bcachefs format --discard scope

[jpsollie]

The--discard parameter scope does not start at the begin of a new group:

linuxserver /home/janpieter # bcachefs format -g SSD --discard /dev/disk/by-partuuid/f8d16ed7-eddc-4ef1-84b7-bc3b74e8b3ca -g HDD /dev/sdd --foreground_target=SSD --promote_target=SSD --background_target=HDD                        External UUID:4ad9b6a6-4911-4b9c-9d05-3c4091e973b4
Internal UUID:df5a0103-6c83-4193-abf3-9602ca9e1b96
Label:
Version:11
Created:Sat Jun 13 07:21:38 2020
Block_size:512
Btree node size:256.0K
Error action:remount-ro
Clean:0
Features:
Metadata replicas:1
Data replicas:1
Metadata checksum type:crc32c (1)
Data checksum type:crc32c (1)
Compression type:none (0)
Foreground write target:Group 0 (SSD)
Background write target:Group 1 (HDD)
Promote target:Group 0 (SSD)
String hash type:siphash (2)
32 bit inodes:0
GC reserve percentage:8%
Root reserve percentage:0%
Devices:2 live, 2 total
Sections:members,disk_groups
Superblock size:976

Members (size 120):
  Device 0:
    UUID:9c41e5d3-979a-467d-b9f5-ac0a31e1d57a
    Size:107.3G
    Bucket size:256.0K
    First bucket:0
    Buckets:439556
    Last mount:(never)
    State:readwrite
    Group:SSD (0)
    Data allowed:journal,btree,data
    Has data:(none)
    Replacement policy:lru
    Discard:1
  Device 1:
    UUID:c2f801f4-b323-4947-bf0b-9921ba482b39
    Size:3.6T
    Bucket size:512.0K
    First bucket:0
    Buckets:7618540
    Last mount:(never)
    State:readwrite
    Group:HDD (1)
    Data allowed:journal,btree,data
    Has data:(none)
    Replacement policy:lru
    Discard:1
initializing new filesystem
mounted with opts: foreground_target=SSD,background_target=HDD,promote_target=SSD
linuxserver /home/janpieter # bcachefs format -g HDD /dev/sdd -g SSD --discard /dev/disk/by-partuuid/f8d16ed7-eddc-4ef1-84b7-bc3b74e8b3ca --foreground_target=SSD --promote_target=SSD --background_target=HDD
External UUID:5044a9f4-27e4-4b73-bb89-4e104ab370eb
Internal UUID:b916dfed-a454-4d55-ac63-c79733474949
Label:
Version:11
Created:Sat Jun 13 07:23:34 2020
Block_size:512
Btree node size:256.0K
Error action:remount-ro
Clean:0
Features:
Metadata replicas:1
Data replicas:1
Metadata checksum type:crc32c (1)
Data checksum type:crc32c (1)
Compression type:none (0)
Foreground write target:Group 1 (SSD)
Background write target:Group 0 (HDD)
Promote target:Group 1 (SSD)
String hash type:siphash (2)
32 bit inodes:0
GC reserve percentage:8%
Root reserve percentage:0%
Devices:2 live, 2 total
Sections:members,disk_groups
Superblock size:976

Members (size 120):
  Device 0:
    UUID:bcce3afc-d31b-49d5-b9f0-bb3231e6431f
    Size:3.6T
    Bucket size:512.0K
    First bucket:0
    Buckets:7618540
    Last mount:(never)
    State:readwrite
    Group:HDD (0)
    Data allowed:journal,btree,data
    Has data:(none)
    Replacement policy:lru
    Discard:0
  Device 1:
    UUID:000c018b-05a0-46e1-b535-e04ae00f70b6
    Size:107.3G
    Bucket size:256.0K
    First bucket:0
    Buckets:439556
    Last mount:(never)
    State:readwrite
    Group:SSD (1)
    Data allowed:journal,btree,data
    Has data:(none)
    Replacement policy:lru
    Discard:1
initializing new filesystem
mounted with opts: foreground_target=SSD,background_target=HDD,promote_target=SSD
linuxserver /home/janpieter #

As a result, all SSD devices have to be specified as the final ones, otherwise the discard flag is taken over by the HDD group.

[koverstreet]

I'm not seeing the problem, the way you want to do it seems more confusing to me. Discard, like all other per device options (-g included!) are scoped to devices that come after that option

[jpsollie]

Discard, like all other per device options (-g included!) are scoped to devices that come after that option

Exactly. When I create group 1, enable discard, and then create group 2, the discard option is promoted to group 2 while I only wanted it for group 1. This is what issue #33 is all about. When we need to consider resetting device options when a group is created, it may be useful to create a feature where you have a global flag when you do not include it in a group. The logic behind "is a global flag" (and thus does not need to be reset for each group) is somewhat confusing, but that's more or less what I'm doing here

[koverstreet]

so let's just add a --nodiscard option if you want discards off for group 2...

[koverstreet]

I think you might be confusing yourself by thinking of the discard option as applying to a group. it doesn't, it applies to a disk.

[jpsollie]

I think you might be confusing yourself by thinking of the discard option as applying to a group. it doesn't, it applies to a disk.

now I am confused indeed: I usually format like --group=ssd --discard <dev1> <dev2> .... --group=hdd <dev3><dev4> ... so the discard option should apply only for dev1?

[koverstreet]

No, it applies to all devices that come after it, until another option overrides (which won't happen since we don't have --nodiscard) - it'll apply to all devices in your example.

On Tue, Jun 30, 2020 at 2:56 PM jpsollie [email protected] wrote:

I think you might be confusing yourself by thinking of the discard option as applying to a group. it doesn't, it applies to a disk.

now I am confused indeed: I usually format like --group=ssd --discard .... --group=hdd ... so the discard option should apply only for dev1?

— You are receiving this because you commented. Reply to this email directly, view it on GitHub https://github.com/koverstreet/bcachefs-tools/issues/33#issuecomment-651981405, or unsubscribe https://github.com/notifications/unsubscribe-auth/AAPGX3WEK6CZJUURUFPJ3N3RZIYOBANCNFSM4N42XNZQ .

[koverstreet]

ok, now rebase and squash your fixup commits so I can see what's actually going on... I don't want to change the discard option and make it inconsistent, but I'll take the other stuff

[jpsollie]

Here you go, I hope it's clear to you