qcow2: relax subclusters allocation dependencies
cover qcow2: relax subclusters allocation dependencies
Hi all!
Parallel small writes to unallocated cluster works bad when subclusters enabled.
Look, without subclusters, one of write requests will allocate the whole cluster, and all other writes to this cluster will be independent of each other, they depend only on the first one that does allocation.
With subclusters, each write to unallocated subcluster will block the whole cluster for parallel writing.
So, assume we write 8 consecutive 4k chunks in parallel:
Without subclusters, one of the chunks will block all the cluster and write L2 entry. The remaining 7 chunks are written in parallel.
With subclusters, each of the chunks will allocate new subcluster and block the whole cluster. All the chunks are dependent on each other and queue depth becomes 1. That's not good.
Let's improve the situation.
Vladimir Sementsov-Ogievskiy (3):
simplebench: add img_bench_templater.py
qcow2: refactor handle_dependencies() loop body
qcow2: handle_dependencies(): relax conflict detection
block/qcow2-cluster.c | 60 +++++++++------ scripts/simplebench/img_bench_templater.py | 85 ++++++++++++++++++++++ scripts/simplebench/table_templater.py | 62 ++++++++++++++++ tests/qemu-iotests/271 | 4 +- tests/qemu-iotests/271.out | 2 - 5 files changed, 187 insertions(+), 26 deletions(-) create mode 100755 scripts/simplebench/img_bench_templater.py create mode 100644 scripts/simplebench/table_templater.py
Signature · 1 line
2.29.2
PATCH 1/3 simplebench: add img_bench_templater.py
Add simple grammar-parsing template benchmark.
--- scripts/simplebench/img_bench_templater.py | 85 ++++++++++++++++++++++ scripts/simplebench/table_templater.py | 62 ++++++++++++++++ 2 files changed, 147 insertions(+) create mode 100755 scripts/simplebench/img_bench_templater.py create mode 100644 scripts/simplebench/table_templater.py
scripts/simplebench/img_bench_templater.py new file +85 -0
@@ -0,0 +1,85 @@ +#!/usr/bin/env python3 +# +# Run img-bench template tests +# +# Copyright (c) 2021 Virtuozzo International GmbH. +# +# This program is free software; you can redistribute it and/or modify +# it under the terms of the GNU General Public License as published by +# the Free Software Foundation; either version 2 of the License, or +# (at your option) any later version. +# +# This program is distributed in the hope that it will be useful, +# but WITHOUT ANY WARRANTY; without even the implied warranty of +# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the +# GNU General Public License for more details. +# +# You should have received a copy of the GNU General Public License +# along with this program. If not, see <http://www.gnu.org/licenses/>. +# + + +import sys +import subprocess +import re +import json + +import simplebench +from results_to_text import results_to_text +from table_templater import Templater + + +def bench_func(env, case): + test = templater.gen(env['data'], case['data']) + + p = subprocess.run(test, shell=True, stdout=subprocess.PIPE, + stderr=subprocess.STDOUT, universal_newlines=True) + + if p.returncode == 0: + try: + m = re.search(r'Run completed in (\d+.\d+) seconds.', p.stdout) + return {'seconds': float(m.group(1))} + except Exception: + return {'error': f'failed to parse qemu-img output: {p.stdout}'} + else: + return {'error': f'qemu-img failed: {p.returncode}: {p.stdout}'} + + +if __name__ == '__main__': + if len(sys.argv) > 1: + print(""" +Usage: no arguments. Just pass template test to stdin. Template test is +a bash script, last command should be qemu-img bench (it's output is parsed +to get a result). For templating use the following synax: + + column templating: {var1|var2|...} - test will use different values in + different columns. You may use several {} constructions in the test, in this + case product of all choice-sets will be used. + + row templating: [var1|var2|...] - similar thing to define rows (test-cases) + +Test tempalate example: + +Assume you want to compare two qemu-img binaries, called qemu-img-old and +qemu-img-new in your build directory in two test-cases with 4K writes and 64K +writes. Test may look like this: + +qemu_img=/path/to/qemu/build/qemu-img-{old|new} +$qemu_img create -f qcow2 /ssd/x.qcow2 1G +$qemu_img bench -c 100 -d 8 [-s 4K|-s 64K] -w -t none -n /ssd/x.qcow2 + +If pass it to stdin of img_bench_templater.py, the resulting comparison table +will contain two columns (for two binaries) and two rows (for two test-cases). +""") + sys.exit() + + templater = Templater(sys.stdin.read()) + + envs = [{'id': ' / '.join(x), 'data': x} for x in templater.columns] + cases = [{'id': ' / '.join(x), 'data': x} for x in templater.rows] + + result = simplebench.bench(bench_func, envs, cases, count=5, + initial_run=False) + print(results_to_text(result)) + with open('results.json', 'w') as f: + json.dump(result, f, indent=4)
scripts/simplebench/table_templater.py new file +62 -0
62 lines not on this page · raw message
Signature · 1 line
2.29.2
Re: [PATCH 1/3] simplebench: add img_bench_templater.py
Quoted 1 line · Vladimir Sementsov-Ogievskiy
Add simple grammar-parsing template benchmark.
This doesn’t really say much, and FWIW, for like ten minutes I thought this would do something completely different than it did (while I was trying to parse the help text).
(I thought this was about formatting an existing test’s output, and that “template” were kind of the wrong word, but then it turned out it’s exactly the right word, only that this is not about using a test’s output as a template, but actually using a template of a test (i.e. a test template, not a template test) to generate test instances to run... Which of course is much cooler.)
