstat/zfs'
SPL_PATH = '/sys/module/spl/parameters'
TUNABLES_PATH = '/sys/module/zfs/parameters'
def load_kstats(section):
path = os.path.join(KSTAT_PATH, section)
with open(path) as f:
return list(f)[2:] # Get rid of header
def get_params(basepath):
"""Collect information on the Solaris Porting Layer (SPL) or the
tunables, depending on the PATH given. Does not check if PATH is
legal.
"""
result = {}
for name in os.listdir(basepath):
path = os.path.join(basepath, name)
with open(path) as f:
value = f.read()
result[name] = value.strip()
return result
def get_spl_params():
return get_params(SPL_PATH)
def get_tunable_params():
return get_params(TUNABLES_PATH)
def get_vdev_params():
return get_params(TUNABLES_PATH)
def get_version_impl(request):
# The original arc_summary called /sbin/modinfo/{spl,zfs} to get
# the version information. We switch to /sys/module/{spl,zfs}/version
# to make sure we get what is really loaded in the kernel
try:
with open("/sys/module/{}/version".format(request)) as f:
return f.read().strip()
except:
return "(unknown)"
def get_descriptions(request):
"""Get the descriptions of the Solaris Porting Layer (SPL) or the
tunables, return with minimal formatting.
"""
if request not in ('spl', 'zfs'):
print('ERROR: description of "{0}" requested)'.format(request))
sys.exit(1)
descs = {}
target_prefix = 'parm:'
# We would prefer to do this with /sys/modules -- see the discussion at
# get_version() -- but there isn't a way to get the descriptions from
# there, so we fall back on modinfo
command = ["/sbin/modinfo", request, "-0"]
info = ''
try:
info = subprocess.run(command, stdout=subprocess.PIPE,
check=True, universal_newlines=True)
raw_output = info.stdout.split('\0')
except subprocess.CalledProcessError:
print("Error: Descriptions not available",
"(can't access kernel module)")
sys.exit(1)
for line in raw_output:
if not line.startswith(target_prefix):
continue
line = line[len(target_prefix):].strip()
name, raw_desc = line.split(':', 1)
desc = raw_desc.rsplit('(', 1)[0]
if desc == '':
desc = '(No description found)'
descs[name.strip()] = desc.strip()
return descs
def handle_unraisableException(exc_type, exc_value=None, exc_traceback=None,
err_msg=None, object=None):
handle_Exception(exc_type, object, exc_traceback)
def handle_Exception(ex_cls, ex, tb):
if ex_cls is KeyboardInterrupt:
sys.exit()
if ex_cls is BrokenPipeError:
# It turns out that while sys.exit() triggers an exception
# not handled message on Python 3.8+, os._exit() does not.
os._exit(0)
if ex_cls is OSError:
if ex.errno == errno.ENOTCONN:
sys.exit()
raise ex
if hasattr(sys,'unraisablehook'): # Python 3.8+
sys.unraisablehook = handle_unraisableException
sys.excepthook = handle_Exception
def cleanup_line(single_line):
"""Format a raw line of data from /proc and isolate the name value
part, returning a tuple with each. Currently, this gets rid of the
middle '4'. For example "arc_no_grow 4 0" returns the tuple
("arc_no_grow", "0").
"""
name, _, value = single_line.split()
return name, value
def draw_graph(kstats_dict):
"""Draw a primitive graph representing the basic information on the
ARC -- its size and the proportion used by MFU and MRU -- and quit.
We use max size of the ARC to calculate how full it is. This is a
very rough representation.
"""
arc_stats = isolate_section('arcstats', kstats_dict)
GRAPH_INDENT = ' '*4
GRAPH_WIDTH = 70
arc_max = int(arc_stats['c_max'])
arc_size = f_bytes(arc_stats['size'])
arc_perc = f_perc(arc_stats['size'], arc_max)
data_size = f_bytes(arc_stats['data_size'])
meta_size = f_bytes(arc_stats['metadata_size'])
dnode_size = f_bytes(arc_stats['dnode_size'])
info_form = ('ARC: {0} ({1}) Data: {2} Meta: {3} Dnode: {4}')
info_line = info_form.format(arc_size, arc_perc, data_size, meta_size,
dnode_size)
info_spc = ' '*int((GRAPH_WIDTH-len(info_line))/2)
info_line = GRAPH_INDENT+info_spc+info_line
graph_line = GRAPH_INDENT+'+'+('-'*(GRAPH_WIDTH-2))+'+'
arc_perc = float(int(arc_stats['size'])/arc_max)
data_perc = float(int(arc_stats['data_size'])/arc_max)
meta_perc = float(int(arc_stats['metadata_size'])/arc_max)
dnode_perc = float(int(arc_stats['dnode_size'])/arc_max)
total_ticks = float(arc_perc)*GRAPH_WIDTH
data_ticks = data_perc*GRAPH_WIDTH
meta_ticks = meta_perc*GRAPH_WIDTH
dnode_ticks = dnode_perc*GRAPH_WIDTH
other_ticks = total_ticks-(data_ticks+meta_ticks+dnode_ticks)
core_form = 'D'*int(data_ticks)+'M'*int(meta_ticks)+'N'*int(dnode_ticks)+\
'O'*int(other_ticks)
core_spc = ' '*(GRAPH_WIDTH-(2+len(core_form)))
core_line = GRAPH_INDENT+'|'+core_form+core_spc+'|'
for line in ('', info_line, graph_line, core_line, graph_line, ''):
print(line)
def f_bytes(byte_string):
"""Return human-readable representation of a byte value in
powers of 2 (eg "KiB" for "kibibytes", etc) to two decimal
points. Values smaller than one KiB are returned without
decimal points. Note "bytes" is a reserved keyword.
"""
prefixes = ([2**80, "YiB"], # yobibytes (yotta)
[2**70, "ZiB"], # zebibytes (zetta)
[2**60, "EiB"], # exbibytes (exa)
[2**50, "PiB"], # pebibytes (peta)
[2**40, "TiB"], # tebibytes (tera)
[2**30, "GiB"], # gibibytes (giga)
[2**20, "MiB"], # mebibytes (mega)
[2**10, "KiB"]) # kibibytes (kilo)
bites = int(byte_string)
if bites >= 2**10:
for limit, unit in prefixes:
if bites >= limit:
value = bites / limit
break
result = '{0:.1f} {1}'.format(value, unit)
else:
result = '{0} Bytes'.format(bites)
return result
def f_hits(hits_string):
"""Create a human-readable representation of the number of hits.
The single-letter symbols used are SI to avoid the confusion caused
by the different "short scale" and "long scale" representations in
English, which use the same words for different values. See
https://en.wikipedia.org/wiki/Names_of_large_numbers and:
https://physics.nist.gov/cuu/Units/prefixes.html
"""
numbers = ([10**24, 'Y'], # yotta (septillion)
[10**21, 'Z'], # zetta (sextillion)
[10**18, 'E'], # exa (quintrillion)
[10**15, 'P'], # peta (quadrillion)
[10**12, 'T'], # tera (trillion)
[10**9, 'G'], # giga (billion)
[10**6, 'M'], # mega (million)
[10**3, 'k']) # kilo (thousand)
hits = int(hits_string)
if hits >= 1000:
for limit, symbol in numbers:
if hits >= limit:
value = hits/limit
break
result = "%0.1f%s" % (value, symbol)
else:
result = "%d" % hits
return result
def f_perc(value1, value2):
"""Calculate percentage and return in human-readable form. If
rounding produces the result '0.0' though the first number is
not zero, include a 'less-than' symbol to avoid confusion.
Division by zero is handled by returning 'n/a'; no error
is called.
"""
v1 = float(value1)
v2 = float(value2)
try:
perc = 100 * v1/v2
except ZeroDivisionError:
result = 'n/a'
else:
result = '{0:0.1f} %'.format(perc)
if result == '0.0 %' and v1 > 0:
result = '< 0.1 %'
return result
def format_raw_line(name, value):
"""For the --raw option for the tunable and SPL outputs, decide on the
correct formatting based on the --alternate flag.
