[PATCH v3] btrfs: shrink delalloc pages instead of full inodes

[PATCH v3] btrfs: shrink delalloc pages instead of full inodes

[Josef Bacik]

Commit 38d715f494f2 ("btrfs: use btrfs_start_delalloc_roots in
shrink_delalloc") cleaned up how we do delalloc shrinking by utilizing
some infrastructure we have in place to flush inodes that we use for
device replace and snapshot.  However this introduced a pretty serious
performance regression.  To reproduce the user untarred the source
tarball of Firefox, and would see it take anywhere from 5 to 20 times as
long to untar in 5.10 compared to 5.9.

The root cause is because before we would generally use the normal
writeback path to reclaim delalloc space, and for this we would provide
it with the number of pages we wanted to flush.  The referenced commit
changed this to flush that many inodes, which drastically increased the
amount of space we were flushing in certain cases, which severely
affected performance.

We cannot revert this patch unfortunately because of

	btrfs: fix deadlock when cloning inline extent and low on free
		metadata space

which requires the ability to skip flushing inodes that are being cloned
in certain scenarios, which means we need to keep using our flushing
infrastructure or risk re-introducing the deadlock.

Instead to fix this problem we can go back to providing
btrfs_start_delalloc_roots with a number of pages to flush, and then set
up a writeback_control and utilize sync_inode() to handle the flushing
for us.  This gives us the same behavior we had prior to the fix, while
still allowing us to avoid the deadlock that was fixed by Filipe.  I
redid the users original test and got the following results on one of
our test machines (256gib of ram, 56 cores, 2tib Intel NVME drive)

5.9		0m54.258s
5.10		1m26.212s
5.10+patch	0m38.800s

5.10+patch is significantly faster than plain 5.9 because of my patch
series "Change data reservations to use the ticketing infra" which
contained the patch that introduced the regression, but generally
improved the overall ENOSPC flushing mechanisms.

CC: stable@vger.kernel.org # 5.10
Reported-by: René Rebe <rene@exactcode.de>
Fixes: 38d715f494f2 ("btrfs: use btrfs_start_delalloc_roots in shrink_delalloc")
Signed-off-by: Josef Bacik <josef@toxicpanda.com>
---
v2->v3:
- modified the changelog to add information about the patches referenced, and
  detail the specs of the machine I used for the performance numbers.

 fs/btrfs/inode.c      | 60 +++++++++++++++++++++++++++++++------------
 fs/btrfs/space-info.c |  4 ++-
 2 files changed, 46 insertions(+), 18 deletions(-)

diff --git a/fs/btrfs/inode.c b/fs/btrfs/inode.c
index 070716650df8..a8e0a6b038d3 100644
--- a/fs/btrfs/inode.c
+++ b/fs/btrfs/inode.c
@@ -9390,7 +9390,8 @@ static struct btrfs_delalloc_work *btrfs_alloc_delalloc_work(struct inode *inode
  * some fairly slow code that needs optimization. This walks the list
  * of all the inodes with pending delalloc and forces them to disk.
  */
-static int start_delalloc_inodes(struct btrfs_root *root, u64 *nr, bool snapshot,
+static int start_delalloc_inodes(struct btrfs_root *root,
+				 struct writeback_control *wbc, bool snapshot,
 				 bool in_reclaim_context)
 {
 	struct btrfs_inode *binode;
@@ -9399,6 +9400,7 @@ static int start_delalloc_inodes(struct btrfs_root *root, u64 *nr, bool snapshot
 	struct list_head works;
 	struct list_head splice;
 	int ret = 0;
+	bool full_flush = wbc->nr_to_write == LONG_MAX;
 
 	INIT_LIST_HEAD(&works);
 	INIT_LIST_HEAD(&splice);
@@ -9427,18 +9429,24 @@ static int start_delalloc_inodes(struct btrfs_root *root, u64 *nr, bool snapshot
 		if (snapshot)
 			set_bit(BTRFS_INODE_SNAPSHOT_FLUSH,
 				&binode->runtime_flags);
-		work = btrfs_alloc_delalloc_work(inode);
-		if (!work) {
-			iput(inode);
-			ret = -ENOMEM;
-			goto out;
-		}
-		list_add_tail(&work->list, &works);
-		btrfs_queue_work(root->fs_info->flush_workers,
-				 &work->work);
-		if (*nr != U64_MAX) {
-			(*nr)--;
-			if (*nr == 0)
+		if (full_flush) {
+			work = btrfs_alloc_delalloc_work(inode);
+			if (!work) {
+				iput(inode);
+				ret = -ENOMEM;
+				goto out;
+			}
+			list_add_tail(&work->list, &works);
+			btrfs_queue_work(root->fs_info->flush_workers,
+					 &work->work);
+		} else {
+			ret = sync_inode(inode, wbc);
+			if (!ret &&
+			    test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
+				     &BTRFS_I(inode)->runtime_flags))
+				ret = sync_inode(inode, wbc);
+			btrfs_add_delayed_iput(inode);
+			if (ret || wbc->nr_to_write <= 0)
 				goto out;
 		}
 		cond_resched();
@@ -9464,18 +9472,29 @@ static int start_delalloc_inodes(struct btrfs_root *root, u64 *nr, bool snapshot
 
 int btrfs_start_delalloc_snapshot(struct btrfs_root *root)
 {
+	struct writeback_control wbc = {
+		.nr_to_write = LONG_MAX,
+		.sync_mode = WB_SYNC_NONE,
+		.range_start = 0,
+		.range_end = LLONG_MAX,
+	};
 	struct btrfs_fs_info *fs_info = root->fs_info;
-	u64 nr = U64_MAX;
 
 	if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
 		return -EROFS;
 