Functionality-wise, as far as I understand (of course I have no knowledge of lark), this looks good to me. And it’s really quite cool.
I just found the documentation confusing, so I have some suggestions for it below.
Quoted 55 lines
Signed-off-by Vladimir Sementsov-Ogievskiy <[email protected]>--- scripts/simplebench/img_bench_templater.py | 85 ++++++++++++++++++++++ scripts/simplebench/table_templater.py | 62 ++++++++++++++++ 2 files changed, 147 insertions(+)create mode 100755
scripts/simplebench/img_bench_templater.pycreate mode 100644scripts/simplebench/table_templater.pyscripts/simplebench/img_bench_templater.py new file +41
-0@@ -0,0 +1,85 @@ +#!/usr/bin/env python3 +# +# Run img-bench template tests +# +# Copyright (c) 2021 Virtuozzo International GmbH. +# +# This program is free software; you can redistribute it and/or modify +# it under the terms of the GNU General Public License as published by +# the Free Software Foundation; either version 2 of the License, or +# (at your option) any later version. +# +# This program is distributed in the hope that it will be useful, +# but WITHOUT ANY WARRANTY; without even the implied warranty of +# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the +# GNU General Public License for more details. +# +# You should have received a copy of the GNU General Public License +# along with this program. If not, see <http://www.gnu.org/licenses/>. +# + + +import sys +import subprocess +import re +import json + +import simplebench +from results_to_text import results_to_text +from table_templater import Templater + + +def bench_func(env, case): + test = templater.gen(env['data'], case['data']) + + p = subprocess.run(test, shell=True, stdout=subprocess.PIPE, + stderr=subprocess.STDOUT, universal_newlines=True) + + if p.returncode == 0: + try: + m = re.search(r'Run completed in (\d+.\d+) seconds.', p.stdout) + return {'seconds': float(m.group(1))}
scripts/simplebench/img_bench_templater.py line 51
Quoted 10 lines
scripts/simplebench/img_bench_templater.py new file +10
-0+ except Exception: + return {'error': f'failed to parse qemu-img output: {p.stdout}'} + else: + return {'error': f'qemu-img failed: {p.returncode}: {p.stdout}'} + + +if __name__ == '__main__': + if len(sys.argv) > 1: + print(""" +Usage: no arguments. Just pass template test to stdin. Template test is
FWIW, I completely misunderstood this.
At first, this sounded really ambiguous to me; then I thought that clearly this must mean that one should pipe the test’s output to this script, i.e.
$ path/to/test.sh | scripts/simplebench/img_bench_templater.py
But now after reading more, I finally understand that this is not what is meant, but actually literally passing some template of a test script to this script, i.e.
$ scripts/simplebench/img_bench_templater.py < path/to/test-template.sh
So, two things; first, I believe it should be a “test template”, not a “template test”, because this is about templates for a test, not about a test that has something to do with templates.
Second, perhaps we should start with what this does.
Perhaps:
“This script generates performance tests from a test template (example below), runs them, and displays the results in a table. The template is read from stdin. It must be written in bash and end with a qemu-img bench invocation (whose result is parsed to get the test instance’s result).”?
scripts/simplebench/img_bench_templater.py line 53
Quoted 2 lines
scripts/simplebench/img_bench_templater.py new file +2
-0+a bash script, last command should be qemu-img bench (it's output is parsed +to get a result). For templating use the following synax:
“Use the following syntax in the template to create the various different test instances:”?
scripts/simplebench/img_bench_templater.py line 61
Quoted 8 lines
scripts/simplebench/img_bench_templater.py new file +8
-0+ + column templating: {var1|var2|...} - test will use different values in + different columns. You may use several {} constructions in the test, in this + case product of all choice-sets will be used. + + row templating: [var1|var2|...] - similar thing to define rows (test-cases) + +Test tempalate example:
*template
scripts/simplebench/img_bench_templater.py line 65
Quoted 4 lines
scripts/simplebench/img_bench_templater.py new file +4
-0+ +Assume you want to compare two qemu-img binaries, called qemu-img-old and +qemu-img-new in your build directory in two test-cases with 4K writes and 64K +writes. Test may look like this:
I’d prefer s/Test/The template/.
scripts/simplebench/img_bench_templater.py line 71
Quoted 6 lines
scripts/simplebench/img_bench_templater.py new file +6
-0+ +qemu_img=/path/to/qemu/build/qemu-img-{old|new} +$qemu_img create -f qcow2 /ssd/x.qcow2 1G +$qemu_img bench -c 100 -d 8 [-s 4K|-s 64K] -w -t none -n /ssd/x.qcow2 + +If pass it to stdin of img_bench_templater.py, the resulting comparison table
s/If pass it/When passing this/
Quoted 4 lines
scripts/simplebench/img_bench_templater.py new file +4
-0+will contain two columns (for two binaries) and two rows (for two test-cases). +""") + sys.exit() +
scripts/simplebench/img_bench_templater.py line 85
Quoted 10 lines
scripts/simplebench/img_bench_templater.py new file +10
-0+ templater = Templater(sys.stdin.read()) + + envs = [{'id': ' / '.join(x), 'data': x} for x in templater.columns] + cases = [{'id': ' / '.join(x), 'data': x} for x in templater.rows] + + result = simplebench.bench(bench_func, envs, cases, count=5, + initial_run=False) + print(results_to_text(result)) + with open('results.json', 'w') as f: + json.dump(result, f, indent=4)
Is this worth documenting?
