OpenZFS 9166 - zfs storage pool checkpoint #7570

dweeezil · 2018-05-26T18:51:27Z

Description

See references below.

Motivation and Context

Details about the motivation of this feature and its usage can
be found in this blogpost:

https://sdimitro.github.io/post/zpool-checkpoint/

A lightning talk of this feature can be found here:
https://www.youtube.com/watch?v=fPQA8K40jAM

Implementation details can be found in big block comment of
spa_checkpoint.c

Side-changes that are relevant to this commit but not explained
elsewhere:

renames metaslab trees to be shorter without losing meaning
space_map_{alloc,truncate}() accept a block size as a
parameter. The reason is that in the current state all space
maps that we allocate through the DMU use a global tunable
(space_map_blksz) which defauls to 4KB. This is ok for
metaslab space maps in terms of bandwirdth since they are
scattered all over the disk. But for other space maps this
default is probably not what we want. Examples are device
removal's vdev_obsolete_sm or vdev_chedkpoint_sm from this
review. Both of these have a 1:1 relationship with each vdev
and could benefit from a bigger block size.

Authored by: Serapheim Dimitropoulos serapheim.dimitro@delphix.com
Reviewed by: Matthew Ahrens mahrens@delphix.com
Reviewed by: John Kennedy john.kennedy@delphix.com
Reviewed by: Dan Kimmel dan.kimmel@delphix.com
Approved by: Richard Lowe richlowe@richlowe.net
OpenZFS-issue: https://illumos.org/issues/9166
OpenZFS-commit: openzfs/openzfs@7159fdb
Ported-by: Tim Chase tim@chase2k.com
Signed-off-by: Tim Chase tim@chase2k.com

Porting notes:

The part of dsl_scan_sync() which handles async destroys has
been moved into the new dsl_process_async_destroys() function.

Renamed module parameter metaslabs_per_vdev to vdev_max_ms_count.

New module parameter:

    zfs_spa_discard_memory_limit,

Desirable administrative feature, particularly to allow reverting pool configuration changes.

How Has This Been Tested?

Light manual testing so far.

Types of changes

Bug fix (non-breaking change which fixes an issue)
New feature (non-breaking change which adds functionality)
Performance enhancement (non-breaking change which improves efficiency)
Code cleanup (non-breaking change which makes code smaller or more readable)
Breaking change (fix or feature that would cause existing functionality to change)
Documentation (a change to man pages or other documentation)

Checklist:

My code follows the ZFS on Linux code style requirements.
I have updated the documentation accordingly.
I have read the contributing document.
I have added tests to cover my changes.
All new and existing tests passed.
All commit messages are properly formatted and contain Signed-off-by.
Change has been approved by a ZFS on Linux member.

dweeezil · 2018-05-26T18:53:17Z

I've only done light manual testing of this feature and it seems to work properly. Linux-specific changes to the ZTS have not yet been made. It's being submitted now to get an initial run through the buildbots to find any gross problems.

sdimitro

Hello there!

Thank you for working on this.
Overall the changes look good.
I also went through the split in the dsl_scan() code that you brought up and they seem fine too.

I found a couple of small omissions that I pointed out below and I also raised some questions (most of them are probably due to my lack of knowledge of Linux dev).

sdimitro · 2018-06-01T17:57:19Z

cmd/ztest/ztest.c

@@ -397,6 +398,8 @@ ztest_info_t ztest_info[] = {
 	ZTI_INIT(ztest_vdev_aux_add_remove, 1, &ztest_opts.zo_vdevtime),
 	ZTI_INIT(ztest_device_removal, 1, &zopt_sometimes),
 	ZTI_INIT(ztest_remap_blocks, 1, &zopt_sometimes),
+	ZTI_INIT(ztest_remap_blocks, 1, &zopt_sometimes),


no need to duplicate ZTI_INIT(ztest_remap_blocks....),

sdimitro · 2018-06-01T18:10:16Z

cmd/ztest/ztest.c

@@ -6977,6 +7074,7 @@ ztest_import(ztest_shared_t *zs)

