Re: [PATCH] Implement sharding for FSFS

Malcolm Rowe <malcolm-svn-dev@farside.org.uk> writes:

>Out of interest, how common is this? I know it was true for original
>Unices, but I thought (without any justification, I admit) that modern
>Unices had done away with it.

I just tried it on linux 2.6.16 (ext3) and it seems to cap the link count
at 32000, actually. Any mkdir attempt after that will result in an error.

>In any case, this only becomes a problem at c. 32 million revisions. I
>don't know of _any_ organisation that's even considering that size of
>repository at the moment... and when they are, they can switch to a
>shard-by-10,000 scheme, for example. The 1,000 figure is just the
>default.

Certainly, as long as it's possible to change shard size when the
limit comes near. And would a user get any indication at all that the
limit was reached, or just see the transaction fail in an unspecified way?

I agree that it's unlikely to ever be a problem, but for large-scale
use, who knows. It could also bother someone who is using Subversion
as a revision control back-end in some application where commits are
very frequent.

---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org
Received on Mon Apr 16 17:40:09 2007