Quoted 23 lines
scripts/simplebench/table_templater.py new file +17
-0@@ -0,0 +1,62 @@ +# Parser for test templates +# +# Copyright (c) 2021 Virtuozzo International GmbH. +# +# This program is free software; you can redistribute it and/or modify +# it under the terms of the GNU General Public License as published by +# the Free Software Foundation; either version 2 of the License, or +# (at your option) any later version. +# +# This program is distributed in the hope that it will be useful, +# but WITHOUT ANY WARRANTY; without even the implied warranty of +# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the +# GNU General Public License for more details. +# +# You should have received a copy of the GNU General Public License +# along with this program. If not, see <http://www.gnu.org/licenses/>. +#
scripts/simplebench/table_templater.py line 27
Quoted 10 lines
scripts/simplebench/table_templater.py new file +10
-0+ +import itertools +from lark import Lark + +grammar = """ +start: ( text | column_switch | row_switch )+ + +column_switch: "{" text ["|" text]+ "}" +row_switch: "[" text ["|" text]+ "]" +text: /[^|{}\[\]]+/
So I have no idea how this really works, of course, but does this mean that the text pattern cannot contain pipe symbols? I.e. that you cannot use pipes in the test template?
Hanna
Quoted 35 lines
scripts/simplebench/table_templater.py new file +35
-0+""" + +parser = Lark(grammar) + +class Templater: + def __init__(self, template): + self.tree = parser.parse(template) + + c_switches = [] + r_switches = [] + for x in self.tree.children: + if x.data == 'column_switch': + c_switches.append([el.children[0].value for el in x.children]) + elif x.data == 'row_switch': + r_switches.append([el.children[0].value for el in x.children]) + + self.columns = list(itertools.product(*c_switches)) + self.rows = list(itertools.product(*r_switches)) + + def gen(self, column, row): + i = 0 + j = 0 + result = [] + + for x in self.tree.children: + if x.data == 'text': + result.append(x.children[0].value) + elif x.data == 'column_switch': + result.append(column[i]) + i += 1 + elif x.data == 'row_switch': + result.append(row[j]) + j += 1 + + return ''.join(result)
Re: [PATCH 1/3] simplebench: add img_bench_templater.py
Quoted 87 lines · Hanna Reitz
Quoted 1 line · Vladimir Sementsov-Ogievskiy
Add simple grammar-parsing template benchmark.
This doesn’t really say much, and FWIW, for like ten minutes I thought this would do something completely different than it did (while I was trying to parse the help text).
(I thought this was about formatting an existing test’s output, and that “template” were kind of the wrong word, but then it turned out it’s exactly the right word, only that this is not about using a test’s output as a template, but actually using a template of a test (i.e. a test template, not a template test) to generate test instances to run... Which of course is much cooler.)
Functionality-wise, as far as I understand (of course I have no knowledge of lark), this looks good to me. And it’s really quite cool.
I just found the documentation confusing, so I have some suggestions for it below.
Quoted 55 lines
Signed-off-by Vladimir Sementsov-Ogievskiy <[email protected]>---
scripts/simplebench/img_bench_templater.py| 85 ++++++++++++++++++++++ scripts/simplebench/table_templater.py | 62 ++++++++++++++++2 files changed, 147 insertions(+)create mode 100755
scripts/simplebench/img_bench_templater.pycreate mode 100644scripts/simplebench/table_templater.pyscripts/simplebench/img_bench_templater.py new file +41
-0@@ -0,0 +1,85 @@ +#!/usr/bin/env python3 +# +# Run img-bench template tests +# +# Copyright (c) 2021 Virtuozzo International GmbH. +# +# This program is free software; you can redistribute it and/or modify +# it under the terms of the GNU General Public License as published by +# the Free Software Foundation; either version 2 of the License, or +# (at your option) any later version. +# +# This program is distributed in the hope that it will be useful, +# but WITHOUT ANY WARRANTY; without even the implied warranty of +# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the +# GNU General Public License for more details. +# +# You should have received a copy of the GNU General Public License +# along with this program. If not, see <http://www.gnu.org/licenses/>. +# + + +import sys +import subprocess +import re +import json + +import simplebench +from results_to_text import results_to_text +from table_templater import Templater + + +def bench_func(env, case): + test = templater.gen(env['data'], case['data']) + + p = subprocess.run(test, shell=True, stdout=subprocess.PIPE, + stderr=subprocess.STDOUT, universal_newlines=True) + + if p.returncode == 0: + try: + m = re.search(r'Run completed in (\d+.\d+) seconds.', p.stdout) + return {'seconds': float(m.group(1))}scripts/simplebench/img_bench_templater.py line 51
Quoted 10 lines
scripts/simplebench/img_bench_templater.py new file +10
-0+ except Exception: + return {'error': f'failed to parse qemu-img output: {p.stdout}'} + else: + return {'error': f'qemu-img failed: {p.returncode}: {p.stdout}'} + + +if __name__ == '__main__': + if len(sys.argv) > 1: + print(""" +Usage: no arguments. Just pass template test to stdin. Template test isFWIW, I completely misunderstood this.
At first, this sounded really ambiguous to me; then I thought that clearly this must mean that one should pipe the test’s output to this script, i.e.
$ path/to/test.sh | scripts/simplebench/img_bench_templater.pyBut now after reading more, I finally understand that this is not what is meant, but actually literally passing some template of a test script to this script, i.e.
$ scripts/simplebench/img_bench_templater.py < path/to/test-template.sh
Quoted 7 lines
So, two things; first, I believe it should be a “test template”, not a “template test”, because this is about templates for a test, not about a test that has something to do with templates.