 	(void) rwlock_destroy(&ztest_name_lock);
 	mutex_destroy(&ztest_vdev_lock);
+	mutex_destroy(&ztest_checkpoint_lock);


Please correct me if I'm missing something here but shouldn't we init the lock in this functions before we destroy it? (similarly to ztest_vdev_lock which is destroyed here but it is also initialized in this functions?)

sdimitro · 2018-06-01T18:11:45Z

cmd/ztest/ztest.c

@@ -6991,6 +7089,7 @@ ztest_init(ztest_shared_t *zs)
 	int i;

 	mutex_init(&ztest_vdev_lock, NULL, MUTEX_DEFAULT, NULL);
+	mutex_init(&ztest_checkpoint_lock, NULL, USYNC_THREAD, NULL);


Should we also destroy destroy this lock in this function the same way we destroy ztest_vdev_lock?

Fixed. These init/destroy mis-merges were due to ztest_import() which was added by the MMP work.

sdimitro · 2018-06-01T18:32:01Z

lib/libzpool/kernel.c

@@ -497,8 +497,6 @@ vn_open(char *path, int x1, int flags, int mode, vnode_t **vpp, int x2, int x3)
 #ifdef __linux__
 		flags |= O_DIRECT;
 #endif
-		/* We shouldn't be writing to block devices in userspace */
-		VERIFY(!(flags & FWRITE));


Out of curiosity, what is this for? Is this related somehow to the rest of the checkpoint changes?

This was an additional sanity check which was added a while back when we only expected zdb to access the block devices read-only from user space. When zhack was added that changed and this check wasn't dropped at the time. One of the new test cases uncovered this since it uses zhack on a pool with disks instead of files.

Makes sense. Thanks!

sdimitro · 2018-06-01T18:39:12Z

module/zfs/dsl_pool.c

@@ -574,6 +579,29 @@ dsl_pool_dirty_delta(dsl_pool_t *dp, int64_t delta)
 		cv_signal(&dp->dp_spaceavail_cv);
 }

+#ifdef ZFS_DEBUG


Is that #ifdef really necessary? This function is only called through an ASSERT.

It prevents an unused function warning in non-debug builds.

sdimitro · 2018-06-01T19:05:43Z

tests/zfs-tests/cmd/randwritecomp/randwritecomp.c

+ */
+#define	__EXTENSIONS__
+
+#include <libzfs_core.h>


I may be missing something but why is libzfs_core.h included here?

IIRC, it was to give access to the typedefs such as uint64_t, etc.

Wouldn't something like stdint.h work?
http://pubs.opengroup.org/onlinepubs/009695399/basedefs/stdint.h.html

It just seems unnecessary to me to include a whole new dependency (libzfs_core) for a small utility like randwritecomp.

Yes indeed. Using libzfs_core.h was expedient at the time. Using stdint.h works perfectly well and also allows to remove -I$(top_srcdir)/lib/libspl/include from Makefile.am. I'll be updating it accordingly (and also adding an include of string.h so grab the declaration of strcmp() (which we apparently need under Linux).

That's great! Thank you Tim

sdimitro · 2018-06-01T19:18:32Z

tests/zfs-tests/tests/functional/pool_checkpoint/pool_checkpoint.kshlib

+	log_must truncate -s $DISKSIZE $FILEDISK1
+	log_must truncate -s $DISKSIZE $FILEDISK2
+
+	log_must zpool create -O sync=disabled $NESTEDPOOL $FILEDISKS


That's probably some illumos/Linux difference that I'm missing but why is sync disabled for these pools?

