pack command...

[John Arbash Meinel]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

On 9/15/2010 11:24 AM, Michael Andronov wrote:
> I have a question about the pack command.
> 
> The manual says that  usually it should not be run manually.
> At the same time, it has helped me already a couple times to recover
> from unnecessary long times during bzr transfer operation.

We do trigger automatic repacking of some of the content based on some
heuristics. Approximately to keep ~log(N_revs) pack files around,
without having to touch all the data all the time.

The specific is to have pack files distributed roughly like the decimal
count. So if you have 2384 revisions, we would try to have 2 1000
revision packs, 3 100 revs, 8 10-revs and 4 1-rev. When you hit 2390,
then we pack the 10 1-rev packs into a single 10-rev pack, etc.

Running 'bzr pack' takes all 2390 and puts it into a single pack file.
Also, push + pull will transfer many revisions into a single pack on the
other side, so it isn't 100% clear cut. But that at least gives you the
approximation.

> 
> I'm wondering if there is a way to detect the point, when running pack
> will really help? In order to run the pack automatically on some of
> the nodes which are keeping repositories.

We do have the above heuristic, though it isn't necessarily optimal for
all cases. The goal is to trade off the effort of recombining everything
with the benefit of having it repacked.

> 
> Also, is pack command locking the repository ? Is it safe to run pack
> on repository from one process, and to make bzr push/pull attempts
> with the same repository from the other process?
> 
> Thanks.
> 
> 

Repositories are generally multi-writer safe. We lock for the one
operation which makes the update atomic. (do the work in a temporary
pack file, rename it into place, update the registry to include it.)

The one effect for pushing and pulling is that the pull may need to
restart if you move the files out from underneath it. The client should
be capable of coping, but it may do extra work. Though if you only
access the content via the 'smart' protocols (bzr+ssh, etc) then the
server should hide the effect (it should start reading from the new
content, and streaming it without interrupting the stream).

John
=:->
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.9 (Cygwin)
Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org/

iEYEARECAAYFAkyRDYIACgkQJdeBCYSNAAMZCwCgz/MPQwFz9lf23lk3pbquWdIv
TbkAn2BB856kAYvbwuIORdG+XYVoO8aI
=QRTI
-----END PGP SIGNATURE-----