Re: [PATCH] repack: respect gc.pid lock
- From
Jeff King <peff@peff.net>
- Date
- Apr 18, 2017, 17:16 UTC
- Message-ID
- <20170418171646.5a5mjhd4qjr6ot7d@sigill.intra.peff.net>
- In-Reply-To
- <d5c43adf0b074c6ebe43439bc3fc7539@exmbdft7.ad.twosigma.com>
On Tue, Apr 18, 2017 at 05:08:14PM +0000, David Turner wrote:
> On 64-bit systems, I think core.packedGitLimit doesn't make a > lot of sense. There is plenty of address space. Why not use it?
That's my gut feeling, too. I'd have a slight worry that the OS's paging behavior may respond differently if we have more memory mapped. But that's not based on numbers, just a fear of the unknown. :)
If we have infinite windows anyway, I suspect we could just mmap entire packfiles and forget about all the window complexity in the first place. Although IIRC some operating systems take a long time to set up large mmaps, and we may only need a small part of a large pack.
> I'll ask our git server administrator to adjust core.packedGitLimit > and turn repacks back on to see if that fixes the issue.
Thanks. Let us know if you get any results, either way.
-Peff