This was an attempt to speed up the "setup" process, which all by itself was taking almost 10 minutes per test in a VM guest. I've not yet determined whether the sync=disabled makes any difference. I added it on both the outer and the nested pools. Speaking of which, are the nested pools really necessary? It was my understanding (and @behlendorf's too, I think) that nested pools weren't guaranteed to be deadlock-free (but maybe they are in illumos).

behlendorf · 2018-06-01T21:26:37Z

tests/zfs-tests/cmd/randwritecomp/Makefile.am

+	-I$(top_srcdir)/lib/libspl/include
+
+pkgexec_PROGRAMS = randwritecomp
+mktree_SOURCES = randwritecomp.c


s/mktree_SOURCES/randwritecomp_SOURCES

Oops, copy/paste bug. Fixed.

behlendorf · 2018-06-01T21:44:48Z

module/zfs/spa_checkpoint.c

+
+#ifdef DEBUG
+	spa_checkpoint_accounting_verify(vd->vdev_spa);
+#endif


Results in a 'spa_checkpoint_accounting_verify' defined but not used [-Wunused-function] warning.

I'd suggest wrapping the function itself with ZFS_DEBUG. You could also

#define spa_checkpoint_accounting_verify ((void) 0)

for the non-debug case to get rid on the unsightly DEBUG wrapper where it's called.

Personally I find this way least ugly:

/* ARGSUSED */ void foo_verify(...) { #ifdef ZFS_DEBUG ... #endif }

Changed to ZFS_DEBUG. Also wrapped the definition of spa_checkpoint_accounting_verify() in an #ifdef ZFS_DEBUG.

behlendorf · 2018-06-01T21:52:44Z

lib/libzpool/kernel.c

@@ -497,8 +497,6 @@ vn_open(char *path, int x1, int flags, int mode, vnode_t **vpp, int x2, int x3)
 #ifdef __linux__
 		flags |= O_DIRECT;
 #endif
-		/* We shouldn't be writing to block devices in userspace */
-		VERIFY(!(flags & FWRITE));


This was an additional sanity check which was added a while back when we only expected zdb to access the block devices read-only from user space. When zhack was added that changed and this check wasn't dropped at the time. One of the new test cases uncovered this since it uses zhack on a pool with disks instead of files.

behlendorf · 2018-06-01T21:58:53Z

tests/zfs-tests/tests/functional/pool_checkpoint/setup.ksh

@@ -0,0 +1,25 @@
+#!/usr/bin/ksh -p


This should be #!/bin/ksh for all the new test scripts. It was causing the CentOS 6 failures.

Yes! I saw that in one of the tests introduced by vdev removal and totally forgot to look for it.

codecov · 2018-06-14T06:56:44Z

Codecov Report

Merging #7570 into master will increase coverage by 0.07%.
The diff coverage is 89.11%.

@@            Coverage Diff             @@
##           master    #7570      +/-   ##
==========================================
+ Coverage   78.04%   78.12%   +0.07%     
==========================================
  Files         366      368       +2     
  Lines      110618   111627    +1009     
==========================================
+ Hits        86329    87205     +876     
- Misses      24289    24422     +133

Flag	Coverage Δ
#kernel	`78.68% <92.94%> (+0.21%)`	⬆️
#user	`67.05% <74.78%> (+0.09%)`	⬆️

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 7637ef8...5dd9c73. Read the comment docs.

Otherwise the output is consumed by the output redirection. Signed-off-by: Tim Chase <tim@chase2k.com> Requires-builders: none

