On 11/14/07, Chuck Lever <chuck.lever@xxxxxxxxxx> wrote: > Jon Smirl wrote: > > On 11/14/07, Chuck Lever <chuck.lever@xxxxxxxxxx> wrote: > >> Jon Smirl wrote: > >>> On 11/14/07, Chuck Lever <chuck.lever@xxxxxxxxxx> wrote: > >>>> On Nov 13, 2007, at 7:04 PM, Jon Smirl wrote: > >>>>> Is it feasible to do something like this in the linux file system > >>>>> architecture? > >>>>> > >>>>> Beagle beats on my disk for an hour when I reboot. Of course I don't > >>>>> like that and I shut Beagle off. > >>>> Leopard, by the way, does exactly this: it has a daemon that starts > >>>> at boot time and taps FSEvents then journals file system changes to a > >>>> well-known file on local disk. > >>> Logging file systems have all of the needed info. Plus they know what > >>> is going on with rollback/replay after a crash. > >> True, but not all file systems have a journal. Consider ext2 or FAT32, > >> both of which are still common. > > > > ext2/FAT32 can use the deamon approach you describe below which also > > works as a short term solution. The Beagle people do have a deamon but > > it can be turned off. Holes where you don't record the inotify events > > and update the index are really bad because they can make files that > > you know are on the disk disappear from the index. I don't believe > > Beagle distinguishes between someone turning it off for a day and then > > turning it back on, vs a reboot. In both cases it says there was a > > window where untracked changes could have happened and it triggers a > > full rescan. > > > > The root problem here is needing a bullet proof inotify log with no > > windows. > > I disagree: we don't need a "bullet-proof" log. We can get a > significant performance improvement even with a permanent dnotify log > implemented in user-space. We already have well-defined fallback > behavior if such a log is missing or incomplete. > > The problem with a permanent inotify log is that it can become > unmanageably enormous, and a performance problem to boot. Recording at > that level of detail makes it more likely that the logger won't be able > to keep up with file system activity. It doesn't have to become enormous, if the checkpoint request is too old just return no-data and trigger a full scan in Beagle. 50K of log data would probably be enough. The main thing you need to cover is the reboot process and files that get touched after the beagle shuts down or before it gets started. For example the log could checkpoint once a minute, in that case you wouldn't need more than two minutes worth of log. Beagle would just remember the last checkpoint it processed and apply reapply changes after it. If someone turns Beagle off for a couple of days it should be expected that they will need a full scan when they turn it back on. > > A lightweight solution gets us most of the way there, is simple to > implement, and doesn't introduce many new issues. As long as it can > tell us precisely where the holes are, it shouldn't be a problem. > > > The only place that is going to happen is inside the file > > system logs. > > As Andi points out, existing block-based journaling implementations > won't easily provide this. And most fs journals are actually pretty > limited in size. > > Alternately, you could insert a stackable file system layer between the > VFS and the on-disk fs to provide more seamless information about updates. > > > We just need an API to say recreate the inotify stream > > from this checkpoint forward. Things like FAT/ext2 will always return > > a no data available error from this API. > > > >>> How about a fs API > >>> where Beagle has a token for a checkpoint, and then it can ask for a > >>> recreation of inotify events from that point forward. It's always > >>> possible for the file system to say I can't do that and trigger a full > >>> rebuild from Beagle. Daemons that aren't coordinated with the file > >>> system have a window during crash/reboot where they can get confused. > >> A reasonably effective solution can be implemented in user space without > >> changes to the file system APIs or implementations. IOW we already have > >> the tools to make something useful. > >> > >> For example, you don't need to record every file system event to make > >> this useful. Listing only directory-level changes (ie "some file