"""
if ARGS.alt:
result = '{0}{1}={2}'.format(INDENT, name, value)
else:
# Right-align the value within the line length if it fits,
# otherwise just separate it from the name by a single space.
fit = LINE_LENGTH - len(INDENT) - len(name)
overflow = len(value) + 1
w = max(fit, overflow)
result = '{0}{1}{2:>{w}}'.format(INDENT, name, value, w=w)
return result
def get_kstats():
"""Collect information on the ZFS subsystem. The step does not perform any
further processing, giving us the option to only work on what is actually
needed. The name "kstat" is a holdover from the Solaris utility of the same
name.
"""
result = {}
for section in SECTION_PATHS.values():
if section not in result:
result[section] = load_kstats(section)
return result
def get_version(request):
"""Get the version number of ZFS or SPL on this machine for header.
Returns an error string, but does not raise an error, if we can't
get the ZFS/SPL version.
"""
if request not in ('spl', 'zfs'):
error_msg = '(ERROR: "{0}" requested)'.format(request)
return error_msg
return get_version_impl(request)
def print_header():
"""Print the initial heading with date and time as well as info on the
kernel and ZFS versions. This is not called for the graph.
"""
# datetime is now recommended over time but we keep the exact formatting
pyzfs: python3 support (build system) Almost all of the Python code in the respository has been updated to be compatibile with Python 2.6, Python 3.4, or newer. The only exceptions are arc_summery3.py which requires Python 3, and pyzfs which requires at least Python 2.7. This allows us to maintain a single version of the code and support most default versions of python. This change does the following: * Sets the default shebang for all Python scripts to python3. If only Python 2 is available, then at install time scripts which are compatible with Python 2 will have their shebangs replaced with /usr/bin/python. This is done for compatibility until Python 2 goes end of life. Since only the installed versions are changed this means Python 3 must be installed on the system for test-runner when testing in-tree. * Added --with-python=<2|3|3.4,etc> configure option which sets the PYTHON environment variable to target a specific python version. By default the newest installed version of Python will be used or the preferred distribution version when creating pacakges. * Fixed --enable-pyzfs configure checks so they are run when --enable-pyzfs=check and --enable-pyzfs=yes. * Enabled pyzfs for Python 3.4 and newer, which is now supported. * Renamed pyzfs package to python<VERSION>-pyzfs and updated to install in the appropriate site location. For example, when building with --with-python=3.4 a python34-pyzfs will be created which installs in /usr/lib/python3.4/site-packages/. * Renamed the following python scripts according to the Fedora guidance for packaging utilities in /bin - dbufstat.py -> dbufstat - arcstat.py -> arcstat - arc_summary.py -> arc_summary - arc_summary3.py -> arc_summary3 * Updated python-cffi package name. On CentOS 6, CentOS 7, and Amazon Linux it's called python-cffi, not python2-cffi. For Python3 it's called python3-cffi or python3x-cffi. * Install one version of arc_summary. Depending on the version of Python available install either arc_summary2 or arc_summary3 as arc_summary. The user output is only slightly different. Reviewed-by: John Ramsden <johnramsden@riseup.net> Reviewed-by: Neal Gompa <ngompa@datto.com> Reviewed-by: loli10K <ezomori.nozomu@gmail.com> Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Closes #8096
2018-10-31 19:22:59 +03:00
# from the older version of arc_summary in case there are scripts
# that expect it in this way
daydate = time.strftime(DATE_FORMAT)
spc_date = LINE_LENGTH-len(daydate)
sys_version = os.uname()
sys_msg = sys_version.sysname+' '+sys_version.release
zfs = get_version('zfs')
spc_zfs = LINE_LENGTH-len(zfs)
machine_msg = 'Machine: '+sys_version.nodename+' ('+sys_version.machine+')'
spl = get_version('spl')
spc_spl = LINE_LENGTH-len(spl)
print('\n'+('-'*LINE_LENGTH))
print('{0:<{spc}}{1}'.format(TITLE, daydate, spc=spc_date))
print('{0:<{spc}}{1}'.format(sys_msg, zfs, spc=spc_zfs))
print('{0:<{spc}}{1}\n'.format(machine_msg, spl, spc=spc_spl))
def print_raw(kstats_dict):
"""Print all available data from the system in a minimally sorted format.
This can be used as a source to be piped through 'grep'.
"""
sections = sorted(kstats_dict.keys())
for section in sections:
print('\n{0}:'.format(section.upper()))
lines = sorted(kstats_dict[section])
for line in lines:
name, value = cleanup_line(line)
print(format_raw_line(name, value))
# Tunables and SPL must be handled separately because they come from a
# different source and have descriptions the user might request
print()
section_spl()
section_tunables()
def isolate_section(section_name, kstats_dict):
"""From the complete information on all sections, retrieve only those
for one section.
"""
try:
section_data = kstats_dict[section_name]
except KeyError:
print('ERROR: Data on {0} not available'.format(section_data))
sys.exit(1)
section_dict = dict(cleanup_line(l) for l in section_data)
return section_dict
# Formatted output helper functions
def prt_1(text, value):
"""Print text and one value, no indent"""
spc = ' '*(LINE_LENGTH-(len(text)+len(value)))
print('{0}{spc}{1}'.format(text, value, spc=spc))
def prt_i1(text, value):
"""Print text and one value, with indent"""
spc = ' '*(LINE_LENGTH-(len(INDENT)+len(text)+len(value)))
print(INDENT+'{0}{spc}{1}'.format(text, value, spc=spc))
def prt_2(text, value1, value2):
"""Print text and two values, no indent"""
values = '{0:>9} {1:>9}'.format(value1, value2)
spc = ' '*(LINE_LENGTH-(len(text)+len(values)+2))
print('{0}{spc} {1}'.format(text, values, spc=spc))
def prt_i2(text, value1, value2):
"""Print text and two values, with indent"""
values = '{0:>9} {1:>9}'.format(value1, value2)
spc = ' '*(LINE_LENGTH-(len(INDENT)+len(text)+len(values)+2))
print(INDENT+'{0}{spc} {1}'.format(text, values, spc=spc))
# The section output concentrates on important parameters instead of
# being exhaustive (that is what the --raw parameter is for)
def section_arc(kstats_dict):
"""Give basic information on the ARC, MRU and MFU. This is the first
and most used section.
