Re: [PATCH] gc: do not warn about too many loose objects

[Elijah Newren <newren@gmail.com>]

On Mon, Jul 16, 2018 at 2:21 PM, Jeff King <peff@peff.net> wrote:
> On Mon, Jul 16, 2018 at 02:09:43PM -0700, Elijah Newren wrote:

>> The point of gc is to: expire reflogs, repack referenced objects, and
>> delete loose objects that (1) are no longer referenced and (2) are
>> "old enough".
>>
>> The "old enough" bit is the special part here.  Before the gc, those
>> objects were referenced (only by old reflog entries) and were found in
>> a pack.  git currently writes those objects out to disk and gives them
>> the age of the packfile they are contained in, which I think is the
>> source of the bug.  We have a reference for those objects from the
>> reflogs, know they aren't referenced anywhere else, so those objects
>> to me are the age of those reflog entries: 90 days.  As such, they are
>> "old enough" and should be deleted.
>
> OK, I see what you're saying, but...
>
>> I never got around to fixing it properly, though, because 'git prune'
>> is such a handy workaround that for now.  Having people nuke all their
>> loose objects is a bit risky, but the only other workaround people had
>> was to re-clone (waiting the requisite 20+ minutes for the repo to
>> download) and throw away their old clone.  (Though some people even
>> went that route, IIRC.)
>
> If we were to delete those objects, wouldn't it be exactly the same as
> running "git prune"? Or setting your gc.pruneExpire to "now" or even "5
> minutes"?  Or are you concerned with taking other objects along for the
> ride that weren't part of old reflogs? I think that's a valid concern,

Yes, I was worried about taking other objects along for the ride that
weren't part of old reflogs.

> but it's also an issue for objects which were previously referenced in
> a reflog, but are part of another current operation.

I'm not certain what you're referring to here.

> Also, what do you do if there weren't reflogs in the repo? Or the
> reflogs were deleted (e.g., because branch deletion drops the associated
> reflog entirely)?

Yes, there are issues this rule won't help with, but in my experience
it was a primary (if not sole) actual cause in practice.  (I believe I
even said elsewhere in this thread that I knew there were unreachable
objects for other reasons and they might also become large in number).
At $DAYJOB we've had multiple people including myself hit the "too
many unreachable loose objects" nasty loop issue (some of us multiple
different times), and as far as I can tell, most (perhaps even all) of
them would have been avoided by just "properly" deleting garbage as
per my object-age-is-reflog-age-if-not-otherwise-referenced rule.

>> With big repos, it's easy to get into situations where there are well
>> more than 10000 objects satisfying these conditions.  In fact, it's
>> not all that rare that the repo has far more loose objects after a git
>> gc than before.
>
> Yes, this is definitely a wart and I think is worth addressing.
>
>> I totally agree with your general plan to put unreferenced loose
>> objects into a pack.  However, I don't think these objects should be
>> part of that pack; they should just be deleted instead.
>
> I assume by "these objects" you mean ones which used to be reachable
> from a reflog, but that reflog entry just expired.  I think you'd be
> sacrificing some race-safety in that case.

Is that inherently any more race unsafe than 'git prune
--expire=2.weeks.ago'?  I thought it'd be racy in the same ways, and
thus a tradeoff folks are already accepting (at least implicitly) when
running git-gc.  Since these objects are actually 90 days old rather
than a mere two weeks, it actually seemed more safe to me.  But maybe
I'm overlooking something with the pack->loose transition that makes
it more racy?

> If the objects went into a pack under a race-proof scheme, would that
> actually bother you? Is it the 10,000 objects that's a problem, or is it
> the horrible I/O from exploding them coupled with the annoying auto-gc
> behavior?

Yeah, good point.  It's mostly the annoying auto-gc behavior and the
horrible I/O of future git operations from having lots of loose
objects.  They've already been paying the cost of storing those
objects in packed form for 90 days; a few more won't hurt much.  I'd
be slightly annoyed knowing that we're storing garbage we don't need
to be, but I don't think it's a real issue.