| 1 |
On 5/12/20 11:22 PM, Michał Górny wrote: |
| 2 |
> On Tue, 2020-05-12 at 13:18 -0700, Zac Medico wrote: |
| 3 |
>> On 5/12/20 12:28 PM, Michał Górny wrote: |
| 4 |
>>> W dniu wto, 12.05.2020 o godzinie 10∶05 -0700, użytkownik Zac Medico |
| 5 |
>>> napisał: |
| 6 |
>>>> On 5/12/20 3:39 AM, Michał Górny wrote: |
| 7 |
>>>>> I'm sorry but I was asking of a more high-level implications. |
| 8 |
>>>>> |
| 9 |
>>>>> I presume that this means that more than files of more than one |
| 10 |
>>>>> package |
| 11 |
>>>>> can be merged simultaneously. However: |
| 12 |
>>>> |
| 13 |
>>>> No, an exclusive lock must be held on vardbapi._fs_lock for this. |
| 14 |
>>>> This |
| 15 |
>>>> is currently required at least to guarantee that access to the config |
| 16 |
>>>> memory file is serialized (config memory is the thing that emerge |
| 17 |
>>>> --noconfmem disables, but --noconfmem does not disable this lock). |
| 18 |
>>>> |
| 19 |
>>>> We assume that it's probably not worthwhile to try to merge files for |
| 20 |
>>>> more than one package at a time, since that would cause them to |
| 21 |
>>>> compete |
| 22 |
>>>> for IO bandwidth. |
| 23 |
>>>> |
| 24 |
>>>>> 1. Are collisions handled correctly then? i.e. if you start |
| 25 |
>>>>> installing |
| 26 |
>>>>> A, and then B, and the two packages collide will portage fail |
| 27 |
>>>>> before |
| 28 |
>>>>> starting to install any file from B? |
| 29 |
>>>> |
| 30 |
>>>> There are no guarantees here. However, the risk is minimal, since |
| 31 |
>>>> it's |
| 32 |
>>>> unlikely that a file collision of this sort would occur. file |
| 33 |
>>>> collisions |
| 34 |
>>>> are a QA problem that is generally detected and corrected log before |
| 35 |
>>>> we |
| 36 |
>>>> would encounter a collision of this sort. |
| 37 |
>>>> |
| 38 |
>>>>> 2. Are preinst/postinst phases called simultaneously or serialized? |
| 39 |
>>>> |
| 40 |
>>>> They're serialized. |
| 41 |
>>> |
| 42 |
>>> Now I'm lost here. Could you try to explain to me, without getting |
| 43 |
>>> into the deep technicalities, how parallel-install achieves better |
| 44 |
>>> speed or at doing what is non-parallel-install so slow? |
| 45 |
>> |
| 46 |
>> It allows preinst/postinst/prerm/postrm phases to run for one package |
| 47 |
>> while files are concurrently being merged or unmerged for another |
| 48 |
>> package. This makes it possible to approach saturation of IO bandwidth. |
| 49 |
> |
| 50 |
> Doesn't this imply that programs run in these phases could fail if |
| 51 |
> Portage is simultaneously replacing their files? |
| 52 |
|
| 53 |
We've got the same potential issue with emerge --jobs, but we use the |
| 54 |
dependency graph to schedule merges such that our dependencies do not |
| 55 |
mutate underneath us. |
| 56 |
-- |
| 57 |
Thanks, |
| 58 |
Zac |
Archives