"""
arc_stats = isolate_section('arcstats', kstats_dict)
memory_all = arc_stats['memory_all_bytes']
memory_free = arc_stats['memory_free_bytes']
memory_avail = arc_stats['memory_available_bytes']
arc_size = arc_stats['size']
arc_target_size = arc_stats['c']
arc_max = arc_stats['c_max']
arc_min = arc_stats['c_min']
dnode_limit = arc_stats['arc_dnode_limit']
print('ARC status:')
prt_i1('Total memory size:', f_bytes(memory_all))
prt_i2('Min target size:', f_perc(arc_min, memory_all), f_bytes(arc_min))
prt_i2('Max target size:', f_perc(arc_max, memory_all), f_bytes(arc_max))
prt_i2('Target size (adaptive):',
f_perc(arc_size, arc_max), f_bytes(arc_target_size))
prt_i2('Current size:', f_perc(arc_size, arc_max), f_bytes(arc_size))
prt_i1('Free memory size:', f_bytes(memory_free))
prt_i1('Available memory size:', f_bytes(memory_avail))
print()
compressed_size = arc_stats['compressed_size']
overhead_size = arc_stats['overhead_size']
bonus_size = arc_stats['bonus_size']
dnode_size = arc_stats['dnode_size']
dbuf_size = arc_stats['dbuf_size']
hdr_size = arc_stats['hdr_size']
l2_hdr_size = arc_stats['l2_hdr_size']
abd_chunk_waste_size = arc_stats['abd_chunk_waste_size']
prt_1('ARC structural breakdown (current size):', f_bytes(arc_size))
prt_i2('Compressed size:',
f_perc(compressed_size, arc_size), f_bytes(compressed_size))
prt_i2('Overhead size:',
f_perc(overhead_size, arc_size), f_bytes(overhead_size))
prt_i2('Bonus size:',
f_perc(bonus_size, arc_size), f_bytes(bonus_size))
prt_i2('Dnode size:',
f_perc(dnode_size, arc_size), f_bytes(dnode_size))
prt_i2('Dbuf size:',
f_perc(dbuf_size, arc_size), f_bytes(dbuf_size))
prt_i2('Header size:',
f_perc(hdr_size, arc_size), f_bytes(hdr_size))
prt_i2('L2 header size:',
f_perc(l2_hdr_size, arc_size), f_bytes(l2_hdr_size))
prt_i2('ABD chunk waste size:',
f_perc(abd_chunk_waste_size, arc_size), f_bytes(abd_chunk_waste_size))
print()
More adaptive ARC eviction Traditionally ARC adaptation was limited to MRU/MFU distribution. But for years people with metadata-centric workload demanded mechanisms to also manage data/metadata distribution, that in original ZFS was just a FIFO. As result ZFS effectively got separate states for data and metadata, minimum and maximum metadata limits etc, but it all required manual tuning, was not adaptive and in its heart remained a bad FIFO. This change removes most of existing eviction logic, rewriting it from scratch. This makes MRU/MFU adaptation individual for data and meta- data, same as the distribution between data and metadata themselves. Since most of required states separation was already done, it only required to make arcs_size state field specific per data/metadata. The adaptation logic is still based on previous concept of ghost hits, just now it balances ARC capacity between 4 states: MRU data, MRU metadata, MFU data and MFU metadata. To simplify arc_c changes instead of arc_p measured in bytes, this code uses 3 variable arc_meta, arc_pd and arc_pm, representing ARC balance between metadata and data, MRU and MFU for data, and MRU and MFU for metadata respectively as 32-bit fixed point fractions. Since we care about the math result only when need to evict, this moves all the logic from arc_adapt() to arc_evict(), that reduces per-block overhead, since per-block operations are limited to stats collection, now moved from arc_adapt() to arc_access() and using cheaper wmsums. This also allows to remove ugly ARC_HDR_DO_ADAPT flag from many places. This change also removes number of metadata specific tunables, part of which were actually not functioning correctly, since not all metadata are equal and some (like L2ARC headers) are not really evictable. Instead it introduced single opaque knob zfs_arc_meta_balance, tuning ARC's reaction on ghost hits, allowing administrator give more or less preference to metadata without setting strict limits. Some of old code parts like arc_evict_meta() are just removed, because since introduction of ABD ARC they really make no sense: only headers referenced by small number of buffers are not evictable, and they are really not evictable no matter what this code do. Instead just call arc_prune_async() if too much metadata appear not evictable. Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Reviewed-by: Allan Jude <allan@klarasystems.com> Signed-off-by: Alexander Motin <mav@FreeBSD.org> Sponsored by: iXsystems, Inc. Closes #14359
2023-03-08 22:17:23 +03:00
meta = arc_stats['meta']
pd = arc_stats['pd']
pm = arc_stats['pm']
data_size = arc_stats['data_size']
metadata_size = arc_stats['metadata_size']
More adaptive ARC eviction Traditionally ARC adaptation was limited to MRU/MFU distribution. But for years people with metadata-centric workload demanded mechanisms to also manage data/metadata distribution, that in original ZFS was just a FIFO. As result ZFS effectively got separate states for data and metadata, minimum and maximum metadata limits etc, but it all required manual tuning, was not adaptive and in its heart remained a bad FIFO. This change removes most of existing eviction logic, rewriting it from scratch. This makes MRU/MFU adaptation individual for data and meta- data, same as the distribution between data and metadata themselves. Since most of required states separation was already done, it only required to make arcs_size state field specific per data/metadata. The adaptation logic is still based on previous concept of ghost hits, just now it balances ARC capacity between 4 states: MRU data, MRU metadata, MFU data and MFU metadata. To simplify arc_c changes instead of arc_p measured in bytes, this code uses 3 variable arc_meta, arc_pd and arc_pm, representing ARC balance between metadata and data, MRU and MFU for data, and MRU and MFU for metadata respectively as 32-bit fixed point fractions. Since we care about the math result only when need to evict, this moves all the logic from arc_adapt() to arc_evict(), that reduces per-block overhead, since per-block operations are limited to stats collection, now moved from arc_adapt() to arc_access() and using cheaper wmsums. This also allows to remove ugly ARC_HDR_DO_ADAPT flag from many places. This change also removes number of metadata specific tunables, part of which were actually not functioning correctly, since not all metadata are equal and some (like L2ARC headers) are not really evictable. Instead it introduced single opaque knob zfs_arc_meta_balance, tuning ARC's reaction on ghost hits, allowing administrator give more or less preference to metadata without setting strict limits. Some of old code parts like arc_evict_meta() are just removed, because since introduction of ABD ARC they really make no sense: only headers referenced by small number of buffers are not evictable, and they are really not evictable no matter what this code do. Instead just call arc_prune_async() if too much metadata appear not evictable. Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Reviewed-by: Allan Jude <allan@klarasystems.com> Signed-off-by: Alexander Motin <mav@FreeBSD.org> Sponsored by: iXsystems, Inc. Closes #14359
2023-03-08 22:17:23 +03:00
anon_data = arc_stats['anon_data']
anon_metadata = arc_stats['anon_metadata']
mfu_data = arc_stats['mfu_data']
mfu_metadata = arc_stats['mfu_metadata']
mfu_edata = arc_stats['mfu_evictable_data']
mfu_emetadata = arc_stats['mfu_evictable_metadata']
More adaptive ARC eviction Traditionally ARC adaptation was limited to MRU/MFU distribution. But for years people with metadata-centric workload demanded mechanisms to also manage data/metadata distribution, that in original ZFS was just a FIFO. As result ZFS effectively got separate states for data and metadata, minimum and maximum metadata limits etc, but it all required manual tuning, was not adaptive and in its heart remained a bad FIFO. This change removes most of existing eviction logic, rewriting it from scratch. This makes MRU/MFU adaptation individual for data and meta- data, same as the distribution between data and metadata themselves. Since most of required states separation was already done, it only required to make arcs_size state field specific per data/metadata. The adaptation logic is still based on previous concept of ghost hits, just now it balances ARC capacity between 4 states: MRU data, MRU metadata, MFU data and MFU metadata. To simplify arc_c changes instead of arc_p measured in bytes, this code uses 3 variable arc_meta, arc_pd and arc_pm, representing ARC balance between metadata and data, MRU and MFU for data, and MRU and MFU for metadata respectively as 32-bit fixed point fractions. Since we care about the math result only when need to evict, this moves all the logic from arc_adapt() to arc_evict(), that reduces per-block overhead, since per-block operations are limited to stats collection, now moved from arc_adapt() to arc_access() and using cheaper wmsums. This also allows to remove ugly ARC_HDR_DO_ADAPT flag from many places. This change also removes number of metadata specific tunables, part of which were actually not functioning correctly, since not all metadata are equal and some (like L2ARC headers) are not really evictable. Instead it introduced single opaque knob zfs_arc_meta_balance, tuning ARC's reaction on ghost hits, allowing administrator give more or less preference to metadata without setting strict limits. Some of old code parts like arc_evict_meta() are just removed, because since introduction of ABD ARC they really make no sense: only headers referenced by small number of buffers are not evictable, and they are really not evictable no matter what this code do. Instead just call arc_prune_async() if too much metadata appear not evictable. Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Reviewed-by: Allan Jude <allan@klarasystems.com> Signed-off-by: Alexander Motin <mav@FreeBSD.org> Sponsored by: iXsystems, Inc. Closes #14359
