Re: [Monotone-devel] Re: speed of "mtn ls branches"

[William Uther]

On 19/01/2008, at 3:04 AM, Zack Weinberg wrote:

> On Jan 18, 2008 5:38 AM, Lapo Luchini <address@hidden> wrote:
> > We should maybe trade in some "security" with speed, e.g.  
> > mantaining a
> > DB table with a "cache of valid and not suspended branches".
>
> I'm seriously wondering whether we oughtn't to cache validity for
> *all* certs in the local database.  Or perhaps go even further and
> refuse to store "bogus" certs at all.
>
> Pro: would speed up just about everything.
> Con: we lose protection against database corruption and possibly
> against local attacks (anyone who can do "mtn db execute" can bork a
> certificate - I actually used this in a test case recently).
>
> Perhaps there is a middle ground, where we continue to do the
> checksum, but not the digital signature validation, except when we
> first hear about a cert.

I believe part of the reason we don't cache 'validity' is that it
actually means two things in the current system:
  i) The signature matches, and
  ii) The lua hook says that we want to trust that person.

We could certainly cache the first.  I think we could also cache the
second if we were careful about invalidating the cache when the lua
hooks change.

Cheers,

Will       :-}