Details about the motivation of this feature and its usage can be found in this blogpost: https://sdimitro.github.io/post/zpool-checkpoint/ A lightning talk of this feature can be found here: https://www.youtube.com/watch?v=fPQA8K40jAM Implementation details can be found in big block comment of spa_checkpoint.c Side-changes that are relevant to this commit but not explained elsewhere: * renames members of "struct metaslab trees to be shorter without losing meaning * space_map_{alloc,truncate}() accept a block size as a parameter. The reason is that in the current state all space maps that we allocate through the DMU use a global tunable (space_map_blksz) which defauls to 4KB. This is ok for metaslab space maps in terms of bandwirdth since they are scattered all over the disk. But for other space maps this default is probably not what we want. Examples are device removal's vdev_obsolete_sm or vdev_chedkpoint_sm from this review. Both of these have a 1:1 relationship with each vdev and could benefit from a bigger block size. Authored by: Serapheim Dimitropoulos <serapheim.dimitro@delphix.com> Reviewed by: Matthew Ahrens <mahrens@delphix.com> Reviewed by: John Kennedy <john.kennedy@delphix.com> Reviewed by: Dan Kimmel <dan.kimmel@delphix.com> Approved by: Richard Lowe <richlowe@richlowe.net> OpenZFS-issue: https://illumos.org/issues/9166 OpenZFS-commit: openzfs/openzfs@7159fdb Ported-by: Tim Chase <tim@chase2k.com> Signed-off-by: Tim Chase <tim@chase2k.com> Porting notes: The part of dsl_scan_sync() which handles async destroys has been moved into the new dsl_process_async_destroys() function. Remove "VERIFY(!(flags & FWRITE))" in "kernel.c" so zhack can write to block device backed pools. ZTS: Fix get_txg() in zpool_sync_001_pos due to "checkpoint_txg". Don't use large dd block sizes on /dev/urandom under Linux in checkpoint_capacity. Adopt Delphix-OS's setting of 4 (spa_asize_inflation = SPA_DVAS_PER_BP + 1) for the checkpoint_capacity test to speed its attempts to fill the pool Create the base and nested pools with sync=disabled to speed up the "setup" phase. Clear labels in test pool between checkpoint tests to avoid duplicate pool issues. The import_rewind_device_replaced test has been marked as "known to fail" for the reasons listed in its DISCLAIMER. New module parameters: zfs_spa_discard_memory_limit, zfs_remove_max_bytes_pause (not documented - debugging only) vdev_max_ms_count (formerly metaslabs_per_vdev) vdev_min_ms_count

Details about the motivation of this feature and its usage can be found in this blogpost: https://sdimitro.github.io/post/zpool-checkpoint/ A lightning talk of this feature can be found here: https://www.youtube.com/watch?v=fPQA8K40jAM Implementation details can be found in big block comment of spa_checkpoint.c Side-changes that are relevant to this commit but not explained elsewhere: * renames members of "struct metaslab trees to be shorter without losing meaning * space_map_{alloc,truncate}() accept a block size as a parameter. The reason is that in the current state all space maps that we allocate through the DMU use a global tunable (space_map_blksz) which defauls to 4KB. This is ok for metaslab space maps in terms of bandwirdth since they are scattered all over the disk. But for other space maps this default is probably not what we want. Examples are device removal's vdev_obsolete_sm or vdev_chedkpoint_sm from this review. Both of these have a 1:1 relationship with each vdev and could benefit from a bigger block size. Porting notes: * The part of dsl_scan_sync() which handles async destroys has been moved into the new dsl_process_async_destroys() function. * Remove "VERIFY(!(flags & FWRITE))" in "kernel.c" so zhack can write to block device backed pools. * ZTS: * Fix get_txg() in zpool_sync_001_pos due to "checkpoint_txg". * Don't use large dd block sizes on /dev/urandom under Linux in checkpoint_capacity. * Adopt Delphix-OS's setting of 4 (spa_asize_inflation = SPA_DVAS_PER_BP + 1) for the checkpoint_capacity test to speed its attempts to fill the pool * Create the base and nested pools with sync=disabled to speed up the "setup" phase. * Clear labels in test pool between checkpoint tests to avoid duplicate pool issues. * The import_rewind_device_replaced test has been marked as "known to fail" for the reasons listed in its DISCLAIMER. * New module parameters: zfs_spa_discard_memory_limit, zfs_remove_max_bytes_pause (not documented - debugging only) vdev_max_ms_count (formerly metaslabs_per_vdev) vdev_min_ms_count Authored by: Serapheim Dimitropoulos <serapheim.dimitro@delphix.com> Reviewed by: Matthew Ahrens <mahrens@delphix.com> Reviewed by: John Kennedy <john.kennedy@delphix.com> Reviewed by: Dan Kimmel <dan.kimmel@delphix.com> Reviewed by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Richard Lowe <richlowe@richlowe.net> Ported-by: Tim Chase <tim@chase2k.com> Signed-off-by: Tim Chase <tim@chase2k.com> OpenZFS-issue: https://illumos.org/issues/9166 OpenZFS-commit: openzfs/openzfs@7159fdb8 Closes #7570

