bzr-git, the git map, fetching revisions, and being very slow

[Russel Winder]

Jelmer,

On Sat, 2009-11-21 at 14:58 -0600, Jelmer Vernooij wrote:
> On Sat, 2009-11-21 at 11:24 +0000, Russel Winder wrote: 
> > Updating the git map, as happens after taking a branch of a Bazaar
> > branch bound to a Git repository which is then bound to the same Git
> > repository and updated, takes ages.  We cannot blame NFS and TDB here as
> > this is on the machine with the real discs locally.  It has taken about
> > 45mins to update the git map for 1749 revisions.
> > 
> > It gets worse :-(  it then fetched 38 revisions and this took about 50
> > mins.
> Please file a bug about this. We should probably get bzr to provide a
> hook that can be used to copy the sha map when cloning a bzr branch. 

Done:  https://bugs.launchpad.net/bzr-git/+bug/486570

Clearly more experimentation is needed to get proper data to investigate
this.  If you let me know what to do and what data to gather, I can have
a go.

> > This seems sadly slow compared to Git.
> It's unfair to compare imports from git to bazaar with native clones
> between git repositories. Compare with pulling from a git repository to
> a git repository using bzr instead.

Yes and no.  Clearly from a science perspective all three scenarios are
different with different variables and hence not directly comparable.
However, as Stephen has already pointed out, this isn't science, this is
user functionality and perception.  It is therefore all about personal
reaction and the knock on effect that has on social perception.  The
user goal is to interact with a Git repository.  Current options are use
a Git clone or use a Bazaar branch. Major factors here are branch model
and speed.  From a user perspective these are what matter, correct
science is not an issue.

For those of us that prefer the Bazaar model of branches to the Git one,
bzr-git is a great boon.  Such users will forgive a lot of performance
issues as long as they don't get too great.  Taking 100min to do what
Git does in <10s is beyond the point of pain.

This is an anecdotal data point, and it is not science, but it is a user
abandoning bzr-git for git because of performance problems.

I will of course undertake any experiment asked of me, and I will try to
keep using Bazaar with bzr-git but with the TDB/NFS problems (*) and now
this, it is very hard to keep the faith.

(*) As we know this is a Linux mmap over NFS problem and nothing that
can be fixed by TDB or bzr-git -- and bzr-svn -- per se, but it is a
*huge* negative factor for me -- and I suspect anyone else using NFS
mounted home filestores.

-- 
Russel.
=============================================================================
Dr Russel Winder      Partner
                                            xmpp: russel at russel.org.uk
Concertant LLP        t: +44 20 7585 2200, +44 20 7193 9203
41 Buckmaster Road,   f: +44 8700 516 084   voip: sip:russel.winder at ekiga.net
London SW11 1EN, UK   m: +44 7770 465 077   skype: russel_winder
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 197 bytes
Desc: This is a digitally signed message part
Url : https://lists.ubuntu.com/archives/bazaar/attachments/20091122/3c873782/attachment.pgp