Re: RFC/Pull Request: Refs db backend

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

 



On Wed, 2015-06-24 at 05:14 -0400, Jeff King wrote:
> On Tue, Jun 23, 2015 at 02:18:36PM -0400, David Turner wrote:
> 
> > > Can you describe a bit more about the reflog handling?
> > > 
> > > One of the problems we've had with large-ref repos is that the reflog
> > > storage is quite inefficient. You can pack all the refs, but you may
> > > still be stuck with a bunch of reflog files with one entry, wasting a
> > > whole inode. Doing a "git repack" when you have a million of those has
> > > horrible cold-cache performance. Basically anything that isn't
> > > one-file-per-reflog would be a welcome change. :)
> > 
> > Reflogs are stored in the database as well.  There is one header entry
> > per ref to indicate that a reflog is present, and then one database
> > entry per reflog entry; the entries are stored consecutively and
> > immediately following the header so that it's fast to iterate over them.
> 
> OK, that make sense. I did notice that the storage for the refdb grows
> rapidly. If I add a millions refs (like refs/tags/$i) with a simple
> reflog message "foo", I ended up with a 500MB database file.
> 
> That's _probably_ OK, because a million is getting into crazy
> territory[1].  But it's 500 bytes per ref, each with one reflog entry.
> Our ideal lower bound is probably something like 100 bytes per reflog
> entry:
> 
>   - 20 bytes for old sha1
>   - 20 bytes for new sha1
>   - ~50 bytes for name, email, timestamp
>   - ~6 bytes for refname (1000000 is the longest unique part)
> 
> That assumes we store binary[2] (and not just the raw reflog lines), and
> reconstruct the reflog lines on the fly. It also assumes we use some
> kind of trie-like storage (where we can amortize the cost of storing
> "refs/tags/" across all of the entries).
> 
> Of course that neglects lmdb's overhead, and the storage of the ref tip
> itself. But it would hopefully give us a ballpark for an optimal
> solution. We don't have to hit that, of course, but it's food for
> thought.
> 
> [1] The homebrew/homebrew repository on GitHub has almost half a million
>     ref updates. Since this is storing not just refs but all ref
>     updates, that's actually the interesting number (and optimizing the
>     per-reflog-entry size is more interesting than the per-ref size).
> 
> [2] I'm hesitant to suggest binary formats in general, but given that
>     this is a blob embedded inside lmdb, I think it's OK. If we were to
>     pursue the log-structured idea I suggested earlier, I'm torn on
>     whether it should be binary or not.

I could try a binary format.  I was optimizing for simplicity,
debuggability, recoverability, compatibility with the choice of the text
format, but I wouldn't have to.  I don't know how much this will save.
Unfortunately, given the way LMDB works, a trie-like storage to save
refs/tags does not seem possible (of course, we could hard-code some
hacks like \001=refs/rags, \002=refs/heads, etc but that is a
micro-optimization that might not be worth it.

Also, the reflog header has some overhead (it's an entire extra record
per ref). The header exists to implement reflog creation/existence
checking.  I didn't really try to understand why we have the distinction
between empty and nonexistent reflogs; I just copied it.  If we didn't
have that distinction, we could eliminate that overhead.

> > Thanks, that's valuable.  For the refs backend, opening the LMDB
> > database for writing is sufficient to block other writers.  Do you think
> > it would be valuable to provide a git hold-ref-lock command that simply
> > reads refs from stdin and keeps them locked until it reads EOF from
> > stdin?  That would allow cross-backend ref locking.
> 
> I'm not sure what you would use it for. If you want to update the refs,
> then you can specify a whole transaction with "git update-ref --stdin",
> and that should work whatever backend you choose. Is there some other
> operation you want where you hold the lock for a longer period of time?

I'm sure I had a reason for this at the time I wrote it, but now I can't
think of what it was.  Nevermind!

--
To unsubscribe from this list: send the line "unsubscribe git" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html



[Index of Archives]     [Linux Kernel Development]     [Gcc Help]     [IETF Annouce]     [DCCP]     [Netdev]     [Networking]     [Security]     [V4L]     [Bugtraq]     [Yosemite]     [MIPS Linux]     [ARM Linux]     [Linux Security]     [Linux RAID]     [Linux SCSI]     [Fedora Users]