update-a-x-i should not run under pbuilder and similar systems

[Junichi Uekawa]

Hi,

At Mon, 23 Feb 2009 11:16:44 +0000,
Enrico Zini wrote:
> 
> Package: apt-xapian-index
> Severity: serious
> 
> On Mon, Feb 23, 2009 at 12:02:32PM +0900, Junichi Uekawa wrote:
> 
> We drifted a bit away from the original bug report, and I think this
> issue deserves its own bug number.
> 
> > > That's a good point.  How come your pbuilder setup installs
> > > apt-xapian-index?  It's usually pulled in by aptitude, but as a
> > > Recommends, not a Depends.  In theory, buildds and pbuilder chroots
> > > shouldn't install Recommends, or am I mistaken?
> > aptitude is depending on apt-xapian-index.
> > Package: aptitude
> [...]
> > Depends: apt-xapian-index, libapt-pkg-libc6.7-6-4.6, libc6 (>= 2.7-1), libcwidget3, libept0 (>= 0.5.26), libgcc1 (>= 1:4.1.1), libncursesw5 (>= 5.6+20071006-3), libsigc++-2.0-0c2
> > a (>= 2.0.2), libstdc++6 (>= 4.2.1), libxapian15, zlib1g (>= 1:1.1.4)
> 
> Oh, no.  I've opened a bug with aptitude asking if it can be downgraded
> to a Recommends (#516719).  I'm happy to have aptitude on my openmoko,
> but I certainly don't want to have apt-xapia-index in it.

Okay.

> > > I'd be happy to work out a way not to have apt-xapian-index run in cases
> > > like pbuilder.  Is invoke-rc.d the only tool that honours the init
> > > script policy?  Would it be acceptable if apt-xapian-index's postinst
> > > checked with policy-rc.d to see if it should run the reindexing or not?
> > For pbuillder specific, it's okay to check for policy-rc.d output.
> 
> Where can I find instructions on how to query policy-rc.d for this
> specific case?
> 
> Can it be just a case of:
> 
> if [ ! -x /usr/sbin/policy-rc.d ] || /usr/sbin/policy-rc.d apt-xapian-index start
> then
> 	update-apt-xapian-index --quiet &
> fi

Yeah, I think so.  policy-rc.d's original design intent is to have a
file in init.d with name 'apt-xapian-index' so that users can run
invoke-rc.d apt-xapian-index.

> > So, you really don't need the data until a client application requests
> > it.  The first invocation can probably wait, and the invocation afterwards
> > can wait
> > 
> > Or you can make installation wait for the rebuild of database.
> > There isn't much point in making it run in background.
> 
> The point is that the rebuild takes time, and a common use case is that
> the data is needed by the package manager just after apt-xapian-index is
> installed for the first time.  Rebuilding in background means that when
> the installation is finished and we're back to the package manager, the
> data will be almost readily available, instead of having to look at a
> progress bar for a minute or two.
> 
> 
> > If it's a long process, you should take care to make it run in the
> > background maybe niced.
> 
> You're right here, I had forgot to nice it.  I've fixed that in git.
> 
> 
> > Well, because you are running the indexing in background, it is
> > reasonable that someone would want to stop it, and if stopping is
> > possible, a restart/start.
> 
> update-apt-xapian-index --stop to stop a running background indexer is
> certainly a good idea: I've added it to the todo list (bug#TODO-TO-BE-ASSIGNED).
> 
> Restarting it is just a matter of running update-apt-xapian-index again.
> 
> > > However, I 100% agree that update-a-x-i must NOT run on chroots and slow
> > > down builds.  Let's find a way to do so.  Here are the options that I
> > > can think of:
> > >  - Study how policy-rc.d works, and query it in postinst to see if I
> > >    should run the update or not
> > >  - What does man-db do to prevent the index update?
> > Probably, man-db just rebuilds the index now in triggers.
> 
> I don't mind to use triggers, but I understand that they are only run at
> the end of the installation, so it would defeat the idea to start it
> soon so that at the end of the installation it's almost done.
> 
> Or is there a way to start triggers sooner?

Hmm... I understand what you're trying to do here now.  However,
package installation is a busy process with a lot of disk/CPU load,
I'm not sure if it's really giving you much advantage by having
started earlier.

> > >  - Disabling the update in postinst via a preseedable debconf
> > >    low-priority question
> > Feasible, but that is probably counterintuitive for users, and I need
> > a hack for apt-xapian-index in pbuilder.
> 
> Indeed.  It could still be useful in other cases, but not for pbuilder.
> 
> 
> It seems that the best way so far is to check policy-rc.d.  I've never
> used it, nor I found documentation about how to query it, so I inferred
> from invoke-rc.d source code.  If you confirm me the code snippet above
> on how to query it, I'll be happy to make an upload right away.

man invoke-rc.d 
refers you to:
/usr/share/doc/sysv-rc/README.policy-rc.d
which is actually available in:
/usr/share/doc/sysv-rc/README.policy-rc.d.gz
(filed a bug here...)

So there is a documentation.

regards,
	junichi
-- 
dancer@{netfort.gr.jp,debian.org}