Second, perhaps we should start with what this does.
Perhaps:
“This script generates performance tests from a test template (example below), runs them, and displays the results in a table. The template is read from stdin. It must be written in bash and end with a
qemu-img benchinvocation (whose result is parsed to get the test instance’s result).”?
Yes, that's correct, thanks for wording
Quoted 77 lines
scripts/simplebench/img_bench_templater.py line 53
Quoted 3 lines
scripts/simplebench/img_bench_templater.py new file +2
-0+a bash script, last command should be qemu-img bench (it's output is parsed +to get a result). For templating use the following synax:“Use the following syntax in the template to create the various different test instances:”?
scripts/simplebench/img_bench_templater.py line 61
Quoted 8 lines
scripts/simplebench/img_bench_templater.py new file +8
-0+ + column templating: {var1|var2|...} - test will use different values in + different columns. You may use several {} constructions in the test, in this + case product of all choice-sets will be used. + + row templating: [var1|var2|...] - similar thing to define rows (test-cases) + +Test tempalate example:*template
scripts/simplebench/img_bench_templater.py line 65
Quoted 4 lines
scripts/simplebench/img_bench_templater.py new file +4
-0+ +Assume you want to compare two qemu-img binaries, called qemu-img-old and +qemu-img-new in your build directory in two test-cases with 4K writes and 64K +writes. Test may look like this:I’d prefer s/Test/The template/.
scripts/simplebench/img_bench_templater.py line 71
Quoted 6 lines
scripts/simplebench/img_bench_templater.py new file +6
-0+ +qemu_img=/path/to/qemu/build/qemu-img-{old|new} +$qemu_img create -f qcow2 /ssd/x.qcow2 1G +$qemu_img bench -c 100 -d 8 [-s 4K|-s 64K] -w -t none -n /ssd/x.qcow2 + +If pass it to stdin of img_bench_templater.py, the resulting comparison tables/If pass it/When passing this/
Quoted 4 lines
scripts/simplebench/img_bench_templater.py new file +4
-0+will contain two columns (for two binaries) and two rows (for two test-cases). +""") + sys.exit() +scripts/simplebench/img_bench_templater.py line 85
Quoted 10 lines
scripts/simplebench/img_bench_templater.py new file +10
-0+ templater = Templater(sys.stdin.read()) + + envs = [{'id': ' / '.join(x), 'data': x} for x in templater.columns] + cases = [{'id': ' / '.join(x), 'data': x} for x in templater.rows] + + result = simplebench.bench(bench_func, envs, cases, count=5, + initial_run=False) + print(results_to_text(result)) + with open('results.json', 'w') as f: + json.dump(result, f, indent=4)Is this worth documenting?
Quoted 27 lines
scripts/simplebench/table_templater.py new file +21
-0@@ -0,0 +1,62 @@ +# Parser for test templates +# +# Copyright (c) 2021 Virtuozzo International GmbH. +# +# This program is free software; you can redistribute it and/or modify +# it under the terms of the GNU General Public License as published by +# the Free Software Foundation; either version 2 of the License, or +# (at your option) any later version. +# +# This program is distributed in the hope that it will be useful, +# but WITHOUT ANY WARRANTY; without even the implied warranty of +# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the +# GNU General Public License for more details. +# +# You should have received a copy of the GNU General Public License +# along with this program. If not, see <http://www.gnu.org/licenses/>. +# + +import itertools +from lark import Lark +
Quoted 9 lines
scripts/simplebench/table_templater.py line 27
Quoted 6 lines
scripts/simplebench/table_templater.py new file +6
-0+grammar = """ +start: ( text | column_switch | row_switch )+ + +column_switch: "{" text ["|" text]+ "}" +row_switch: "[" text ["|" text]+ "]" +text: /[^|{}\[\]]+/So I have no idea how this really works, of course, but does this mean that the
textpattern cannot contain pipe symbols? I.e. that you cannot use pipes in the test template?
Hmm. I didn't try. I hope lark is smart enough to keep pipes that are out of {} [] as is.. But of course, you can't hope that pipe inside {} or [] will work as bash-pipe.
Same thing with other special symbols ("{}" and "[]"). I don't want to care about this too much now. This simple grammar works good for test template in patch 03. If we need something more, we can add a kind of special symbols escaping later.
Quoted 37 lines
Quoted 36 lines
scripts/simplebench/table_templater.py new file +35
-0+""" + +parser = Lark(grammar) + +class Templater: + def __init__(self, template): + self.tree = parser.parse(template) + + c_switches = [] + r_switches = [] + for x in self.tree.children: + if x.data == 'column_switch': + c_switches.append([el.children[0].value for el in x.children]) + elif x.data == 'row_switch': + r_switches.append([el.children[0].value for el in x.children]) + + self.columns = list(itertools.product(*c_switches)) + self.rows = list(itertools.product(*r_switches)) + + def gen(self, column, row): + i = 0 + j = 0 + result = [] + + for x in self.tree.children: + if x.data == 'text': + result.append(x.children[0].value) + elif x.data == 'column_switch': + result.append(column[i]) + i += 1 + elif x.data == 'row_switch': + result.append(row[j]) + j += 1 + + return ''.join(result)
Signature · 2 lines
Best regards, Vladimir
Re: [PATCH 1/3] simplebench: add img_bench_templater.py
Quoted 2 lines · Vladimir Sementsov-Ogievskiy
Quoted 1 line · Hanna Reitz
On 24.07.21 15:38, Vladimir Sementsov-Ogievskiy wrote:
[...]
