On Tue, Jun 7, 2011 at 4:50 PM, Ric Wheeler <ricwheeler@xxxxxxxxx> wrote: > On 06/07/2011 09:01 AM, Amir G. wrote: >> >> On Tue, Jun 7, 2011 at 1:09 PM, Lukas Czerner<lczerner@xxxxxxxxxx> wrote: >>> >>> On Tue, 7 Jun 2011, Amir G. wrote: >>> >>>> On Tue, Jun 7, 2011 at 8:17 AM, Andreas Dilger<adilger@xxxxxxxxx> >>>> wrote: >>>>> >>>>> On 2011-06-06, at 2:55 PM, Ted Ts'o wrote: >>>>>> >>>>>> On Mon, Jun 06, 2011 at 10:31:33AM -0500, Eric Sandeen wrote: >>>>>>>> >>>>>>>> For one reason, a snapshot file format is currently an indirect file >>>>>>>> and big_alloc doesn't support indirect mapped files. >>>>>>>> I am not saying it cannot be done, but if it does, there would be >>>>>>>> several obstacles to cross. >>>>>>> >>>>>>> I know I'm kind of just throwing a bomb out here, but I am very >>>>>>> concerned >>>>>>> about the ever-growing feature (in)compatibility matrix in ext4. >>>>>> >>>>>> bigalloc doesn't support indirect blocks mainly because it was faster >>>>>> to get things working if I didn't have to worry about indirect blocks. >>>>>> It wouldn't be _that_ hard to make bigalloc work on indirect blocks. >>>>>> I'll get around to it at some point. >>>>> >>>>> My main concern isn't about whether bigalloc grows support for >>>>> indirect- >>>>> mapped files, but rather the opposite - that snapshots gain support for >>>>> extent-mapped files. In fact, since extent-mapped files can be 16TB in >>>>> size, it might make sense that the snapshots are _always_ extent-mapped >>>>> files, and we don't need to deal with the new block-mapped files with >>>>> 4-triple-indirect blocks layout at all? Since snapshots are only going >>>>> into ext4, and ext4 + e2fsprogs already support extents, there wouldn't >>>>> be any issue about compatibility? >>>>> >>>>> The only concern might be that mapping fragmented files into extents is >>>>> more effort, which makes me wonder about whether we should introduce >>>>> the >>>>> "block-mapped extents" that I proposed in the past, to allow efficient >>>>> mapping of files (or parts thereof) that are highly fragmented, but >>>>> still >>>>> keeping the benefits of extents (internal redundancy, 48-bit physical >>>>> block numbers, and while we are adding a new extent format it could be >>>>> designed to add 48-bit logical block numbers. >>>>> >>>> You are right about snapshot file being a highly fragmented file by >>>> design, >>>> so single block mapping is an advantage. The down side is that deleting >>>> an extent mapped file, requires mapping all blocks one-by-one to >>>> snapshot >>>> file, which is not efficient and makes deletes slow. >>>> So having a format optimized for both single and multi block mapping >>>> would be >>>> best. >>>> >>>> The reason I DO NOT want to change the snapshot file format at this >>>> moment >>>> is that it will make us lose all the stabilization that snapshot feature >>>> gained >>>> during 1 year in production as next3. >>>> You see, ext4_free_blocks() cares not if blocks are deleted from >>>> indirect or >>>> extent mapped files and from there on, the code that maps those blocks >>>> to >>>> the special snapshot file is the same in next3 and ext4. >>>> >>> But the problem is, that you will not be able to change it in the future >>> or at least not without adding more incompatibility flags, which is >>> exactly the point of this thread. I just wonder if it would not be >>> better to do it now, because now is the right time. Although I do not >>> know how much work will that require. >>> >> There are no compatibility issues. >> ext4 fs is either 32bit or 64bit and you cannot convert between the 2 >> formats. >> 32bit ext4 has snapshots support with indirect mapped snapshot files. >> 64bit ext4 has no snapshots support. >> if in the future, be it near or far, 64bit ext4 will have snapshots >> support with >> a new snapshot file format, then 64bit feature + snapshots feature will >> prevent the present (i.e. next) kernel from mouting that fs rw. >> which is exactly the same as older kernel will prevent mounting a 32bit >> ext4 >> with snapshots rw. >> >> Amir. > > Hi Amir, > > I really am not comfortable with having two formats for snapshots. > > Why not just do one 64 bit format and skip the 32 bit one? Well for 2 reasons mainly: 1. Something like that could hold back the feature further more and maybe even to eternity and some people do want to use it this lifetime. 2. There are performance implications that need to be studied. An indirect format gives me the ability to maps blocks of different block groups without taking a global lock (not doing that yet). With extent tree format, a global lock is needed for re-balancing the tree, so concurrent COW operations on different blocks in different block groups are bound to contend the same global lock, which is something I am trying to see if can be avoided. > > This seems like a recipe for end user confusion and pain :) > I honestly don't see how the internal format of a snapshot file affects the end user in any way. What happens in 32bit ext4 stays in 32bit ext4. There is no migration of formats whatsoever to 64bit ext4. The only pain caused by 2 formats is having to maintain the code for 2 formats. But the fact of the matter is that indirect mapped file code is there to stay, so having the snapshot file use it for now, is not much of a maintenance burden later. All it takes is an EXTENT_FL flag to distinguish between an indirect mapped snapshot to a future extent mapped (v2) snapshot. > thanks! > > Ric > > -- To unsubscribe from this list: send the line "unsubscribe linux-ext4" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