in > >> this directory has changed") is enough to prune most of Beagle's work > >> when it starts up. > >> > >>> Without low level support like this Beagle is forced to do a rescan on > >>> every boot. Since I crash my machine all of the time the disk load > >>> from rebooting is intolerable and I turn Beagle off. Even just turning > >>> the machine on in the morning generates an annoyingly large load on > >>> the disk. > >> Understood. The need is clear. > >> > >> My Dad's WinXP system takes 10 minutes after every start-up before it's > >> usable, simply because the virus scanner has to check every file in the > >> system. Same problem! > >> > >>>> I don't see why this couldn't be done on Linux as well. > >>>> > >>>>> ---------- Forwarded message ---------- > >>>>> From: Jon Smirl <jonsmirl@xxxxxxxxx> > >>>>> Date: Nov 13, 2007 4:44 PM > >>>>> Subject: Re: Strange "beagle" interaction.. > >>>>> To: Linus Torvalds <torvalds@xxxxxxxxxxxxxxxxxxxx> > >>>>> Cc: "J. Bruce Fields" <bfields@xxxxxxxxxxxx>, Junio C Hamano > >>>>> <gitster@xxxxxxxxx>, Git Mailing List <git@xxxxxxxxxxxxxxx>, Johannes > >>>>> Schindelin <Johannes.Schindelin@xxxxxx> > >>>>> > >>>>> > >>>>> On 11/13/07, Linus Torvalds <torvalds@xxxxxxxxxxxxxxxxxxxx> wrote: > >>>>>> On Tue, 13 Nov 2007, J. Bruce Fields wrote: > >>>>>>> Last I ran across this, I believe I found it was adding extended > >>>>>>> attributes to the file. > >>>>>> Yeah, I just straced it and found the same thing. It's saving > >>>>>> fingerprints > >>>>>> and mtimes to files in the extended attributes. > >>>>> Things like Beagle need a guaranteed log of global inotify events. > >>>>> That would let them efficiently find changes made since the last time > >>>>> they updated their index. > >>>>> > >>>>> Right now every time Beagle starts it hasn't got a clue what has > >>>>> changed in the file system since it was last run. This forces Beagle > >>>>> to rescan the entire filesystem every time it is started. The xattrs > >>>>> are used as cache to reduce this load somewhat. > >>>>> > >>>>> A better solution would be for the kernel to log inotify events to > >>>>> disk in a manner that survives reboots. When Beagle starts it would > >>>>> locate its last checkpoint and then process the logged inotify events > >>>>> from that time forward. This inotify logging needs to be bullet proof > >>>>> or it will mess up your Beagle index. > >>>>> > >>>>> Logged files systems already contain the logged inotify data (in their > >>>>> own internal form). There's just no universal API for retrieving it in > >>>>> a file system independent manner. > >>>>> > >>>>>>> Yeah, I just turned off beagle. It looked to me like it was doing > >>>>>>> something wrongheaded. > >>>>>> Gaah. The problem is, setting xattrs does actually change ctime. > >>>>>> Which > >>>>>> means that if we want to make git play nice with beagle, I guess > >>>>>> we have > >>>>>> to just remove the comparison of ctime. > >>>>>> > >>>>>> Oh, well. Git doesn't *require* it, but I like the notion of > >>>>>> checking the > >>>>>> inode really really carefully. But it looks like it may not be an > >>>>>> option, > >>>>>> because of file indexers hiding stuff behind our backs. > >>>>>> > >>>>>> Or we could just tell people not to run beagle on their git trees, > >>>>>> but I > >>>>>> suspect some people will actually *want* to. Even if it flushes > >>>>>> their disk > >>>>>> caches. > >>>>>> > >>>>>> Linus > >>>>>> - > >>>>>> To unsubscribe from this list: send the line "unsubscribe git" in > >>>>>> the body of a message to majordomo@xxxxxxxxxxxxxxx > >>>>>> More majordomo info at http://vger.kernel.org/majordomo-info.html > >>>>>> > >>>>> -- > >>>>> Jon Smirl > >>>>> jonsmirl@xxxxxxxxx > >>>>> > >>>>> > >>>>> -- > >>>>> Jon Smirl > >>>>> jonsmirl@xxxxxxxxx > >>>>> - > >>>>> To unsubscribe from this list: send the line "unsubscribe linux- > >>>>> fsdevel" in > >>>>> the body of a message to majordomo@xxxxxxxxxxxxxxx > >>>>> More majordomo info at http://vger.kernel.org/majordomo-info.html > >>>> -- > >>>> Chuck Lever > >>>> chuck[dot]lever[at]oracle[dot]com > >>>> > >>>> > >>>> > >>>> > >>> > >> > > > > > > -- Jon Smirl jonsmirl@xxxxxxxxx - To unsubscribe from this list: send the line "unsubscribe linux-fsdevel" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