Quoted 10 lines
Quoted 10 lines
scripts/simplebench/table_templater.py new file +9
-0+import itertools +from lark import Lark + +grammar = """ +start: ( text | column_switch | row_switch )+ + +column_switch: "{" text ["|" text]+ "}" +row_switch: "[" text ["|" text]+ "]" +text: /[^|{}\[\]]+/
Quoted 4 lines
Quoted 2 lines
So I have no idea how this really works, of course, but does this mean that the
textpattern cannot contain pipe symbols? I.e. that you cannot use pipes in the test template?Hmm. I didn't try. I hope lark is smart enough to keep pipes that are out of {} [] as is.. But of course, you can't hope that pipe inside {} or [] will work as bash-pipe.
Yep, sure. It’s just that the text nonterminal symbol doesn’t look like it could match anything with a pipe in it.
Quoted 1 line
Same thing with other special symbols ("{}" and "[]"). I don't want to care about this too much now. This simple grammar works good for test template in patch 03. If we need something more, we can add a kind of special symbols escaping later.
But yes, if someone trips over this (i.e. we ourselves), we can still fix it then.
Hanna
Re: [PATCH 1/3] simplebench: add img_bench_templater.py
Quoted 16 lines · Hanna Reitz
Quoted 2 lines · Vladimir Sementsov-Ogievskiy
Quoted 1 line · Hanna Reitz
On 24.07.21 15:38, Vladimir Sementsov-Ogievskiy wrote:
[...]
Quoted 10 lines
Quoted 10 lines
scripts/simplebench/table_templater.py new file +9
-0+import itertools +from lark import Lark + +grammar = """ +start: ( text | column_switch | row_switch )+ + +column_switch: "{" text ["|" text]+ "}" +row_switch: "[" text ["|" text]+ "]" +text: /[^|{}\[\]]+/
Quoted 6 lines
Quoted 4 lines
Quoted 2 lines
So I have no idea how this really works, of course, but does this mean that the
textpattern cannot contain pipe symbols? I.e. that you cannot use pipes in the test template?Hmm. I didn't try. I hope lark is smart enough to keep pipes that are out of {} [] as is.. But of course, you can't hope that pipe inside {} or [] will work as bash-pipe.
Yep, sure. It’s just that the
textnonterminal symbol doesn’t look like it could match anything with a pipe in it.
Oops, right. To avoid it, we'll have to split text into "text_inside_brackets" and "text_outside_of_brackets".. As well, let's postpone this complication until we really need it.
Quoted 7 lines
Quoted 2 lines
Same thing with other special symbols ("{}" and "[]"). I don't want to care about this too much now. This simple grammar works good for test template in patch 03. If we need something more, we can add a kind of special symbols escaping later.
But yes, if someone trips over this (i.e. we ourselves), we can still fix it then.
Hanna
Signature · 2 lines
Best regards, Vladimir
PATCH 2/3 qcow2: refactor handle_dependencies() loop body
In reply to Vladimir Sementsov-OgievskiyNo logic change, just prepare for the following commit.
--- block/qcow2-cluster.c | 49 ++++++++++++++++++++++++------------------- 1 file changed, 28 insertions(+), 21 deletions(-)
block/qcow2-cluster.c +28 -21
@@ -1400,29 +1400,36 @@ static int handle_dependencies(BlockDriverState *bs, uint64_t guest_offset,if (end <= old_start || start >= old_end) {/* No intersection */ - }else { -if (start < old_start) { -/* Stop at the start of a running allocation */ - bytes = old_start - start; - }else { - bytes =0 ; - } +continue ; + } -/* Stop if already an l2meta exists. After yielding, it wouldn't -* be valid any more, so we'd have to clean up the old L2Metas -* and deal with requests depending on them before starting to -* gather new ones. Not worth the trouble. */ -if (bytes ==0 && *m) { - *cur_bytes =0 ; -return 0 ; - } +/* Conflict */ -if (bytes ==0 ) { -/* Wait for the dependency to complete. We need to recheck -* the free/allocated clusters when we continue. */ -qemu_co_queue_wait (&old_alloc->dependent_requests, &s->lock); -return -EAGAIN ; - } +if (start < old_start) { +/* Stop at the start of a running allocation */ + bytes = old_start - start; + }else { + bytes =0 ; + } + +/* +* Stop if already an l2meta exists. After yielding, it wouldn't +* be valid any more, so we'd have to clean up the old L2Metas +* and deal with requests depending on them before starting to +* gather new ones. Not worth the trouble. +*/ +if (bytes ==0 && *m) { + *cur_bytes =0 ; +return 0 ; + } + +if (bytes ==0 ) { +/* +* Wait for the dependency to complete. We need to recheck +* the free/allocated clusters when we continue. +*/ +qemu_co_queue_wait (&old_alloc->dependent_requests, &s->lock); +return -EAGAIN ; } }
Signature · 1 line
2.29.2
Re: [PATCH 2/3] qcow2: refactor handle_dependencies() loop body
Quoted 16 lines · Vladimir Sementsov-Ogievskiy
No logic change, just prepare for the following commit.