-	return start_delalloc_inodes(root, &nr, true, false);
+	return start_delalloc_inodes(root, &wbc, true, false);
 }
 
 int btrfs_start_delalloc_roots(struct btrfs_fs_info *fs_info, u64 nr,
 			       bool in_reclaim_context)
 {
+	struct writeback_control wbc = {
+		.nr_to_write = (nr == U64_MAX) ? LONG_MAX : (unsigned long)nr,
+		.sync_mode = WB_SYNC_NONE,
+		.range_start = 0,
+		.range_end = LLONG_MAX,
+	};
 	struct btrfs_root *root;
 	struct list_head splice;
 	int ret;
@@ -9489,6 +9508,13 @@ int btrfs_start_delalloc_roots(struct btrfs_fs_info *fs_info, u64 nr,
 	spin_lock(&fs_info->delalloc_root_lock);
 	list_splice_init(&fs_info->delalloc_roots, &splice);
 	while (!list_empty(&splice) && nr) {
+		/*
+		 * Reset nr_to_write here so we know that we're doing a full
+		 * flush.
+		 */
+		if (nr == U64_MAX)
+			wbc.nr_to_write = LONG_MAX;
+
 		root = list_first_entry(&splice, struct btrfs_root,
 					delalloc_root);
 		root = btrfs_grab_root(root);
@@ -9497,9 +9523,9 @@ int btrfs_start_delalloc_roots(struct btrfs_fs_info *fs_info, u64 nr,
 			       &fs_info->delalloc_roots);
 		spin_unlock(&fs_info->delalloc_root_lock);
 
-		ret = start_delalloc_inodes(root, &nr, false, in_reclaim_context);
+		ret = start_delalloc_inodes(root, &wbc, false, in_reclaim_context);
 		btrfs_put_root(root);
-		if (ret < 0)
+		if (ret < 0 || wbc.nr_to_write <= 0)
 			goto out;
 		spin_lock(&fs_info->delalloc_root_lock);
 	}
diff --git a/fs/btrfs/space-info.c b/fs/btrfs/space-info.c
index 67e55c5479b8..e8347461c8dd 100644
--- a/fs/btrfs/space-info.c
+++ b/fs/btrfs/space-info.c
@@ -532,7 +532,9 @@ static void shrink_delalloc(struct btrfs_fs_info *fs_info,
 
 	loops = 0;
 	while ((delalloc_bytes || dio_bytes) && loops < 3) {
-		btrfs_start_delalloc_roots(fs_info, items, true);
+		u64 nr_pages = min(delalloc_bytes, to_reclaim) >> PAGE_SHIFT;
+
+		btrfs_start_delalloc_roots(fs_info, nr_pages, true);
 
 		loops++;
 		if (wait_ordered && !trans) {
-- 
2.26.2

Re: [PATCH v3] btrfs: shrink delalloc pages instead of full inodes

[David Sterba]

On Thu, Jan 07, 2021 at 05:08:30PM -0500, Josef Bacik wrote:
> Commit 38d715f494f2 ("btrfs: use btrfs_start_delalloc_roots in
> shrink_delalloc") cleaned up how we do delalloc shrinking by utilizing
> some infrastructure we have in place to flush inodes that we use for
> device replace and snapshot.  However this introduced a pretty serious
> performance regression.  To reproduce the user untarred the source
> tarball of Firefox, and would see it take anywhere from 5 to 20 times as
> long to untar in 5.10 compared to 5.9.
> 
> The root cause is because before we would generally use the normal
> writeback path to reclaim delalloc space, and for this we would provide
> it with the number of pages we wanted to flush.  The referenced commit
> changed this to flush that many inodes, which drastically increased the
> amount of space we were flushing in certain cases, which severely
> affected performance.
> 
> We cannot revert this patch unfortunately because of
> 
> 	btrfs: fix deadlock when cloning inline extent and low on free
> 		metadata space
> 
> which requires the ability to skip flushing inodes that are being cloned
> in certain scenarios, which means we need to keep using our flushing
> infrastructure or risk re-introducing the deadlock.
> 
> Instead to fix this problem we can go back to providing
> btrfs_start_delalloc_roots with a number of pages to flush, and then set
> up a writeback_control and utilize sync_inode() to handle the flushing
> for us.  This gives us the same behavior we had prior to the fix, while
> still allowing us to avoid the deadlock that was fixed by Filipe.  I
> redid the users original test and got the following results on one of
> our test machines (256gib of ram, 56 cores, 2tib Intel NVME drive)
> 
> 5.9		0m54.258s
> 5.10		1m26.212s
> 5.10+patch	0m38.800s
> 
> 5.10+patch is significantly faster than plain 5.9 because of my patch
> series "Change data reservations to use the ticketing infra" which
> contained the patch that introduced the regression, but generally
> improved the overall ENOSPC flushing mechanisms.
> 
> CC: stable@vger.kernel.org # 5.10
> Reported-by: René Rebe <rene@exactcode.de>
> Fixes: 38d715f494f2 ("btrfs: use btrfs_start_delalloc_roots in shrink_delalloc")
> Signed-off-by: Josef Bacik <josef@toxicpanda.com>
> ---
> v2->v3:
> - modified the changelog to add information about the patches referenced, and
>   detail the specs of the machine I used for the performance numbers.

Great, thanks. Meanwhile I did some other tests, 'dbench 32' is
basically the same and async random write with 'fio --rw=randwrite
--size=4g --ioengine=libaio' as well.

I'm going to send another rc3 pull request with this patch so we can get
it to 5.10 stable.