On 13/06/16 13:05, "[email protected] on behalf of Roman Haefeli" <[email protected] on behalf of [email protected]> wrote:
>On Fri, 2016-06-10 at 09:44 +0000, Dmitry Mishin wrote: >> >> On 09/06/16 18:27, "[email protected] on behalf of Roman >> Haefeli" >> <[email protected] on behalf of [email protected]> wrote: >> >> > >> > On Fri, 2016-06-03 at 15:10 +0000, Dmitry Mishin wrote: >> > > >> > > Hi, >> > > >> > > On 03/06/16 11:52, "[email protected] on behalf of Roman >> > > Haefeli" >> > > <[email protected] on behalf of [email protected]> wrote: >> > > >> > > > >> > > > >> > > > Dear All, >> > > > >> > > > We're considering creating regular snapshots of our containers. >> > > > The >> > > > setup would include deleting the oldest snapshots. While >> > > > playing >> > > > around >> > > > with creating and deleting snapshots, I noticed that the >> > > > root.hdd >> > > > file >> > > > keeps growing. It seems there is no point at all in deleting >> > > > old >> > > > snapshots for the only purpose of saving disk space. Then I >> > > > tried >> > > > to >> > > > reclaim some space back with 'vzctl compact <name>', but it >> > > > wasn't >> > > > able >> > > > to reclaim anything. Only when I deleted all snapshots, 'vzctl >> > > > compact >> > > > <name>' seems to be effective. >> > > > >> > > > Now, my question is, will there be support in the future to >> > > > reclaim >> > > > space, even if there are snapshots? >> > > It's a bit tricky, and that'a why there were no such plans so >> > > far. >> > > Could you please explain why are you considering regular >> > > snapshots? >> > > What >> > > is a use case? >> > Our ploop containers have their image on an NFS store. We once had >> > an >> > outage of the NFS server and thus a lot of images with broken >> > filesystems. We do have backups, but sometimes it would be easier >> > to >> > simply switch to a recent snapshot. >> > >> > Why is it considered special to have a snapshots of the - let's say >> > - >> > past 7 days? >> Mb because of habit to use simfs? >> >> Please forgive me, if the question sounded offensive, it was not >> supposed >> to be such in any way - the only reason why I ask such questions is >> that >> we as developers lack the knowledge of whole range of community use >> cases. >> And these questions allow us to clarify is the particular use case a >> pattern and why. > >Thanks, I didn't take it as offensive. > >> Thank you for the answer, I've got the idea. >> >> Back to the topic. >> >> Snapshot by its definition is an ability to make a read-only dump of >> a >> Container's (and thus all its hdds) at some time point. The keyword >> here >> is "read-only". Every feature or tool which use this functionality - >> rely >> on this assumption. So, any functionality which would change a 'read- >> only' >> delta - breaks this assumption and thus leads to issues with other >> functionality. And that's why nobody performs compacting of >> snapshots. >> >> The proper analogy here is 'backup' - nobody expects that compact of >> a >> virtual disk will lead to drop of 'unused' data from backups as well. >> >> Possible solution here - is to do not snapshot/backup unused data. >> It doesn't worth to modify a snapshot action for this - because: >> 1) it makes it to be a non-atomic action (only guest OS knows what >> data is >> unused) and >> 2) it can be easily emulated be a sequence of 2 commands: >> a) ploop balloon discard >> b) vzctl snapshot >> >> I.e. do discard just before snapshot. >> >> So, I'd recommend: >> 1) at some point, do a Container's backup, merge all snapshots and >> run >> compact >> 2) after that, do compact just before any snapshot. >> >> It will not solve the issue completely (because of defragmentation >> issue), >> but IMHO, it is the best what can be achieved now. > >Thanks for the explanations. I think I understood the part with >compacting before snapshotting. However, is it correct that if a create >a 10GB and then take a snapshot, delete the file and take another >snapshot and continue rotating snapshots (creating new ones while >merging the oldest ones) the 10GB would end up in root.hdd even if no >snapshot is "actively" referencing this data anymore? So even if I >compact before I snapshot, there will be wasted space. Is this correct? Yes, you are right - this issue affects ploop (but not affect Vms with qcow2 images format). Unfortunately, this makes whole advice useless - one still needs to merge all snapshots and run ploop discrad on an image without them periodically. > >> And back to the off-topic again: >> What are NFS benefits, which outweigh their instability? > >Good question. We never had troubles with NFS before, but the hardware >is aging. When we set up the Hostnodes, we wanted to have the >possibility to quickly migrate between Hostnodes, so we decided to put >containers on NFS export shared between hostnodes. Got it, thank you. > >Roman _______________________________________________ Users mailing list [email protected] https://lists.openvz.org/mailman/listinfo/users
