On Wed, 2010-06-02 at 11:14:38 -0600, LaMont Jones wrote: > Package: dpkg > Version: 1.15.6 > Severity: normal > > With the introduction of fsync() calls to protect data, applications > that do potentially large apt-get install invocations may not want > to incur the penalty of fsync() calls from dpkg. > > In the case of building a livecd, this can be the difference between > a 10 minute apt invocation, and 21 minutes. > > The use cases where --force-unsafe-io would make sense are those > where apt is modifiying an underlying (possibly empty) base chroot > and the results will be thrown away before the next apt invocation > such that a crash just means "start from the known good base". The > two that jump to mind are: > 1. livecd builds > 2. buildd chroots where we are using lvm snapshots (or pbuilder or > whatever) and will be discarding the resulting chroot upon either > completion or machine reboot). > > Obviously, we want the default to be safe, and the option name to be > scary enough that end users don't think it's a good option to use in > daily life. I spoke with Colin Watson, and he suggested the option > be called --force-unsafe-io.
Right, that's what we discussed with Colin at some point, as a last resort "solution", but I added support for using sync() instead of fsync() on Linux to avoid the massive slow down. Have you tested 1.15.7.2 (which includes that change), and do the performance issues persist there. We actually got pretty good results from several testers on ext4, so if there's no need I'd just avoid adding such option. thanks, guillem -- To UNSUBSCRIBE, email to debian-bugs-dist-requ...@lists.debian.org with a subject of "unsubscribe". Trouble? Contact listmas...@lists.debian.org