2023-03-08 22:17:23 +03:00
mru_data = arc_stats['mru_data']
mru_metadata = arc_stats['mru_metadata']
mru_edata = arc_stats['mru_evictable_data']
mru_emetadata = arc_stats['mru_evictable_metadata']
More adaptive ARC eviction Traditionally ARC adaptation was limited to MRU/MFU distribution. But for years people with metadata-centric workload demanded mechanisms to also manage data/metadata distribution, that in original ZFS was just a FIFO. As result ZFS effectively got separate states for data and metadata, minimum and maximum metadata limits etc, but it all required manual tuning, was not adaptive and in its heart remained a bad FIFO. This change removes most of existing eviction logic, rewriting it from scratch. This makes MRU/MFU adaptation individual for data and meta- data, same as the distribution between data and metadata themselves. Since most of required states separation was already done, it only required to make arcs_size state field specific per data/metadata. The adaptation logic is still based on previous concept of ghost hits, just now it balances ARC capacity between 4 states: MRU data, MRU metadata, MFU data and MFU metadata. To simplify arc_c changes instead of arc_p measured in bytes, this code uses 3 variable arc_meta, arc_pd and arc_pm, representing ARC balance between metadata and data, MRU and MFU for data, and MRU and MFU for metadata respectively as 32-bit fixed point fractions. Since we care about the math result only when need to evict, this moves all the logic from arc_adapt() to arc_evict(), that reduces per-block overhead, since per-block operations are limited to stats collection, now moved from arc_adapt() to arc_access() and using cheaper wmsums. This also allows to remove ugly ARC_HDR_DO_ADAPT flag from many places. This change also removes number of metadata specific tunables, part of which were actually not functioning correctly, since not all metadata are equal and some (like L2ARC headers) are not really evictable. Instead it introduced single opaque knob zfs_arc_meta_balance, tuning ARC's reaction on ghost hits, allowing administrator give more or less preference to metadata without setting strict limits. Some of old code parts like arc_evict_meta() are just removed, because since introduction of ABD ARC they really make no sense: only headers referenced by small number of buffers are not evictable, and they are really not evictable no matter what this code do. Instead just call arc_prune_async() if too much metadata appear not evictable. Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Reviewed-by: Allan Jude <allan@klarasystems.com> Signed-off-by: Alexander Motin <mav@FreeBSD.org> Sponsored by: iXsystems, Inc. Closes #14359
2023-03-08 22:17:23 +03:00
mfug_data = arc_stats['mfu_ghost_data']
mfug_metadata = arc_stats['mfu_ghost_metadata']
mrug_data = arc_stats['mru_ghost_data']
mrug_metadata = arc_stats['mru_ghost_metadata']
unc_data = arc_stats['uncached_data']
unc_metadata = arc_stats['uncached_metadata']
caches_size = int(anon_data)+int(anon_metadata)+\
int(mfu_data)+int(mfu_metadata)+int(mru_data)+int(mru_metadata)+\
int(unc_data)+int(unc_metadata)
prt_1('ARC types breakdown (compressed + overhead):', f_bytes(caches_size))
prt_i2('Data size:',
f_perc(data_size, caches_size), f_bytes(data_size))
prt_i2('Metadata size:',
f_perc(metadata_size, caches_size), f_bytes(metadata_size))
print()
prt_1('ARC states breakdown (compressed + overhead):', f_bytes(caches_size))
More adaptive ARC eviction Traditionally ARC adaptation was limited to MRU/MFU distribution. But for years people with metadata-centric workload demanded mechanisms to also manage data/metadata distribution, that in original ZFS was just a FIFO. As result ZFS effectively got separate states for data and metadata, minimum and maximum metadata limits etc, but it all required manual tuning, was not adaptive and in its heart remained a bad FIFO. This change removes most of existing eviction logic, rewriting it from scratch. This makes MRU/MFU adaptation individual for data and meta- data, same as the distribution between data and metadata themselves. Since most of required states separation was already done, it only required to make arcs_size state field specific per data/metadata. The adaptation logic is still based on previous concept of ghost hits, just now it balances ARC capacity between 4 states: MRU data, MRU metadata, MFU data and MFU metadata. To simplify arc_c changes instead of arc_p measured in bytes, this code uses 3 variable arc_meta, arc_pd and arc_pm, representing ARC balance between metadata and data, MRU and MFU for data, and MRU and MFU for metadata respectively as 32-bit fixed point fractions. Since we care about the math result only when need to evict, this moves all the logic from arc_adapt() to arc_evict(), that reduces per-block overhead, since per-block operations are limited to stats collection, now moved from arc_adapt() to arc_access() and using cheaper wmsums. This also allows to remove ugly ARC_HDR_DO_ADAPT flag from many places. This change also removes number of metadata specific tunables, part of which were actually not functioning correctly, since not all metadata are equal and some (like L2ARC headers) are not really evictable. Instead it introduced single opaque knob zfs_arc_meta_balance, tuning ARC's reaction on ghost hits, allowing administrator give more or less preference to metadata without setting strict limits. Some of old code parts like arc_evict_meta() are just removed, because since introduction of ABD ARC they really make no sense: only headers referenced by small number of buffers are not evictable, and they are really not evictable no matter what this code do. Instead just call arc_prune_async() if too much metadata appear not evictable. Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Reviewed-by: Allan Jude <allan@klarasystems.com> Signed-off-by: Alexander Motin <mav@FreeBSD.org> Sponsored by: iXsystems, Inc. Closes #14359
2023-03-08 22:17:23 +03:00
prt_i2('Anonymous data size:',
f_perc(anon_data, caches_size), f_bytes(anon_data))
prt_i2('Anonymous metadata size:',
f_perc(anon_metadata, caches_size), f_bytes(anon_metadata))
s = 4294967296
v = (s-int(pd))*(s-int(meta))/s
prt_i2('MFU data target:', f_perc(v, s),
f_bytes(v / 65536 * caches_size / 65536))
prt_i2('MFU data size:',
f_perc(mfu_data, caches_size), f_bytes(mfu_data))
prt_i2('MFU evictable data size:',
f_perc(mfu_edata, caches_size), f_bytes(mfu_edata))
