As a follow-up to various fsync topics in-flight I've been encouraging
those involved to come up with some way to configure fsync() in a way
that'll make holistic sense in the end-state.
Continuing a discussion from [1] currently we have:
; Defaults to 'false'
core.fsyncObjectFiles = [true|false]
In master..next this has been extended to this by Neeraj:
core.fsyncObjectFiles = [true|false|batch]
Which, as an aside I hadn't considered before and I think we need to
change before it lands on "master", we really don't want config users
want to enable that makes older versions hard die. It's annoying to want
to configure a new thing and not being able to put it in .gitconfig
because older versions die on it:
$ git -c core.fsyncObjectFiles=batch status; echo $?
fatal: bad boolean config value 'batch' for 'core.fsyncobjectfiles'
128
Then there's Eric Wong's proposed[2]:
core.fsync = <bool>
And now Patrick Steinhardt has a proposal to extend Neeraj's with[3]:
; Like core.fsyncObjectFiles, but apparently for .git/refs, not
; .git/objects (but see my confusion on that topic in [1])
core.fsyncRefFiles = [<bool>|batch]
I think this sort of config schema would make everyone above happy
It would:
A) Be easy to extend for any future fsync behavior we'd reasonably
implement
B) Not make older git versions die. It's fine if they warn(), but not die.
C) Has some pretty contrived key names, but I'm trying to maintain the
constraint that you can set both fsck.X=Y and
e.g. fetch.fsck.X=Y. I.e. we should be able to configure things
globally *and* per-command, like color.*, fsck.* etc.
Proposal:
; Turns on/off all fsync, whatever the method is. I.e. allows you to
; never make any fsync() calls whatsoever (which we have another
; in-flight topic for).
; The "false" was controversial, and we could just leave it
; unimplemented
core.fsync = <bool>
; Optional, by default we'd use the most pedantic (I'd call our
; current "loose", whether we want to forward-support it is another
; matter.
;
; Whatever names we pick an option like this should ignore (or at most
; warn about) values it doesn't know about, not hard die on it.
;
; Here "bach" is what Neeraj and Patrick are pursuing, a hypothetical
; POSIX would be a pedantic way of exhaustively fsyncing everything.
;
; We'd leave door open to e.g. setting it to "linux:ext4" or whatever,
; to do only the work needed on some specific popular FS
core.fsyncMethod = loose | POSIX | batch | linux:ext4 | NTFS | ...
; Turn on or off entire categories of files we'd like to sync. This
; way Neeraj's and Patrick's approach would be to set
; core.fsyncMethod=batch, and then core.fsyncGroup=files &
; core.fsyncGroup=refs.
; If we learn about a new core.fsyncGroup = xyz in the future a <bool>
; in "core.fsyncGroupDefault" will prevail. I.e. if true it's
; included, if false not.
;
; Whether "false" or "true" is the default depends on
; core.fsyncMethod. For POSIX it would be true, for "loose" it's
; false.
core.fsyncGroup = files
core.fsyncGroup = refs
core.fsyncGroup = objects
I'm not sure I like calling it "group". Maybe "class", "category"? Doing
it with this structure is extensible to the two-level keys, as noted
above.
; Our existing config knob. When "false" synonymous with:
;
; core.fsync = true
; core.fsyncMethod = loose
; core.fsyncGroup = pack
;
; When "true" synonymous with the same as the above, plus:
; core.fsyncGroup = loose
;
: Or something like that. I.e. we'll fsync *.pack, *.bitmap etc, and ;
; probably some other stuff, but not loose objects etc.
;
; Whatever we fsync now exactly this schema should be generic enough
; to support it.
core.fsyncObjectFiles = <bool>
; A namespace for core.fsyncMethod = <X>. Specific methods will
; own this namespace and can configure whatever they want.
fsyncMethod.<x>.<a> = <b>
E.g. we might have:
fsyncMethod.POSIX.content = true
fsyncMethod.POSIX.metadata = false
If we know we'd like to (depending on other config) to fsync things
exhaustively or not, but do different things depending on file content
or metadata. I.e. maybe your FS's fsync() on a file fd always implies a
sync of the metadata, and maybe not.
; Change whatever fsync configuration you want per-command, similar to
; fsck.* and fetch.fsck.*
transfer.fsyncGroup=*
fetch.fsyncGroup=*
...
1. https://lore.kernel.org/git/211110.86v910gi9a.gmgdl@xxxxxxxxxxxxxxxxxxx/
2. https://lore.kernel.org/git/20211028002102.19384-1-e@xxxxxxxxx/
3. https://lore.kernel.org/git/cover.1636544377.git.ps@xxxxxx/