Signed-off-by Vladimir Sementsov-Ogievskiy <[email protected]>--- block/qcow2-cluster.c | 49 ++++++++++++++++++++++++------------------- 1 file changed, 28 insertions(+), 21 deletions(-)block/qcow2-cluster.c +0
-1@@ -1400,29 +1400,36 @@ static int handle_dependencies(BlockDriverState *bs, uint64_t guest_offset,if (end <= old_start || start >= old_end) {/* No intersection */ - }else {
block/qcow2-cluster.c static int handle_dependencies(BlockDriverState *bs, uint64_t guest_offset, · line 1411
Quoted 10 lines
block/qcow2-cluster.c +2
-7-if (start < old_start) { -/* Stop at the start of a running allocation */ - bytes = old_start - start; - }else { - bytes =0 ; - } +continue ; + } -/* Stop if already an l2meta exists. After yielding, it wouldn't
Pre-existing, but...
Quoted 14 lines
block/qcow2-cluster.c +1
-12-* be valid any more, so we'd have to clean up the old L2Metas -* and deal with requests depending on them before starting to -* gather new ones. Not worth the trouble. */ -if (bytes ==0 && *m) { - *cur_bytes =0 ; -return 0 ; - } +/* Conflict */ -if (bytes ==0 ) { -/* Wait for the dependency to complete. We need to recheck -* the free/allocated clusters when we continue. */ -qemu_co_queue_wait (&old_alloc->dependent_requests, &s->lock); -return -EAGAIN ;
block/qcow2-cluster.c static int handle_dependencies(BlockDriverState *bs, uint64_t guest_offset, · line 1416
Quoted 10 lines
block/qcow2-cluster.c +9
-1- } +if (start < old_start) { +/* Stop at the start of a running allocation */ + bytes = old_start - start; + }else { + bytes =0 ; + } + +/* +* Stop if already an l2meta exists. After yielding, it wouldn't
...might as well fix the grammar: Stop if an l2meta already exists.
Quoted 6 lines
block/qcow2-cluster.c +6
-0+* be valid any more, so we'd have to clean up the old L2Metas +* and deal with requests depending on them before starting to +* gather new ones. Not worth the trouble. +*/ +if (bytes ==0 && *m) { + *cur_bytes =0 ;
block/qcow2-cluster.c static int handle_dependencies(BlockDriverState *bs, uint64_t guest_offset, · line 1432
Quoted 10 lines
block/qcow2-cluster.c +10
-0+return 0 ; + } + +if (bytes ==0 ) { +/* +* Wait for the dependency to complete. We need to recheck +* the free/allocated clusters when we continue. +*/ +qemu_co_queue_wait (&old_alloc->dependent_requests, &s->lock); +return -EAGAIN ;
So the change adds a short-circuiting 'continue', then reduces the indentation of the rest of the loop body.
Signature · 3 lines
Eric Blake, Principal Software Engineer Red Hat, Inc. +1-919-301-3266 Virtualization: qemu.org | libvirt.org
Re: [PATCH 2/3] qcow2: refactor handle_dependencies() loop body
In reply to Vladimir Sementsov-OgievskiyQuoted 6 lines · Vladimir Sementsov-Ogievskiy
No logic change, just prepare for the following commit.
Signed-off-by Vladimir Sementsov-Ogievskiy <[email protected]>--- block/qcow2-cluster.c | 49 ++++++++++++++++++++++++------------------- 1 file changed, 28 insertions(+), 21 deletions(-)
PATCH 3/3 qcow2: handle_dependencies(): relax conflict detection
In reply to Vladimir Sementsov-OgievskiyThere is no conflict and no dependency if we have parallel writes to different subclusters of one cluster when cluster itself is already allocated. So, relax extra dependency.
Measure performance:
First, prepare build/qemu-img-old and build/qemu-img-new images.
cd scripts/simplebench
./img_bench_templater.py
Paste the following to stdin of running script:
qemu_img=../../build/qemu-img-{old|new}
$qemu_img create -f qcow2 -o extended_l2=on /ssd/x.qcow2 1G
$qemu_img bench -c 100000 -d 8 [-s 2K|-s 2K -o 512|-s $((1024*2+512))] \
-w -t none -n /ssd/x.qcow2The result:
All results are in seconds
------------------ --------- ---------
old new
-s 2K 6.7 ± 15% 6.2 ± 12%
-7%
-s 2K -o 512 13 ± 3% 11 ± 5%
-16%
-s $((1024*2+512)) 9.5 ± 4% 8.4
-12%------------------ --------- ---------
So small writes are more independent now and that helps to keep deeper io queue which improves performance.
271 iotest output becomes racy for three allocation in one cluster. Second and third writes may finish in different order. Second and third requests don't depend on each other any more. Still they both depend on first request anyway. Keep only one for consistent output.
--- block/qcow2-cluster.c | 11 +++++++++++ tests/qemu-iotests/271 | 4 +--- tests/qemu-iotests/271.out | 2 -- 3 files changed, 12 insertions(+), 5 deletions(-)
block/qcow2-cluster.c +11 -0
@@ -1403,6 +1403,17 @@ static int handle_dependencies(BlockDriverState *bs, uint64_t guest_offset,continue ; } +if (old_alloc->keep_old_clusters && + (end <=l2meta_cow_start (old_alloc) || + start >=l2meta_cow_end (old_alloc))) + { +/* +* Clusters intersect but COW areas don't. And cluster itself is +* already allocated. So, there is no actual conflict. +*/ +continue ; + } +/* Conflict */ if (start < old_start) {
tests/qemu-iotests/271 +1 -3
@@ -866,7 +866,7 @@ echo _concurrent_io() { -# Allocate three subclusters in the same cluster. +# Allocate two subclusters in the same cluster. # This works because handle_dependencies() checks whether the requests # allocate the same cluster, even if the COW regions don't overlap (in # this case they don't).