dweeezil force-pushed the illumos-9166 branch 8 times, most recently from 1399feb to 68ff1ff Compare May 30, 2018 16:05

ahrens requested a review from sdimitro May 30, 2018 17:08

dweeezil force-pushed the illumos-9166 branch 2 times, most recently from 1615849 to 7db8ffc Compare June 1, 2018 03:11

sdimitro reviewed Jun 1, 2018

View reviewed changes

behlendorf requested changes Jun 1, 2018

View reviewed changes

dweeezil force-pushed the illumos-9166 branch 4 times, most recently from df76923 to bbbd614 Compare June 7, 2018 03:54

ahrens added the Type: Feature Feature request or new feature label Jun 7, 2018

dweeezil force-pushed the illumos-9166 branch 3 times, most recently from ef59091 to 3b69b11 Compare June 13, 2018 22:44

dweeezil force-pushed the illumos-9166 branch 3 times, most recently from 5f0c492 to cddfe80 Compare June 19, 2018 14:17

Use "eval" in history_002_pos for log_must

0d44335

Otherwise the output is consumed by the output redirection. Signed-off-by: Tim Chase <tim@chase2k.com> Requires-builders: none

dweeezil force-pushed the illumos-9166 branch 3 times, most recently from 97c4672 to 3397340 Compare June 24, 2018 20:32

dweeezil force-pushed the illumos-9166 branch from 3397340 to 500fe38 Compare June 24, 2018 22:56

dweeezil force-pushed the illumos-9166 branch from 500fe38 to 5dd9c73 Compare June 25, 2018 01:12

behlendorf approved these changes Jun 25, 2018

View reviewed changes

sdimitro approved these changes Jun 26, 2018

View reviewed changes

behlendorf closed this in 88eaf61 Jun 26, 2018

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

OpenZFS 9166 - zfs storage pool checkpoint #7570

OpenZFS 9166 - zfs storage pool checkpoint #7570

dweeezil commented May 26, 2018

dweeezil commented May 26, 2018

sdimitro left a comment

sdimitro Jun 1, 2018

dweeezil Jun 2, 2018

sdimitro Jun 1, 2018

dweeezil Jun 2, 2018

sdimitro Jun 1, 2018

dweeezil Jun 2, 2018 •

edited

Loading

sdimitro Jun 1, 2018

behlendorf Jun 1, 2018

sdimitro Jun 1, 2018

sdimitro Jun 1, 2018

dweeezil Jun 2, 2018

sdimitro Jun 1, 2018

dweeezil Jun 2, 2018

sdimitro Jun 4, 2018

dweeezil Jun 4, 2018

sdimitro Jun 4, 2018

sdimitro Jun 1, 2018

dweeezil Jun 2, 2018

behlendorf Jun 1, 2018

dweeezil Jun 2, 2018

behlendorf Jun 1, 2018

ahrens Jun 1, 2018

dweeezil Jun 2, 2018

behlendorf Jun 1, 2018

behlendorf Jun 1, 2018

dweeezil Jun 2, 2018

codecov bot commented Jun 14, 2018 •

edited

Loading

OpenZFS 9166 - zfs storage pool checkpoint #7570

OpenZFS 9166 - zfs storage pool checkpoint #7570

Conversation

dweeezil commented May 26, 2018

Description

Motivation and Context

How Has This Been Tested?

Types of changes

Checklist:

dweeezil commented May 26, 2018

sdimitro left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

dweeezil Jun 2, 2018 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

codecov bot commented Jun 14, 2018 • edited Loading

Codecov Report

dweeezil Jun 2, 2018 •

edited

Loading

codecov bot commented Jun 14, 2018 •

edited

Loading