Dear Anne,
I beg to be of a different opinion:
- three is a very small number
- the oldest, that is proposed to be removed, could be the best (or only one) working well on the current machine: the more recent ones might have been tried and found unsatisfactory.
If it is decided to go this way, the user / owner should be allowed to override the choice of removal, e.g. prompt him to answer to: "Is it okay to remove kernel-x.y.z.?" y/N
Kind regards, Dick Gevers
(ennael) We should keep at least new kernel and the one before. In case of regression (mainly hardware support for example), we can switch back to the one before
(annew) Three has worked very well for me in the past, but this is an arbitrary number. A larger number could be chosen, or, better still, the use could choose between 3 (in my opinion, the minimum) or, say, 5. In cases where hardware problems are significant, the larger number may well be chosen.
The only time I had a significant problem was one where a particular kernel caused a lot of problems on my system. On that occasion I removed the troublesome kernel by hand, so the 3 rule still worked for me. However, since the troublesome one would usually be the newest one, creating an escape route from it, by blacklisting such a kernel, may be more difficult to achieve.
(dvgevers) Dear Annes ;) My experience is mostly Cauldron, where it's all developed. But I do have one Mga1 machine with HW from the previous millennium and rarely does a new kernel work w/o probs. It boots up in ~20 minutes every 1/2 or 1 1/2 years: YMMV. All I'm saying is: please don't make the playground too small for our "kids". Thank you.
(ovitters) Suggest to limiting number of installed kernels on these criteria: 1. Is it the latest yes/no (always keep latest, irrespective when it has been last booted) 2. When has the kernel last been booted (e.g. have systemd touch the vmlinuz file or something) 3. Always keep 4 kernels (so always latest, plus 3 ordered by last time they've been booted)
(anaselli) Well I am in favor of having a number, we just have to choose the default one, so i believe 2 or 3 is correct to boot back to the old system. The big issue is not into our stable in which we should not have big problems, but in cauldron. So we could decide to have N=0 (do not remove anything, e.g. how it works now) N>0 (N old kernels). I believe someone would have instead N<0 (-or INF- do not remove anything, e.g. how it works now) N>=0 (N old kernels) adding also "leave only the last kernel" option.
(mailedfist) I agree with Anne's suggestion that the default be 3. Provided that the mechanism to set N (via MCC, I suggest) is straightforward, it would seem that having 0<N<255 should allow everyone to be happy. If we must use N=0 for 'keep everything' then so-be-it, but it wouldn't be so hard to have a checkbox for that instead, would it?
Please Don't Make It Mandatory
If you're going to implement this feature, please either make it optional, or alterantively allow the number of kept kernels to be specified up to any arbitrary limit (say 100 kernels or more). I don't want to be stuck in a system I cannot boot because all the kernels are incompatible. I recall a bugzilla.redhat.com bug where one commentator noted that he'll be in the problem when the new kernel will be coming because his only working kernel (the oldest) will be removed.
Remember that Linux is about choice.
shlomif 20:16, 15 June 2012 (UTC)