@@ -876,7 +876,6 @@ break write_aio A aio_write -P 10 30k 2k wait_break A aio_write -P 11 20k 2k -aio_write -P 12 40k 2k resume A aio_flush EOF
@@ -888,7 +887,6 @@ cat <<EOF open -o driver=$IMGFMT $TEST_IMG read -q -P 10 30k 2k read -q -P 11 20k 2k -read -q -P 12 40k 2k EOF }
tests/qemu-iotests/271.out +0 -2
@@ -721,6 +721,4 @@ wrote 2048/2048 bytes at offset 30720 2 KiB, X ops; XX:XX:XX.X (XXX YYY/sec and XXX ops/sec) wrote 2048/2048 bytes at offset 20480 2 KiB, X ops; XX:XX:XX.X (XXX YYY/sec and XXX ops/sec) -wrote 2048/2048 bytes at offset 40960 -2 KiB, X ops; XX:XX:XX.X (XXX YYY/sec and XXX ops/sec) *** done
Signature · 1 line
2.29.2
Re: [PATCH 3/3] qcow2: handle_dependencies(): relax conflict detection
Quoted 2 lines · Vladimir Sementsov-Ogievskiy
There is no conflict and no dependency if we have parallel writes to different subclusters of one cluster when cluster itself is already
when the cluster itself
Quoted 21 lines
allocated. So, relax extra dependency.
Measure performance:
First, prepare build/qemu-img-old and build/qemu-img-new images.cd
scripts/simplebench
./img_bench_templater.pyPaste the following to stdin of running script:
qemu_img=../../build/qemu-img-{old|new}
$qemu_img create -f qcow2 -o extended_l2=on /ssd/x.qcow2 1G$qemu_img bench -c 100000 -d 8 [-s 2K|-s 2K -o 512|-s $((1024*2+512))] \ -w -t none -n /ssd/x.qcow2The result:
All results are in seconds
------------------ --------- ---------
old new
Review
Quoted 7 lines
(quoted diff) +0
-4-s 2K 6.7 ± 15% 6.2 ± 12% -7% -s 2K -o 512 13 ± 3% 11 ± 5% -16% -s $((1024*2+512)) 9.5 ± 4% 8.4 -12% ------------------ --------- ---------
Cool improvement.
Quoted 8 lines
So small writes are more independent now and that helps to keep deeper io queue which improves performance.
271 iotest output becomes racy for three allocation in one cluster. Second and third writes may finish in different order. Second and third requests don't depend on each other any more. Still they both depend on first request anyway. Keep only one for consistent output.
Interesting fallout. Yes, it looks like the test is still robust enough without the extra request.
Quoted 7 lines
Signed-off-by Vladimir Sementsov-Ogievskiy <[email protected]>--- block/qcow2-cluster.c | 11 +++++++++++ tests/qemu-iotests/271 | 4 +--- tests/qemu-iotests/271.out | 2 -- 3 files changed, 12 insertions(+), 5 deletions(-)
Signature · 3 lines
Eric Blake, Principal Software Engineer Red Hat, Inc. +1-919-301-3266 Virtualization: qemu.org | libvirt.org
Re: [PATCH 3/3] qcow2: handle_dependencies(): relax conflict detection
In reply to Vladimir Sementsov-OgievskiyQuoted 31 lines · Vladimir Sementsov-Ogievskiy
There is no conflict and no dependency if we have parallel writes to different subclusters of one cluster when cluster itself is already allocated. So, relax extra dependency.
Measure performance:
First, prepare build/qemu-img-old and build/qemu-img-new images.cd
scripts/simplebench
./img_bench_templater.pyPaste the following to stdin of running script:
qemu_img=../../build/qemu-img-{old|new}
$qemu_img create -f qcow2 -o extended_l2=on /ssd/x.qcow2 1G$qemu_img bench -c 100000 -d 8 [-s 2K|-s 2K -o 512|-s $((1024*2+512))] \ -w -t none -n /ssd/x.qcow2The result:
All results are in seconds
------------------ --------- ---------
old new(quoted diff) +0
-4-s 2K 6.7 ± 15% 6.2 ± 12% -7% -s 2K -o 512 13 ± 3% 11 ± 5% -16% -s $((1024*2+512)) 9.5 ± 4% 8.4 -12% ------------------ --------- ---------
Quoted 7 lines
So small writes are more independent now and that helps to keep deeper io queue which improves performance.
271 iotest output becomes racy for three allocation in one cluster. Second and third writes may finish in different order. Second and third requests don't depend on each other any more. Still they both depend on first request anyway. Keep only one for consistent output.
I mean, we could also just filter the result (s/\(20480\|40960\)/FILTERED/ or something). Perhaps there was some idea behind doing three writes, I don’t know exactly.
I think I’d prefer a filter, because I guess this is the only test that actually will do two subcluster requests in parallel...?