More adaptive ARC eviction Traditionally ARC adaptation was limited to MRU/MFU distribution. But for years people with metadata-centric workload demanded mechanisms to also manage data/metadata distribution, that in original ZFS was just a FIFO. As result ZFS effectively got separate states for data and metadata, minimum and maximum metadata limits etc, but it all required manual tuning, was not adaptive and in its heart remained a bad FIFO. This change removes most of existing eviction logic, rewriting it from scratch. This makes MRU/MFU adaptation individual for data and meta- data, same as the distribution between data and metadata themselves. Since most of required states separation was already done, it only required to make arcs_size state field specific per data/metadata. The adaptation logic is still based on previous concept of ghost hits, just now it balances ARC capacity between 4 states: MRU data, MRU metadata, MFU data and MFU metadata. To simplify arc_c changes instead of arc_p measured in bytes, this code uses 3 variable arc_meta, arc_pd and arc_pm, representing ARC balance between metadata and data, MRU and MFU for data, and MRU and MFU for metadata respectively as 32-bit fixed point fractions. Since we care about the math result only when need to evict, this moves all the logic from arc_adapt() to arc_evict(), that reduces per-block overhead, since per-block operations are limited to stats collection, now moved from arc_adapt() to arc_access() and using cheaper wmsums. This also allows to remove ugly ARC_HDR_DO_ADAPT flag from many places. This change also removes number of metadata specific tunables, part of which were actually not functioning correctly, since not all metadata are equal and some (like L2ARC headers) are not really evictable. Instead it introduced single opaque knob zfs_arc_meta_balance, tuning ARC's reaction on ghost hits, allowing administrator give more or less preference to metadata without setting strict limits. Some of old code parts like arc_evict_meta() are just removed, because since introduction of ABD ARC they really make no sense: only headers referenced by small number of buffers are not evictable, and they are really not evictable no matter what this code do. Instead just call arc_prune_async() if too much metadata appear not evictable. Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Reviewed-by: Allan Jude <allan@klarasystems.com> Signed-off-by: Alexander Motin <mav@FreeBSD.org> Sponsored by: iXsystems, Inc. Closes #14359
2023-03-08 22:17:23 +03:00
prt_i1('MFU ghost data size:', f_bytes(mfug_data))
v = (s-int(pm))*int(meta)/s
prt_i2('MFU metadata target:', f_perc(v, s),
f_bytes(v / 65536 * caches_size / 65536))
prt_i2('MFU metadata size:',
f_perc(mfu_metadata, caches_size), f_bytes(mfu_metadata))
prt_i2('MFU evictable metadata size:',
f_perc(mfu_emetadata, caches_size), f_bytes(mfu_emetadata))
More adaptive ARC eviction Traditionally ARC adaptation was limited to MRU/MFU distribution. But for years people with metadata-centric workload demanded mechanisms to also manage data/metadata distribution, that in original ZFS was just a FIFO. As result ZFS effectively got separate states for data and metadata, minimum and maximum metadata limits etc, but it all required manual tuning, was not adaptive and in its heart remained a bad FIFO. This change removes most of existing eviction logic, rewriting it from scratch. This makes MRU/MFU adaptation individual for data and meta- data, same as the distribution between data and metadata themselves. Since most of required states separation was already done, it only required to make arcs_size state field specific per data/metadata. The adaptation logic is still based on previous concept of ghost hits, just now it balances ARC capacity between 4 states: MRU data, MRU metadata, MFU data and MFU metadata. To simplify arc_c changes instead of arc_p measured in bytes, this code uses 3 variable arc_meta, arc_pd and arc_pm, representing ARC balance between metadata and data, MRU and MFU for data, and MRU and MFU for metadata respectively as 32-bit fixed point fractions. Since we care about the math result only when need to evict, this moves all the logic from arc_adapt() to arc_evict(), that reduces per-block overhead, since per-block operations are limited to stats collection, now moved from arc_adapt() to arc_access() and using cheaper wmsums. This also allows to remove ugly ARC_HDR_DO_ADAPT flag from many places. This change also removes number of metadata specific tunables, part of which were actually not functioning correctly, since not all metadata are equal and some (like L2ARC headers) are not really evictable. Instead it introduced single opaque knob zfs_arc_meta_balance, tuning ARC's reaction on ghost hits, allowing administrator give more or less preference to metadata without setting strict limits. Some of old code parts like arc_evict_meta() are just removed, because since introduction of ABD ARC they really make no sense: only headers referenced by small number of buffers are not evictable, and they are really not evictable no matter what this code do. Instead just call arc_prune_async() if too much metadata appear not evictable. Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Reviewed-by: Allan Jude <allan@klarasystems.com> Signed-off-by: Alexander Motin <mav@FreeBSD.org> Sponsored by: iXsystems, Inc. Closes #14359
2023-03-08 22:17:23 +03:00
prt_i1('MFU ghost metadata size:', f_bytes(mfug_metadata))
v = int(pd)*(s-int(meta))/s
prt_i2('MRU data target:', f_perc(v, s),
f_bytes(v / 65536 * caches_size / 65536))
prt_i2('MRU data size:',
f_perc(mru_data, caches_size), f_bytes(mru_data))
prt_i2('MRU evictable data size:',
f_perc(mru_edata, caches_size), f_bytes(mru_edata))
More adaptive ARC eviction Traditionally ARC adaptation was limited to MRU/MFU distribution. But for years people with metadata-centric workload demanded mechanisms to also manage data/metadata distribution, that in original ZFS was just a FIFO. As result ZFS effectively got separate states for data and metadata, minimum and maximum metadata limits etc, but it all required manual tuning, was not adaptive and in its heart remained a bad FIFO. This change removes most of existing eviction logic, rewriting it from scratch. This makes MRU/MFU adaptation individual for data and meta- data, same as the distribution between data and metadata themselves. Since most of required states separation was already done, it only required to make arcs_size state field specific per data/metadata. The adaptation logic is still based on previous concept of ghost hits, just now it balances ARC capacity between 4 states: MRU data, MRU metadata, MFU data and MFU metadata. To simplify arc_c changes instead of arc_p measured in bytes, this code uses 3 variable arc_meta, arc_pd and arc_pm, representing ARC balance between metadata and data, MRU and MFU for data, and MRU and MFU for metadata respectively as 32-bit fixed point fractions. Since we care about the math result only when need to evict, this moves all the logic from arc_adapt() to arc_evict(), that reduces per-block overhead, since per-block operations are limited to stats collection, now moved from arc_adapt() to arc_access() and using cheaper wmsums. This also allows to remove ugly ARC_HDR_DO_ADAPT flag from many places. This change also removes number of metadata specific tunables, part of which were actually not functioning correctly, since not all metadata are equal and some (like L2ARC headers) are not really evictable. Instead it introduced single opaque knob zfs_arc_meta_balance, tuning ARC's reaction on ghost hits, allowing administrator give more or less preference to metadata without setting strict limits. Some of old code parts like arc_evict_meta() are just removed, because since introduction of ABD ARC they really make no sense: only headers referenced by small number of buffers are not evictable, and they are really not evictable no matter what this code do. Instead just call arc_prune_async() if too much metadata appear not evictable. Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Reviewed-by: Allan Jude <allan@klarasystems.com> Signed-off-by: Alexander Motin <mav@FreeBSD.org> Sponsored by: iXsystems, Inc. Closes #14359
2023-03-08 22:17:23 +03:00
prt_i1('MRU ghost data size:', f_bytes(mrug_data))
v = int(pm)*int(meta)/s
prt_i2('MRU metadata target:', f_perc(v, s),
f_bytes(v / 65536 * caches_size / 65536))
prt_i2('MRU metadata size:',
f_perc(mru_metadata, caches_size), f_bytes(mru_metadata))
prt_i2('MRU evictable metadata size:',
f_perc(mru_emetadata, caches_size), f_bytes(mru_emetadata))
