[OpenIndiana-discuss] Time slider missing some snapshots and not deleting snapshots

[Martyn Klassen]

I guess I spoke too soon. After letting time-slider run for the day, I ended up with way too many frequent snapshot and a bunch of missing snapshots, so something is more seriously wrong with the recursion algorithms being used by time-slider. I even ran another test - with 5 minute interval frequent snapshots - intentionally changing every filesystem so that no snapshot is ever zero sized and still some filesystems are not getting snapshots. Missing snapshots always occurs in filesystems preceding those with children, which seems to point to recursion issues. My guess is the deletion problems are related.

If anyone has any suggestions on how to debug these time-slider issues they would be appreciated. 

Martyn

On 2012-06-21, at 13:56 , Martyn Klassen wrote:

> Time slider is missing the creation of some and deletion of all snapshot of filesystems on my data pool. The snapshots are only create for the last filesystem in a tree, for example with pool data and filesystems
> 
> data/fsA
> data/fsB/fs1
> data/fsC
> data/fsD/fs1
> data/fsE
> 
> only data/fsB/fs1 and data/fsD/fs1 get a snapshot and the snapshots are not deleted properly, ie frequent snapshots just keep getting added beyond the 3 that are supposed to be kept. 
> 
> After playing with various setting, I found a solution that appears to be working for me. So far it is handling frequent snapshots correctly.
> 
> The short version:
> Do not set com.sun:auto-snapshot properties on the pool, ie data, but on filesystems created on the pool, ie fsA - fsB.
> 
> The long version:
> The issue seems to be related to inheriting com.sun:auto-snapshot and the recursion algorithms being used to make and destroy snapshots.
> 
> I found this old post http://mail.opensolaris.org/pipermail/indiana-discuss/2009-December/017161.html which made me investigate whether the issue is related to the snapshot properties being inherited directly from the pool and auto-snapshots recursion algorithm. The time slider code base has obvious changed since that post, but I decided to try the following anyway
> 
> zfs create data/nosnap
> zfs set com.sun:auto-snapshot=false data/nosnap
> zfs set com.sun:auto-snapshot:hourly=false data/nosnap
> zfs set  com.sun:auto-snapshot:frequent=false  data/nosnap
> zfs set  com.sun:auto-snapshot:daily=false data/nosnap
> zfs set  com.sun:auto-snapshot:weekly=false  data/nosnap
> zfs set  com.sun:auto-snapshot:monthly=false data/nosnap
> 
> Basically the idea is that the data/nosnap filesystem with all auto-snapshot properties set to false should force auto-snapshot to address each filesystem separately. and not act recursively on data. With this change time-slider starting making snapshots for data/fsD/fs1 and data/fsE get snapshots, but none of the rest, and none of the snapshots get deleted properly.
> 
> My next attempt was to clear all the com.sun:auto-snapshot properties from data and set them on the filesystems fsA, fsB, fsC, fsD, and fsE, and then all the snapshots were create and destroyed as expected.
> 
> zfs set com.sun:auto-snapshot=true data/fsA
> zfs set com.sun:auto-snapshot=true data/fsB
> zfs set com.sun:auto-snapshot=true data/fsC
> zfs set com.sun:auto-snapshot=true data/fsD
> zfs set com.sun:auto-snapshot=true data/fsE
> 
> Obviously this is more tedious that just setting the properties on data, but at least the snapshots seem to work for me now. Hopefully this is useful for anyone else running into this issue, and if someone can point me in the right direction to submit a bug report on this, maybe it can get fixed in a future release.
> 
> Just a bit more about my configuration for reference:
> 
> # uname
> SunOS ****** 5.11 oi_151a4 i86pc i386 i86pc Solaris
> # pkg info time-slider
>          Name: desktop/time-slider
>       Summary: Time Slider ZFS snapshot management for GNOME
>   Description: Time Slider ZFS snapshot management for GNOME
>      Category: Applications/Configuration and Preferences
>         State: Installed
>     Publisher: openindiana.org
>       Version: 0.2.97
> Build Release: 5.11
>        Branch: 0.151.1.4
> Packaging Date: May  2, 2012 10:25:35 PM 
>          Size: 715.80 kB
>          FMRI: pkg://openindiana.org/desktop/time-slider@0.2.97,5.11-0.151.1.4:20120502T222535Z
> 
> 
> _______________________________________________
> OpenIndiana-discuss mailing list
> OpenIndiana-discuss at openindiana.org
> http://openindiana.org/mailman/listinfo/openindiana-discuss