[pve-devel] applied: [PATCH installer] abort installation if memory is less than 1GB

[Thomas Lamprecht]

On 7/3/19 12:06 PM, Oguz Bektas wrote:
> On Wed, Jul 03, 2019 at 11:49:00AM +0200, Thomas Lamprecht wrote:
>> On 7/3/19 11:50 AM, Oguz Bektas wrote:
>>>> I do not want to die here, if one wants to continue, why not (e.g.,
>>>> the limit is not exactly 1024 but rather somewhere beteween 850-900
>>>> MB, and also then the error is IMO not reasonable, proxinstall +
>>>> gtk-webkit + base system need ~ 300 MB memory, and that the page
>>>> cache flushes result in memory almost OOM-like errors seems wrong.)
>>>>
>>>> I mean, the check is still OK, as this is our min. system requirement,
>>>> and it does normally not make sense to install PVE/PMG to anything with
>>>> less memory, besides from testing :)
>>> the reason i added the die is, it'll fail at the last step during
>>> unsquashfs if there's not enough memory (gets oom killed). this is kind
>>> of annoying and you end up with an unbootable system on top, so it's a
>>> waste of time. why wouldn't we prevent that?
>>
>> as I already stated: 1024 is not the hardlimit, less works and we may even
>> be able to work with much less if I have some time to look at this in the
>> future. It'll be always good to remind people about less memory than the
>> minimal system requirements state, though. We just warn here, if it still
>> works then good, if not the user was warned, I do not see the problem?
> it's not a "problem" per se, it's just that i'd like to avoid wasting
> time doing an entire install just to have it get oom-killed in the last
> step which is really frustrating... i guess a warning is fine for now
> until we can improve the memory usage or find a better solution in the
> future.
> 

again what's the issue, you get the warning, if you read it you get reminded
that you assigned too little memory to a _test vm_ (as this normally _only_
happens there), _nobody_ forces you to continue the installation there, if
you do not like to waste time then stop right there, and re-start with more
memory, IMO, you do not win anything with die here, just restrict people who
explicitly want to test something. 99.999% of all user won't run into this
for production systems, it'll always just be on testing systems.