1 |
Dnia October 8, 2019 7:13:03 AM UTC, SoEasyTo Mirrors Manager <mirrors@××××××××.com> napisał(a): |
2 |
>On 2019-10-08 05:33, Michał Górny wrote : |
3 |
> |
4 |
>> Hello, everyone. |
5 |
>> |
6 |
>> TL;DR: shortly, distfiles will need to be present under two paths for |
7 |
>> the transitional period. Would you prefer us using hardlinks or |
8 |
>> symlinks for that? |
9 |
>> |
10 |
>> We're planning to start deploying a new GLEP 75-based [1] mirror |
11 |
>layout |
12 |
>> to our mirrors soonish. This implies a transitional period during |
13 |
>> which |
14 |
>> we'll be using both old and new layouts, so all file entries will be |
15 |
>> duplicated. The plan is roughly to: |
16 |
>> |
17 |
>> 1. Enable new split layout in emirrordist, and start using both |
18 |
>> simultaneously for newly-mirrored files. |
19 |
>> |
20 |
>> 2. Duplicate the existing distfiles to new layout. |
21 |
>> |
22 |
>> 3. Live with both layouts for some longish time, to support people |
23 |
>> using |
24 |
>> old Portage versions. |
25 |
>> |
26 |
>> 4. Eventually disable the old (flat) layout and start removing files. |
27 |
>> |
28 |
>> The basic problem is whether to use hardlinks or symlinks |
29 |
>> for the duplicate files. I've elaborate more on both solutions in |
30 |
>[2] |
31 |
>> but I'll summarize shortly here. |
32 |
>> |
33 |
>> Hardlinks have the advantage that for mirrors enabling -H, they avoid |
34 |
>> extra space usage and extra traffic. However, we don't really know |
35 |
>how |
36 |
>> many mirrors enable that, and I suspect it's around half of them. |
37 |
>> At initial deployment time, rsync will just hardlink files in new |
38 |
>> layout |
39 |
>> to existing entries, and at cleanup time it will just unlink old |
40 |
>> entries. |
41 |
>> |
42 |
>> For mirrors not enabling -H, hardlinks will mean all distfiles being |
43 |
>> transferred again during deployment time. Furthermore, through all |
44 |
>> transitional period all files will be duplicated, and so duplicated |
45 |
>> will |
46 |
>> be space usage. Cleanup should be lightweight though. |
47 |
>> |
48 |
>> Symlinks have the advantage that we know that all or almost all |
49 |
>mirrors |
50 |
>> enable them. They are lightweight at deployment time since it's just |
51 |
>> a matter of rsync copying symlinks, and they definitely won't cause |
52 |
>> double space usage. However, they will cause all files being |
53 |
>> retransferred at cleanup time -- due to symlinks being replaced by |
54 |
>real |
55 |
>> files. |
56 |
>> |
57 |
>> Technically, I suppose we could avoid that by splitting that into two |
58 |
>> stages, repeated for smaller groups of files. Firstly, replace |
59 |
>> symlinks |
60 |
>> with hardlinks which will make it light for at least some of the |
61 |
>> errors. |
62 |
>> Then, remove old files and jump over to the next group. For mirrors |
63 |
>> not |
64 |
>> using -H, this will still mean double transfer but we'd limit double |
65 |
>> space usage to one group at a time, and only for a short period. |
66 |
>> |
67 |
>> If any mirrors sync over rsync without using -l (talking about |
68 |
>private |
69 |
>> mirrors here), they will not get the new layout at all which is going |
70 |
> |
71 |
>> to |
72 |
>> suck for their users. |
73 |
>> |
74 |
>> Which way do you prefer? |
75 |
> |
76 |
>For soeasyto mirror, we are already using both -H and --links, and the |
77 |
>mirror |
78 |
>is hosted on a single partition, so, in order to preserve bandwith as |
79 |
>you |
80 |
>suggested, it's better to use hardlinks, keeping in mind that could |
81 |
>cause |
82 |
>server "overload" as per [1], but it is not an issue here. |
83 |
> |
84 |
>One question remains though: how will the layout.conf be created ? Is |
85 |
>it |
86 |
>by the |
87 |
>mirror maintainer, or only by the master distfiles, and then all |
88 |
>mirrors |
89 |
>will |
90 |
>automatically replicate it ? Because it could be interesting to let the |
91 |
> |
92 |
>mirror |
93 |
>maintainer decide whether to use split or flat layout depending on |
94 |
>their |
95 |
>usage |
96 |
>of hardlinks / symlinks, and leave the choice by providing a master for |
97 |
> |
98 |
>flat, |
99 |
>hybrid, and split layouts ? |
100 |
|
101 |
We are replicating layout.conf along with distfiles from the master mirror. This makes sense for the majority of the mirrors since they're going to replicate the structure of master mirror as well. |
102 |
|
103 |
You can technically override this locally but you'd also have to adjust fetch procedure to account for the changed layout, i.e. run your custom tooling. |
104 |
|
105 |
If you do not rsync from master mirror and e.g. use emirrordist locally, you'd have to create layout.conf yourself. The future version of emirrordist will respect your setting there (the patches are not merged yet). |
106 |
|
107 |
|
108 |
> |
109 |
>> [1] https://www.gentoo.org/glep/glep-0075.html |
110 |
>> [2] https://bugs.gentoo.org/534528#c38 |
111 |
|
112 |
|
113 |
-- |
114 |
Best regards, |
115 |
Michał Górny |