1 |
On 04/05/2016 11:37 AM, Alexis Ballier wrote: |
2 |
> On Sunday, April 3, 2016 8:07:13 PM CEST, Ulrich Mueller wrote: |
3 |
>>> In two weeks from now, the council will meet again. This is the time |
4 |
>>> to raise and prepare items that the council should put on the agenda |
5 |
>>> to discuss or vote on. |
6 |
>> |
7 |
>> I would like the council to follow up on the results of robbat2's |
8 |
>> portage repo usage survey: |
9 |
>> https://archives.gentoo.org/gentoo-project/message/c2ffa62837fd4cbdd42945bf57b09b25 |
10 |
>> |
11 |
>> |
12 |
>> The following two points should be discussed and possibly be voted on: |
13 |
>> |
14 |
>> 1. Should we continue providing ChangeLog files in the rsync |
15 |
>> distribution? |
16 |
> |
17 |
> If space is the sole consideration, then, changelogs can be removed |
18 |
> from manifests, and $PM default rsync command can exclude |
19 |
> '*/*/ChangeLog*' (or, at least, leave the possibility to do it). This |
20 |
> solution seems to be what would please everyone since robbat2's survey |
21 |
> results can be interpreted in many ways, but for Q2, more than 50% |
22 |
> voted "something but only if it were optional". |
23 |
> |
24 |
> |
25 |
> However, I think I recall a nice recap of some infra guy on how much |
26 |
> time it took to generate them. IIRC it was bearable at the moment (a |
27 |
> few hours) but still slow. Assuming infra hardware stays the same, |
28 |
> where will we be in 1, 2 or 5 years wrt to generating changelogs ? Is |
29 |
> there something that can be improved on the software side or are we |
30 |
> just bound to have slower and slower rsync distribution generation ? |
31 |
> If so, how much slower does it get over time ? |
32 |
If one had lots of time for fixing up things ... an incremental |
33 |
thick-manifest update in lockstep with new commits would be possible. So |
34 |
maybe each commit lags by a few seconds to a few minutes (if it's a |
35 |
single commit touching lots of files). |
36 |
|
37 |
But there would always be a thick-manifest copy available that would be |
38 |
behind only by a small time window, and could be propagated to rsync |
39 |
mirrors easily. The problem with that approach is that someone would |
40 |
need to write such code, test it, etc.etc. |
41 |
|
42 |
|
43 |
> What I mean there is that if changelog generation takes 5 days then we |
44 |
> don't have much of a choice but dropping them. |
45 |
> |
46 |
> Alexis. |
47 |
> |
48 |
Pff. Worst case you throw more hardware at it ;) |