Preserve oldest snapshot when forget policies not satisfied

[kapitainsky]

I would like to suggest restic implemented logic to try to keep the oldest snapshot

restic/restic#4366

At the moment with rustic with for example forget policies:

[forget]
keep-last = 2
keep-daily = 2
keep-weekly = 2
keep-monthly = 2

$ rustic forget
using config ./rustic.toml
[INFO] repository rclone:OD01-ENC:restic-test: password is correct.
[INFO] using cache at /Users/kptsky/Library/Caches/rustic/xxxx
snapshots for (host [kp-laptop.local], label [], paths [/])

ID	Time	Host	Label	Tags	Paths	Action	Reason
005a026e	2024-02-01 18:32:35	kp-laptop.local		test_rustic	/	keep	last
							daily
							weekly
							monthly
f2c0df87	2024-01-31 15:09:11	kp-laptop.local		test_rustic	/	keep	last
							daily
							monthly
2690ada9	2024-01-31 04:41:52	kp-laptop.local		test_rustic	/	remove
6fa4307a	2024-01-30 08:05:13	kp-laptop.local		test_rustic	/	remove
1dcc6ab7	2024-01-28 09:12:15	kp-laptop.local		test_rustic	/	keep	weekly
2bc5f01c	2023-12-31 20:44:27	kp-laptop.local		test_rustic	/	remove

on the 1st of Feb 2024 the last snapshot from 31st of Dec is removed - it is only 31 days old....

I think it is something from restic that can be copied into rustic - much more logical IMO.

[aawsome]

@kapitainsky What do you expect as outcome here? I think restic does exactly the same here - there are 2 last, 2 daily, 2 weekly and 2 monthly snapshots kept...

[kapitainsky]

The latest restic would not remove the latest snapshot - see restic PR I linked

[aawsome]

That logic you have linked is already implemented in rustic (it was implemented in rustic before it was added to restic). And in this example it doesn't apply, as all keep-* values are satisfied.

[kapitainsky]

And let's be pragmatic and logical - I want to keep last 2 monthly snapshots. In human terms it means I do not expect 1 month old snapshot to be removed (given it is the last one).

This is where pure algorithm logic does not work in real life.

IMO it is very practical to not touch the oldest snapshot - this is what restic does nowadays.

Look at my example - I want to keep 2 months snaps - suddenly I have only last few days. And before running forget I had 1 month.

[aawsome]

@kapitainsky Did you read my last comment?

I do think that rustic behaves exactly like restic here and that restic also would remove that last snapshot. Simply because it keeps 2 monthly snapshots, as requested: The one from 2024-02-01 and the one from 2024-01-31.

Can you explain why you expect rustic to keep the snapshot from 2023-12-31 which would be the third monthly snapshot?

[kapitainsky]

Because between 2024-02-01 and 2024-01-31 there is 1 day difference? And you are dropping one from 2023-12-31? Which maybe satisfy some artificial logic of 2 months but defies any common sense.

Let's say I started my rustic backup on 31/DEC/2023 - Hurra! my data is safe now. And I set forget policies to keep last 2 months. So now I know that I can restore my files back up to 2 month. All was fine until 1/FEB/2024 when suddenly computer decided you now have only last two days of backups:) WOW. Because my 31/DEC/2023 backup (despite to be less than 2 months old) has been removed.

Very simple adjustment to prevent it is (like restic) to not to remove the oldest snapshot if it does satisfy set policies. It creates sort of exception to algorithmic removal process but makes it much more human friendly.

[aawsome]

If you want to keep monthly snapshots which are less then 2 month old, the keep-within-monthly option would be the right choice. Or, you tell rustic to keep 3 monthly snaphots knowing that in the beginning of a month the monthly snapshot is an early one.

Very simple adjustment to prevent it is (like restic) to not to remove the oldest snapshot if it does satisfy set policies. It creates sort of exception to algorithmic removal process but makes it much more human friendly.

Please run restic forget --dry-runfor your example. It will produce exactly the same result as rustic.

[aawsome]

I agree that this keep-monthly option may be irritating when you first start. But I decided to act exactly like restic here in order to not confuse people who run restic and rustic on the same repositories. The only thing I changed compared to restic was that rustic always keeps the oldest snapshots if there are not enough snapshots satisfying the --keep-* criteria. For restic, this was added with the mentioned PR about a half year ago. Now rustic and restic behave exactly identically.

[aawsome]

FWIW, the logic you are requesting and which has been added recently to restic has been added to rustic here:
bf21f8b#diff-ffefaccfb096399d6312ab3946952e890c6cacb953210caed589cbf7609596aeR230
(snapshots are treated as candidates for a keep-* if they are either the oldest snapshot, the youngest snapshot or if the hour/day/week/.. doesn't match the last processed snapshot - Note that the oldest snapshot is in this list!)