btrfs: zoned: enable metadata over-commit for non-ZNS setup
authorNaohiro Aota <naohiro.aota@wdc.com>
Tue, 10 Jan 2023 06:04:32 +0000 (15:04 +0900)
committerDavid Sterba <dsterba@suse.com>
Wed, 11 Jan 2023 19:04:25 +0000 (20:04 +0100)
The commit 79417d040f4f ("btrfs: zoned: disable metadata overcommit for
zoned") disabled the metadata over-commit to track active zones properly.

However, it also introduced a heavy overhead by allocating new metadata
block groups and/or flushing dirty buffers to release the space
reservations. Specifically, a workload (write only without any sync
operations) worsen its performance from 343.77 MB/sec (v5.19) to 182.89
MB/sec (v6.0).

The performance is still bad on current misc-next which is 187.95 MB/sec.
And, with this patch applied, it improves back to 326.70 MB/sec (+73.82%).

This patch introduces a new fs_info->flag BTRFS_FS_NO_OVERCOMMIT to
indicate it needs to disable the metadata over-commit. The flag is enabled
when a device with max active zones limit is loaded into a file-system.

Fixes: 79417d040f4f ("btrfs: zoned: disable metadata overcommit for zoned")
CC: stable@vger.kernel.org # 6.0+
Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Signed-off-by: Naohiro Aota <naohiro.aota@wdc.com>
Signed-off-by: David Sterba <dsterba@suse.com>
fs/btrfs/fs.h
fs/btrfs/space-info.c
fs/btrfs/zoned.c

index a749367e5ae2a2b9090a44ec8c0dfc8536f92131..37b86acfcbcf880ded4a2fbed2407a0e6f905982 100644 (file)
@@ -119,6 +119,12 @@ enum {
        /* Indicate that we want to commit the transaction. */
        BTRFS_FS_NEED_TRANS_COMMIT,
 
+       /*
+        * Indicate metadata over-commit is disabled. This is set when active
+        * zone tracking is needed.
+        */
+       BTRFS_FS_NO_OVERCOMMIT,
+
 #if BITS_PER_LONG == 32
        /* Indicate if we have error/warn message printed on 32bit systems */
        BTRFS_FS_32BIT_ERROR,
index d28ee4e36f3d907e7ce9f8354e56a9f1810be2fc..69c09508afb506ac8121caeae39fb2e7071362c0 100644 (file)
@@ -407,7 +407,8 @@ int btrfs_can_overcommit(struct btrfs_fs_info *fs_info,
                return 0;
 
        used = btrfs_space_info_used(space_info, true);
-       if (btrfs_is_zoned(fs_info) && (space_info->flags & BTRFS_BLOCK_GROUP_METADATA))
+       if (test_bit(BTRFS_FS_NO_OVERCOMMIT, &fs_info->flags) &&
+           (space_info->flags & BTRFS_BLOCK_GROUP_METADATA))
                avail = 0;
        else
                avail = calc_available_free_space(fs_info, space_info, flush);
index a759668477bb2e10eee5f6b60fb03463cd81fcf8..1f503e8e42d483a765b71b9b5910fce0ede0ea6f 100644 (file)
@@ -539,6 +539,8 @@ int btrfs_get_dev_zone_info(struct btrfs_device *device, bool populate_cache)
                }
                atomic_set(&zone_info->active_zones_left,
                           max_active_zones - nactive);
+               /* Overcommit does not work well with active zone tacking. */
+               set_bit(BTRFS_FS_NO_OVERCOMMIT, &fs_info->flags);
        }
 
        /* Validate superblock log */