Date: Fri, 16 Mar 2001 21:08:24 -0500 From: James Housley <jim@thehousleys.net> To: Dima Dorfman <dima@unixfreak.org> Cc: freebsd-arch@freebsd.org Subject: Re: Inherate nodump cause significant slow down of dump Message-ID: <3AB2C718.B31B29C3@thehousleys.net> References: <20010317015657.C47D03E1E@bazooka.unixfreak.org>
next in thread | previous in thread | raw e-mail | index | archive | help
Dima Dorfman wrote:
>
> [ Should this really be on -arch? ]
>
> James Housley <jim@thehousleys.net> writes:
> > Dima Dorfman wrote:
> > > I'm not terribly opposed to this, but OTOH I don't see why it's
> > > necessary. The slowdown only occurs if you set nodump on a directory.
> > > With the old behavior, this wouldn't do anything useful (AFAIK), so if
> > > you have nodump on a directory, you probably expect the new behavior.
> > > The latter has a problem with taking more time than it should (I think
> > > it actually takes the same amount of time as dumping everything under
> > > the nodump'd directory would). In other words, the only thing that's
> > > broken is the new feature.
> > >
> >
> > I haven't tested this yet, but should be able to tomorrow. But I
> > believe that if I do chflags -R nodump /usr/ports it will not a
> > significant more amount of time to process, throught Phase II, then with
> > out the nodump flags under the old version.
>
> I just tried it, and, as I suspected, it isn't the case. The slowdown
> is caused by the fact that for every directory with subdirectories and
> the nodump flag set, mapdirs returns 1, which causes it to be run
> again. You can test this theory by commenting out lines 250 and 251
> in rev. 1.12 of traverse.c (this will fix the slowdown, but break
> everything else).
Okay.
>
> > I am also guessing the with the new version chflags -R nodump /usr/ports
> > will be about the same amount of time as with the old code. I am not
> > opposed to the feature. Acutally I was suprised that that is not the
> > way it worked when I first found the flag. However, there has to be a
> > better way to implement it.
>
> dump's tree-walker is braindead. It can be implemented by adding
> another pass; I don't know if that will make it better or worse. If I
> can't fix this, I'll try that.
>
Talking about 1.12 traverse.c, about line #160 in mapfiles()
/*
* All dirs go in dumpdirmap; only inodes that are to
* be dumped go in usedinomap and dumpinomap, however.
*/
if (mode == IFDIR)
SETINO(ino, dumpdirmap);
------------------------^^^^^^^^^^^^^^^^^^^^^^^^
if (WANTTODUMP(dp)) {
SETINO(ino, usedinomap);
SETINO(ino, dumpinomap);
if (mode != IFREG && mode != IFDIR && mode !=
IFLNK)
*tapesize += 1;
else
*tapesize += blockest(dp);
continue;
}
if (mode == IFDIR)
anydirskipped = 1;
Why is a directory added without checking for the nodump flag? Because
isn't then just removed later in mapdirs()? Wouldn't this work better?
I don't fully understand all the code yet, but I am looking.
/*
* All dirs go in dumpdirmap; only inodes that are to
* be dumped go in usedinomap and dumpinomap, however.
*/
if (mode == IFDIR)
- SETINO(ino, dumpdirmap);
+ if (WANTTODUMP(dp))
+ SETINO(ino, dumpdirmap);
+ else {
+ anydirskipped = 1;
+ continue;
+ }
if (WANTTODUMP(dp)) {
SETINO(ino, usedinomap);
SETINO(ino, dumpinomap);
if (mode != IFREG && mode != IFDIR && mode !=
IFLNK)
*tapesize += 1;
else
*tapesize += blockest(dp);
continue;
}
if (mode == IFDIR)
anydirskipped = 1;
Jim
--
/"\ ASCII Ribbon Campaign .
\ / - NO HTML/RTF in e-mail .
X - NO Word docs in e-mail .
/ \ -----------------------------------------------------------------
jeh@FreeBSD.org http://www.FreeBSD.org The Power to Serve
jim@TheHousleys.Net http://www.TheHousleys.net
---------------------------------------------------------------------
If it happens once, it's a bug.
If it happens twice, it's a feature.
If it happens more than twice, it's windows.
-- Luiz de Barros
To Unsubscribe: send mail to majordomo@FreeBSD.org
with "unsubscribe freebsd-arch" in the body of the message
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?3AB2C718.B31B29C3>