Hanna
Quoted 69 lines
Signed-off-by Vladimir Sementsov-Ogievskiy <[email protected]>--- block/qcow2-cluster.c | 11 +++++++++++ tests/qemu-iotests/271 | 4 +--- tests/qemu-iotests/271.out | 2 -- 3 files changed, 12 insertions(+), 5 deletions(-)block/qcow2-cluster.c +11
-0@@ -1403,6 +1403,17 @@ static int handle_dependencies(BlockDriverState *bs, uint64_t guest_offset,continue ; } +if (old_alloc->keep_old_clusters && + (end <=l2meta_cow_start (old_alloc) || + start >=l2meta_cow_end (old_alloc))) + { +/* +* Clusters intersect but COW areas don't. And cluster itself is +* already allocated. So, there is no actual conflict. +*/ +continue ; + } +/* Conflict */ if (start < old_start) {tests/qemu-iotests/271 +1
-3@@ -866,7 +866,7 @@ echo _concurrent_io() { -# Allocate three subclusters in the same cluster. +# Allocate two subclusters in the same cluster. # This works because handle_dependencies() checks whether the requests # allocate the same cluster, even if the COW regions don't overlap (in # this case they don't).@@ -876,7 +876,6 @@ break write_aio A aio_write -P 10 30k 2k wait_break A aio_write -P 11 20k 2k -aio_write -P 12 40k 2k resume A aio_flush EOF@@ -888,7 +887,6 @@ cat <<EOF open -o driver=$IMGFMT $TEST_IMG read -q -P 10 30k 2k read -q -P 11 20k 2k -read -q -P 12 40k 2k EOF }tests/qemu-iotests/271.out +0
-2@@ -721,6 +721,4 @@ wrote 2048/2048 bytes at offset 30720 2 KiB, X ops; XX:XX:XX.X (XXX YYY/sec and XXX ops/sec) wrote 2048/2048 bytes at offset 20480 2 KiB, X ops; XX:XX:XX.X (XXX YYY/sec and XXX ops/sec) -wrote 2048/2048 bytes at offset 40960 -2 KiB, X ops; XX:XX:XX.X (XXX YYY/sec and XXX ops/sec) *** done
Re: [PATCH 3/3] qcow2: handle_dependencies(): relax conflict detection
Quoted 35 lines · Hanna Reitz
Quoted 34 lines · Vladimir Sementsov-Ogievskiy
There is no conflict and no dependency if we have parallel writes to different subclusters of one cluster when cluster itself is already allocated. So, relax extra dependency.
Measure performance:
First, prepare build/qemu-img-old and build/qemu-img-new images.cd
scripts/simplebench
./img_bench_templater.pyPaste the following to stdin of running script:
qemu_img=../../build/qemu-img-{old|new}
$qemu_img create -f qcow2 -o extended_l2=on /ssd/x.qcow2 1G $qemu_img bench -c 100000 -d 8 [-s 2K|-s 2K -o 512|-s $((1024*2+512))] \
-w -t none -n /ssd/x.qcow2The result:
All results are in seconds
------------------ --------- ---------
old new
-s 2K 6.7 ± 15% 6.2 ± 12%
-7%
-s 2K -o 512 13 ± 3% 11 ± 5%
-16%
-s $((1024*2+512)) 9.5 ± 4% 8.4
-12%
------------------ --------- ---------So small writes are more independent now and that helps to keep deeper io queue which improves performance.
Quoted 9 lines
Quoted 4 lines
271 iotest output becomes racy for three allocation in one cluster. Second and third writes may finish in different order. Second and third requests don't depend on each other any more. Still they both depend on first request anyway. Keep only one for consistent output.
I mean, we could also just filter the result (
s/\(20480\|40960\)/FILTERED/or something). Perhaps there was some idea behind doing three writes, I don’t know exactly.I think I’d prefer a filter, because I guess this is the only test that actually will do two subcluster requests in parallel...?
Reasonable, will do
Quoted 67 lines
Quoted 66 lines
Signed-off-by Vladimir Sementsov-Ogievskiy <[email protected]>---
block/qcow2-cluster.c | 11 +++++++++++
tests/qemu-iotests/271 | 4 +---
tests/qemu-iotests/271.out | 2 --3 files changed, 12 insertions(+), 5 deletions(-)block/qcow2-cluster.c +11
-0@@ -1403,6 +1403,17 @@ static int handle_dependencies(BlockDriverState *bs, uint64_t guest_offset, continue; } + if (old_alloc->keep_old_clusters && + (end <= l2meta_cow_start(old_alloc) || + start >= l2meta_cow_end(old_alloc))) + { + /* + * Clusters intersect but COW areas don't. And cluster itself is + * already allocated. So, there is no actual conflict. + */ + continue; + } + /* Conflict */ if (start < old_start) {tests/qemu-iotests/271 +1
-1@@ -866,7 +866,7 @@ echo _concurrent_io() { -# Allocate three subclusters in the same cluster. +# Allocate two subclusters in the same cluster. # This works because handle_dependencies() checks whether the requests # allocate the same cluster, even if the COW regions don't overlap (in # this case they don't).@@ -876,7 +876,6 @@ break write_aio A
aio_write-P 10 30k 2k
wait_breakA
aio_write-P 11 20k 2k
-aio_write -P 12 40k 2k
resume A
aio_flush
EOFtests/qemu-iotests/271 +0
-0@@ -888,7 +887,6 @@ cat <<EOFopen -o driver=$IMGFMT $TEST_IMG
read -q -P 10 30k 2k
read -q -P 11 20k 2k
-read -q -P 12 40k 2k
EOF
}tests/qemu-iotests/271.out +0
-0@@ -721,6 +721,4 @@ wrote 2048/2048 bytes at offset 30720 2 KiB, X ops; XX:XX:XX.X (XXX YYY/sec and XXX ops/sec)wrote 2048/2048 bytes at offset 20480
2 KiB, X ops; XX:XX:XX.X (XXX YYY/sec and XXX ops/sec) -wrote 2048/2048 bytes at offset 40960
-2 KiB, X ops; XX:XX:XX.X (XXX YYY/sec and XXX ops/sec) *** done
Signature · 2 lines
Best regards, Vladimir