More adaptive ARC eviction Traditionally ARC adaptation was limited to MRU/MFU distribution. But for years people with metadata-centric workload demanded mechanisms to also manage data/metadata distribution, that in original ZFS was just a FIFO. As result ZFS effectively got separate states for data and metadata, minimum and maximum metadata limits etc, but it all required manual tuning, was not adaptive and in its heart remained a bad FIFO. This change removes most of existing eviction logic, rewriting it from scratch. This makes MRU/MFU adaptation individual for data and meta- data, same as the distribution between data and metadata themselves. Since most of required states separation was already done, it only required to make arcs_size state field specific per data/metadata. The adaptation logic is still based on previous concept of ghost hits, just now it balances ARC capacity between 4 states: MRU data, MRU metadata, MFU data and MFU metadata. To simplify arc_c changes instead of arc_p measured in bytes, this code uses 3 variable arc_meta, arc_pd and arc_pm, representing ARC balance between metadata and data, MRU and MFU for data, and MRU and MFU for metadata respectively as 32-bit fixed point fractions. Since we care about the math result only when need to evict, this moves all the logic from arc_adapt() to arc_evict(), that reduces per-block overhead, since per-block operations are limited to stats collection, now moved from arc_adapt() to arc_access() and using cheaper wmsums. This also allows to remove ugly ARC_HDR_DO_ADAPT flag from many places. This change also removes number of metadata specific tunables, part of which were actually not functioning correctly, since not all metadata are equal and some (like L2ARC headers) are not really evictable. Instead it introduced single opaque knob zfs_arc_meta_balance, tuning ARC's reaction on ghost hits, allowing administrator give more or less preference to metadata without setting strict limits. Some of old code parts like arc_evict_meta() are just removed, because since introduction of ABD ARC they really make no sense: only headers referenced by small number of buffers are not evictable, and they are really not evictable no matter what this code do. Instead just call arc_prune_async() if too much metadata appear not evictable. Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Reviewed-by: Allan Jude <allan@klarasystems.com> Signed-off-by: Alexander Motin <mav@FreeBSD.org> Sponsored by: iXsystems, Inc. Closes #14359
2023-03-08 22:17:23 +03:00
prt_i1('MRU ghost metadata size:', f_bytes(mrug_metadata))
prt_i2('Uncached data size:',
More adaptive ARC eviction Traditionally ARC adaptation was limited to MRU/MFU distribution. But for years people with metadata-centric workload demanded mechanisms to also manage data/metadata distribution, that in original ZFS was just a FIFO. As result ZFS effectively got separate states for data and metadata, minimum and maximum metadata limits etc, but it all required manual tuning, was not adaptive and in its heart remained a bad FIFO. This change removes most of existing eviction logic, rewriting it from scratch. This makes MRU/MFU adaptation individual for data and meta- data, same as the distribution between data and metadata themselves. Since most of required states separation was already done, it only required to make arcs_size state field specific per data/metadata. The adaptation logic is still based on previous concept of ghost hits, just now it balances ARC capacity between 4 states: MRU data, MRU metadata, MFU data and MFU metadata. To simplify arc_c changes instead of arc_p measured in bytes, this code uses 3 variable arc_meta, arc_pd and arc_pm, representing ARC balance between metadata and data, MRU and MFU for data, and MRU and MFU for metadata respectively as 32-bit fixed point fractions. Since we care about the math result only when need to evict, this moves all the logic from arc_adapt() to arc_evict(), that reduces per-block overhead, since per-block operations are limited to stats collection, now moved from arc_adapt() to arc_access() and using cheaper wmsums. This also allows to remove ugly ARC_HDR_DO_ADAPT flag from many places. This change also removes number of metadata specific tunables, part of which were actually not functioning correctly, since not all metadata are equal and some (like L2ARC headers) are not really evictable. Instead it introduced single opaque knob zfs_arc_meta_balance, tuning ARC's reaction on ghost hits, allowing administrator give more or less preference to metadata without setting strict limits. Some of old code parts like arc_evict_meta() are just removed, because since introduction of ABD ARC they really make no sense: only headers referenced by small number of buffers are not evictable, and they are really not evictable no matter what this code do. Instead just call arc_prune_async() if too much metadata appear not evictable. Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Reviewed-by: Allan Jude <allan@klarasystems.com> Signed-off-by: Alexander Motin <mav@FreeBSD.org> Sponsored by: iXsystems, Inc. Closes #14359
2023-03-08 22:17:23 +03:00
f_perc(unc_data, caches_size), f_bytes(unc_data))
prt_i2('Uncached metadata size:',
f_perc(unc_metadata, caches_size), f_bytes(unc_metadata))
print()
print('ARC hash breakdown:')
prt_i1('Elements max:', f_hits(arc_stats['hash_elements_max']))
prt_i2('Elements current:',
f_perc(arc_stats['hash_elements'], arc_stats['hash_elements_max']),
f_hits(arc_stats['hash_elements']))
prt_i1('Collisions:', f_hits(arc_stats['hash_collisions']))
prt_i1('Chain max:', f_hits(arc_stats['hash_chain_max']))
prt_i1('Chains:', f_hits(arc_stats['hash_chains']))
print()
print('ARC misc:')
prt_i1('Memory throttles:', arc_stats['memory_throttle_count'])
prt_i1('Memory direct reclaims:', arc_stats['memory_direct_count'])
prt_i1('Memory indirect reclaims:', arc_stats['memory_indirect_count'])
prt_i1('Deleted:', f_hits(arc_stats['deleted']))
prt_i1('Mutex misses:', f_hits(arc_stats['mutex_miss']))
prt_i1('Eviction skips:', f_hits(arc_stats['evict_skip']))
Add L2ARC arcstats for MFU/MRU buffers and buffer content type Currently the ARC state (MFU/MRU) of cached L2ARC buffer and their content type is unknown. Knowing this information may prove beneficial in adjusting the L2ARC caching policy. This commit adds L2ARC arcstats that display the aligned size (in bytes) of L2ARC buffers according to their content type (data/metadata) and according to their ARC state (MRU/MFU or prefetch). It also expands the existing evict_l2_eligible arcstat to differentiate between MFU and MRU buffers. L2ARC caches buffers from the MRU and MFU lists of ARC. Upon caching a buffer, its ARC state (MRU/MFU) is stored in the L2 header (b_arcs_state). The l2_m{f,r}u_asize arcstats reflect the aligned size (in bytes) of L2ARC buffers according to their ARC state (based on b_arcs_state). We also account for the case where an L2ARC and ARC cached MRU or MRU_ghost buffer transitions to MFU. The l2_prefetch_asize reflects the alinged size (in bytes) of L2ARC buffers that were cached while they had the prefetch flag set in ARC. This is dynamically updated as the prefetch flag of L2ARC buffers changes. When buffers are evicted from ARC, if they are determined to be L2ARC eligible then their logical size is recorded in evict_l2_eligible_m{r,f}u arcstats according to their ARC state upon eviction. Persistent L2ARC: When committing an L2ARC buffer to a log block (L2ARC metadata) its b_arcs_state and prefetch flag is also stored. If the buffer changes its arcstate or prefetch flag this is reflected in the above arcstats. However, the L2ARC metadata cannot currently be updated to reflect this change. Example: L2ARC caches an MRU buffer. L2ARC metadata and arcstats count this as an MRU buffer. The buffer transitions to MFU. The arcstats are updated to reflect this. Upon pool re-import or on/offlining the L2ARC device the arcstats are cleared and the buffer will now be counted as an MRU buffer, as the L2ARC metadata were not updated. Bug fix: - If l2arc_noprefetch is set, arc_read_done clears the L2CACHE flag of an ARC buffer. However, prefetches may be issued in a way that arc_read_done() is bypassed. Instead, move the related code in l2arc_write_eligible() to account for those cases too. Also add a test and update manpages for l2arc_mfuonly module parameter, and update the manpages and code comments for l2arc_noprefetch. Move persist_l2arc tests to l2arc. Reviewed-by: Ryan Moeller <freqlabs@FreeBSD.org> Reviewed-by: Richard Elling <Richard.Elling@RichardElling.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Signed-off-by: George Amanakis <gamanakis@gmail.com> Closes #10743
2020-09-14 20:10:44 +03:00
prt_i1('Eviction skips due to L2 writes:',
f_hits(arc_stats['evict_l2_skip']))
prt_i1('L2 cached evictions:', f_bytes(arc_stats['evict_l2_cached']))
prt_i1('L2 eligible evictions:', f_bytes(arc_stats['evict_l2_eligible']))
prt_i2('L2 eligible MFU evictions:',
f_perc(arc_stats['evict_l2_eligible_mfu'],
arc_stats['evict_l2_eligible']),
f_bytes(arc_stats['evict_l2_eligible_mfu']))
prt_i2('L2 eligible MRU evictions:',
f_perc(arc_stats['evict_l2_eligible_mru'],
arc_stats['evict_l2_eligible']),
f_bytes(arc_stats['evict_l2_eligible_mru']))
prt_i1('L2 ineligible evictions:',
f_bytes(arc_stats['evict_l2_ineligible']))
print()
def section_archits(kstats_dict):
"""Print information on how the caches are accessed ("arc hits").
