1 |
On Mon, Apr 2, 2012 at 7:49 AM, Carsten Otto |
2 |
<otto@××××××××××××××××××××××.de> wrote: |
3 |
> Dear Gentoo admins, |
4 |
> |
5 |
> it seems you made some changes to the data you distribute as |
6 |
> "gentoo-portage". |
7 |
|
8 |
Well, the infra team didn't make any changes. The gentoo-portage tree |
9 |
is a living thing that adapts as the software becomes more complex. |
10 |
There was a change to better protect users against timestamp oddities |
11 |
(https://bugs.gentoo.org/show_bug.cgi?id=409445) |
12 |
|
13 |
> Before the change we had to manually take care that the corresponding |
14 |
> file system has enough inodes, as gentoo-portage seems to have very many |
15 |
> files with only a tiny size. The recent change seems to have increased |
16 |
> the number of inodes substantially. This caused our file system to run |
17 |
> out of inodes and we had to manually create a fresh file system |
18 |
> (extending the number of inodes is not possible without also increasing |
19 |
> the size of the file system). |
20 |
|
21 |
We don't suggest any particular filesystem. That is up to you to |
22 |
implement for your own needs. I'm sorry that you ran out of inodes |
23 |
(our infrastructure did too, our ramdisks became full) |
24 |
|
25 |
> To my knowledge there was no announcement for this change. With such an |
26 |
> announcement the time of having incomplete data (roughly one day in our |
27 |
> case) you may have observed for our mirror (which is part of some |
28 |
> round-robin DNS scheme in the gentoo.org domain) could have been lowered |
29 |
> to a few minutes. |
30 |
|
31 |
Your mirror is not part of the rsync.gentoo.org rotation. We reserve |
32 |
that for officially managed Gentoo hosts only. |
33 |
|
34 |
> Furthermore, it seems that your master mirror currently is overloaded. |
35 |
> My guess is that many mirrors experience problems similar to ours and |
36 |
> re-sync more often than normally. |
37 |
|
38 |
There was just a bug that was causing metadata/md5-cache to be |
39 |
resynced everytime. (https://bugs.gentoo.org/show_bug.cgi?id=410505) |
40 |
This may have explained the load, but it shouldn't have been too bad, |
41 |
that box that you access only transfers 15G/day. |
42 |
|
43 |
> |
44 |
> Please tell us ("us" as in "all the mirror admins") what happened and |
45 |
> how to deal with the situation. Please also take care to communicate |
46 |
> changes in the future. |
47 |
|
48 |
Again, sorry for the trouble(s). This was the first time that such a |
49 |
large increase in filesize happened. To be honest, we did not expect |
50 |
fallout like this. Now we know for next time, if there is a next time. |
51 |
Also, there is plans to reduce the footprint of files/inodes in the |
52 |
short term so it should be back to previous sizes in some time. |
53 |
|
54 |
-Jeremy |