"""
arc_stats = isolate_section('arcstats', kstats_dict)
all_accesses = int(arc_stats['hits'])+int(arc_stats['iohits'])+\
int(arc_stats['misses'])
prt_1('ARC total accesses:', f_hits(all_accesses))
ta_todo = (('Total hits:', arc_stats['hits']),
('Total I/O hits:', arc_stats['iohits']),
('Total misses:', arc_stats['misses']))
for title, value in ta_todo:
prt_i2(title, f_perc(value, all_accesses), f_hits(value))
print()
dd_total = int(arc_stats['demand_data_hits']) +\
int(arc_stats['demand_data_iohits']) +\
int(arc_stats['demand_data_misses'])
prt_2('ARC demand data accesses:', f_perc(dd_total, all_accesses),
f_hits(dd_total))
dd_todo = (('Demand data hits:', arc_stats['demand_data_hits']),
('Demand data I/O hits:', arc_stats['demand_data_iohits']),
('Demand data misses:', arc_stats['demand_data_misses']))
for title, value in dd_todo:
prt_i2(title, f_perc(value, dd_total), f_hits(value))
print()
dm_total = int(arc_stats['demand_metadata_hits']) +\
int(arc_stats['demand_metadata_iohits']) +\
int(arc_stats['demand_metadata_misses'])
prt_2('ARC demand metadata accesses:', f_perc(dm_total, all_accesses),
f_hits(dm_total))
dm_todo = (('Demand metadata hits:', arc_stats['demand_metadata_hits']),
('Demand metadata I/O hits:',
arc_stats['demand_metadata_iohits']),
('Demand metadata misses:', arc_stats['demand_metadata_misses']))
for title, value in dm_todo:
prt_i2(title, f_perc(value, dm_total), f_hits(value))
print()
pd_total = int(arc_stats['prefetch_data_hits']) +\
int(arc_stats['prefetch_data_iohits']) +\
int(arc_stats['prefetch_data_misses'])
prt_2('ARC prefetch data accesses:', f_perc(pd_total, all_accesses),
f_hits(pd_total))
pd_todo = (('Prefetch data hits:', arc_stats['prefetch_data_hits']),
('Prefetch data I/O hits:', arc_stats['prefetch_data_iohits']),
('Prefetch data misses:', arc_stats['prefetch_data_misses']))
for title, value in pd_todo:
prt_i2(title, f_perc(value, pd_total), f_hits(value))
print()
pm_total = int(arc_stats['prefetch_metadata_hits']) +\
int(arc_stats['prefetch_metadata_iohits']) +\
int(arc_stats['prefetch_metadata_misses'])
prt_2('ARC prefetch metadata accesses:', f_perc(pm_total, all_accesses),
f_hits(pm_total))
pm_todo = (('Prefetch metadata hits:',
arc_stats['prefetch_metadata_hits']),
('Prefetch metadata I/O hits:',
arc_stats['prefetch_metadata_iohits']),
('Prefetch metadata misses:',
arc_stats['prefetch_metadata_misses']))
for title, value in pm_todo:
prt_i2(title, f_perc(value, pm_total), f_hits(value))
print()
all_prefetches = int(arc_stats['predictive_prefetch'])+\
int(arc_stats['prescient_prefetch'])
prt_2('ARC predictive prefetches:',
f_perc(arc_stats['predictive_prefetch'], all_prefetches),
f_hits(arc_stats['predictive_prefetch']))
prt_i2('Demand hits after predictive:',
f_perc(arc_stats['demand_hit_predictive_prefetch'],
arc_stats['predictive_prefetch']),
f_hits(arc_stats['demand_hit_predictive_prefetch']))
prt_i2('Demand I/O hits after predictive:',
f_perc(arc_stats['demand_iohit_predictive_prefetch'],
arc_stats['predictive_prefetch']),
f_hits(arc_stats['demand_iohit_predictive_prefetch']))
never = int(arc_stats['predictive_prefetch']) -\
int(arc_stats['demand_hit_predictive_prefetch']) -\
int(arc_stats['demand_iohit_predictive_prefetch'])
prt_i2('Never demanded after predictive:',
f_perc(never, arc_stats['predictive_prefetch']),
f_hits(never))
print()
prt_2('ARC prescient prefetches:',
f_perc(arc_stats['prescient_prefetch'], all_prefetches),
f_hits(arc_stats['prescient_prefetch']))
prt_i2('Demand hits after prescient:',
f_perc(arc_stats['demand_hit_prescient_prefetch'],
arc_stats['prescient_prefetch']),
f_hits(arc_stats['demand_hit_prescient_prefetch']))
prt_i2('Demand I/O hits after prescient:',
f_perc(arc_stats['demand_iohit_prescient_prefetch'],
arc_stats['prescient_prefetch']),
f_hits(arc_stats['demand_iohit_prescient_prefetch']))
never = int(arc_stats['prescient_prefetch'])-\
int(arc_stats['demand_hit_prescient_prefetch'])-\
int(arc_stats['demand_iohit_prescient_prefetch'])
prt_i2('Never demanded after prescient:',
f_perc(never, arc_stats['prescient_prefetch']),
f_hits(never))
print()
print('ARC states hits of all accesses:')
cl_todo = (('Most frequently used (MFU):', arc_stats['mfu_hits']),
('Most recently used (MRU):', arc_stats['mru_hits']),
('Most frequently used (MFU) ghost:',
arc_stats['mfu_ghost_hits']),
('Most recently used (MRU) ghost:',
arc_stats['mru_ghost_hits']),
('Uncached:', arc_stats['uncached_hits']))
for title, value in cl_todo:
prt_i2(title, f_perc(value, all_accesses), f_hits(value))
print()
def section_dmu(kstats_dict):
"""Collect information on the DMU"""
zfetch_stats = isolate_section('zfetchstats', kstats_dict)
zfetch_access_total = int(zfetch_stats['hits']) +\
int(zfetch_stats['future']) + int(zfetch_stats['stride']) +\
int(zfetch_stats['past']) + int(zfetch_stats['misses'])
prt_1('DMU predictive prefetcher calls:', f_hits(zfetch_access_total))
prt_i2('Stream hits:',
f_perc(zfetch_stats['hits'], zfetch_access_total),
f_hits(zfetch_stats['hits']))
future = int(zfetch_stats['future']) + int(zfetch_stats['stride'])
prt_i2('Hits ahead of stream:', f_perc(future, zfetch_access_total),
f_hits(future))
prt_i2('Hits behind stream:',
f_perc(zfetch_stats['past'], zfetch_access_total),
f_hits(zfetch_stats['past']))
prt_i2('Stream misses:',
f_perc(zfetch_stats['misses'], zfetch_access_total),
f_hits(zfetch_stats['misses']))
prt_i2('Streams limit reached:',
f_perc(zfetch_stats['max_streams'], zfetch_stats['misses']),
f_hits(zfetch_stats['max_streams']))
prt_i1('Stream strides:', f_hits(zfetch_stats['stride']))
prt_i1('Prefetches issued', f_hits(zfetch_stats['io_issued']))
print()
def section_l2arc(kstats_dict):
"""Collect information on L2ARC device if present. If not, tell user
that we're skipping the section.
"""
# The L2ARC statistics live in the same section as the normal ARC stuff
arc_stats = isolate_section('arcstats', kstats_dict)
if arc_stats['l2_size'] == '0':
print('L2ARC not detected, skipping section\n')
return
l2_errors = int(arc_stats['l2_writes_error']) +\
int(arc_stats['l2_cksum_bad']) +\
int(arc_stats['l2_io_error'])
l2_access_total = int(arc_stats['l2_hits'])+int(arc_stats['l2_misses'])
health = 'HEALTHY'
if l2_errors > 0:
health = 'DEGRADED'
prt_1('L2ARC status:', health)
l2_todo = (('Low memory aborts:', 'l2_abort_lowmem'),
('Free on write:', 'l2_free_on_write'),
('R/W clashes:', 'l2_rw_clash'),
('Bad checksums:', 'l2_cksum_bad'),
('Read errors:', 'l2_io_error'),
('Write errors:', 'l2_writes_error'))
for title, value in l2_todo:
prt_i1(title, f_hits(arc_stats[value]))
print()
prt_1('L2ARC size (adaptive):', f_bytes(arc_stats['l2_size']))
prt_i2('Compressed:', f_perc(arc_stats['l2_asize'], arc_stats['l2_size']),
f_bytes(arc_stats['l2_asize']))
prt_i2('Header size:',
f_perc(arc_stats['l2_hdr_size'], arc_stats['l2_size']),
f_bytes(arc_stats['l2_hdr_size']))
Add L2ARC arcstats for MFU/MRU buffers and buffer content type Currently the ARC state (MFU/MRU) of cached L2ARC buffer and their content type is unknown. Knowing this information may prove beneficial in adjusting the L2ARC caching policy. This commit adds L2ARC arcstats that display the aligned size (in bytes) of L2ARC buffers according to their content type (data/metadata) and according to their ARC state (MRU/MFU or prefetch). It also expands the existing evict_l2_eligible arcstat to differentiate between MFU and MRU buffers. L2ARC caches buffers from the MRU and MFU lists of ARC. Upon caching a buffer, its ARC state (MRU/MFU) is stored in the L2 header (b_arcs_state). The l2_m{f,r}u_asize arcstats reflect the aligned size (in bytes) of L2ARC buffers according to their ARC state (based on b_arcs_state). We also account for the case where an L2ARC and ARC cached MRU or MRU_ghost buffer transitions to MFU. The l2_prefetch_asize reflects the alinged size (in bytes) of L2ARC buffers that were cached while they had the prefetch flag set in ARC. This is dynamically updated as the prefetch flag of L2ARC buffers changes. When buffers are evicted from ARC, if they are determined to be L2ARC eligible then their logical size is recorded in evict_l2_eligible_m{r,f}u arcstats according to their ARC state upon eviction. Persistent L2ARC: When committing an L2ARC buffer to a log block (L2ARC metadata) its b_arcs_state and prefetch flag is also stored. If the buffer changes its arcstate or prefetch flag this is reflected in the above arcstats. However, the L2ARC metadata cannot currently be updated to reflect this change. Example: L2ARC caches an MRU buffer. L2ARC metadata and arcstats count this as an MRU buffer. The buffer transitions to MFU. The arcstats are updated to reflect this. Upon pool re-import or on/offlining the L2ARC device the arcstats are cleared and the buffer will now be counted as an MRU buffer, as the L2ARC metadata were not updated. Bug fix: - If l2arc_noprefetch is set, arc_read_done clears the L2CACHE flag of an ARC buffer. However, prefetches may be issued in a way that arc_read_done() is bypassed. Instead, move the related code in l2arc_write_eligible() to account for those cases too. Also add a test and update manpages for l2arc_mfuonly module parameter, and update the manpages and code comments for l2arc_noprefetch. Move persist_l2arc tests to l2arc. Reviewed-by: Ryan Moeller <freqlabs@FreeBSD.org> Reviewed-by: Richard Elling <Richard.Elling@RichardElling.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Signed-off-by: George Amanakis <gamanakis@gmail.com> Closes #10743
2020-09-14 20:10:44 +03:00
prt_i2('MFU allocated size:',
f_perc(arc_stats['l2_mfu_asize'], arc_stats['l2_asize']),
f_bytes(arc_stats['l2_mfu_asize']))
prt_i2('MRU allocated size:',
f_perc(arc_stats['l2_mru_asize'], arc_stats['l2_asize']),
f_bytes(arc_stats['l2_mru_asize']))
prt_i2('Prefetch allocated size:',
f_perc(arc_stats['l2_prefetch_asize'], arc_stats['l2_asize']),
f_bytes(arc_stats['l2_prefetch_asize']))
prt_i2('Data (buffer content) allocated size:',
f_perc(arc_stats['l2_bufc_data_asize'], arc_stats['l2_asize']),
f_bytes(arc_stats['l2_bufc_data_asize']))
prt_i2('Metadata (buffer content) allocated size:',
f_perc(arc_stats['l2_bufc_metadata_asize'], arc_stats['l2_asize']),
f_bytes(arc_stats['l2_bufc_metadata_asize']))
print()
prt_1('L2ARC breakdown:', f_hits(l2_access_total))
prt_i2('Hit ratio:',
f_perc(arc_stats['l2_hits'], l2_access_total),
f_hits(arc_stats['l2_hits']))
prt_i2('Miss ratio:',
f_perc(arc_stats['l2_misses'], l2_access_total),
f_hits(arc_stats['l2_misses']))
print()
print('L2ARC I/O:')
prt_i2('Reads:',
f_bytes(arc_stats['l2_read_bytes']),
f_hits(arc_stats['l2_hits']))
prt_i2('Writes:',
f_bytes(arc_stats['l2_write_bytes']),
f_hits(arc_stats['l2_writes_sent']))
print()
print('L2ARC evicts:')
prt_i1('L1 cached:', f_hits(arc_stats['l2_evict_l1cached']))
prt_i1('While reading:', f_hits(arc_stats['l2_evict_reading']))
print()
def section_spl(*_):
"""Print the SPL parameters, if requested with alternative format
and/or descriptions. This does not use kstats.
"""
if sys.platform.startswith('freebsd'):
# No SPL support in FreeBSD
return
spls = get_spl_params()
keylist = sorted(spls.keys())
print('Solaris Porting Layer (SPL):')
if ARGS.desc:
descriptions = get_descriptions('spl')
for key in keylist:
value = spls[key]
if ARGS.desc:
try:
print(INDENT+'#', descriptions[key])
except KeyError:
print(INDENT+'# (No description found)') # paranoid
print(format_raw_line(key, value))
print()
def section_tunables(*_):
"""Print the tunables, if requested with alternative format and/or
descriptions. This does not use kstasts.
"""
tunables = get_tunable_params()
keylist = sorted(tunables.keys())
print('Tunables:')
if ARGS.desc:
descriptions = get_descriptions('zfs')
for key in keylist:
value = tunables[key]
if ARGS.desc:
try:
print(INDENT+'#', descriptions[key])
except KeyError:
print(INDENT+'# (No description found)') # paranoid
print(format_raw_line(key, value))
print()
def section_zil(kstats_dict):
"""Collect information on the ZFS Intent Log. Some of the information
taken from https://github.com/openzfs/zfs/blob/master/include/sys/zil.h
"""
zil_stats = isolate_section('zil', kstats_dict)
prt_1('ZIL committed transactions:',
f_hits(zil_stats['zil_itx_count']))
prt_i1('Commit requests:', f_hits(zil_stats['zil_commit_count']))
prt_i1('Flushes to stable storage:',
f_hits(zil_stats['zil_commit_writer_count']))
prt_i2('Transactions to SLOG storage pool:',
f_bytes(zil_stats['zil_itx_metaslab_slog_bytes']),
f_hits(zil_stats['zil_itx_metaslab_slog_count']))
prt_i2('Transactions to non-SLOG storage pool:',
f_bytes(zil_stats['zil_itx_metaslab_normal_bytes']),
f_hits(zil_stats['zil_itx_metaslab_normal_count']))
print()
section_calls = {'arc': section_arc,
'archits': section_archits,
'dmu': section_dmu,
'l2arc': section_l2arc,
'spl': section_spl,
'tunables': section_tunables,
'zil': section_zil}
def main():
"""Run program. The options to draw a graph and to print all data raw are
treated separately because they come with their own call.
"""
kstats = get_kstats()
if ARGS.graph:
draw_graph(kstats)
sys.exit(0)
print_header()
if ARGS.raw:
print_raw(kstats)
elif ARGS.section:
try:
section_calls[ARGS.section](kstats)
except KeyError:
print('Error: Section "{0}" unknown'.format(ARGS.section))
sys.exit(1)
elif ARGS.page:
print('WARNING: Pages are deprecated, please use "--section"\n')
pages_to_calls = {1: 'arc',
2: 'archits',
3: 'l2arc',
4: 'dmu',
5: 'vdev',
6: 'tunables'}
try:
call = pages_to_calls[ARGS.page]
except KeyError:
print('Error: Page "{0}" not supported'.format(ARGS.page))
sys.exit(1)
else:
section_calls[call](kstats)
else:
# If no parameters were given, we print all sections. We might want to
# change the sequence by hand
calls = sorted(section_calls.keys())
for section in calls:
section_calls[section](kstats)
sys.exit(0)
if __name__ == '__main__':
main()