1 |
TL;DR summary: |
2 |
Yes, stuff has broken, but I'd call them reasonable teething issues well |
3 |
distributed through the stack, and to be compared to the CVS server |
4 |
moves from a decade ago, rather than CVS just before the Git switch. |
5 |
|
6 |
On Sun, Dec 13, 2015 at 06:36:41PM +0100, Patrick Lauer wrote: |
7 |
... (mail re-ordered for related issues) |
8 |
> Once that was 'fixed' there was the fun of [2], which made emerge --sync |
9 |
> very expensive because it refetched lots of files. Every time. |
10 |
The fix for this caused these two bugs: |
11 |
> And fixing that introduces [7] some more regressions that broke updating |
12 |
> @system for about 3.5 days. |
13 |
> - a few days of grub being uninstallable (iow, making installing |
14 |
> impossible for many users) |
15 |
> And the manifest issues are still [9] making life exciting. |
16 |
Bug #567920 describes the issue very succinctly (mtime of a deleted file |
17 |
needs to be included in the new Manifest mtime calculation). |
18 |
|
19 |
Both of them can be worked around if the entire path (all staging nodes, |
20 |
servers and end users) uses --checksum, but that's even more expensive. |
21 |
|
22 |
I have another work-around idea, and that's simply appending a comment |
23 |
of the latest commit per directory to the changelog, because that will |
24 |
trigger the manifest being different ;-). |
25 |
|
26 |
> The fix to that fix (notice a pattern here?) broke rsync for *all* users |
27 |
> [8]. Almost as if no one ever tests things in a test environment ... but |
28 |
> hey, we're agile, let's fix stuff in production! |
29 |
We still never figured out how .git came to be added to the outgoing |
30 |
data. It was NOT the rsync into staging directory, because it was only |
31 |
the directory structure, and none of the files. --exclude=.git WAS used |
32 |
in most of the ryncs, but not the final ones from staging to tree |
33 |
distribution. |
34 |
|
35 |
> - about 3 months of emerge --changelog being broken, just to be broken |
36 |
> in a different way |
37 |
This change (order of changelog entries) was explicitly to reduce your |
38 |
complaint of prior excess traffic. Why Portage's parsing of the |
39 |
ChangeLogs is still not handling it is an open question. |
40 |
|
41 |
> - 3.5 days of emerge @system being broken |
42 |
> - about a day of emerge --sync needing manual interaction to be able to |
43 |
> update again |
44 |
You missed one: |
45 |
- rsync generation now halts if somebody committed some breakage to the |
46 |
tree (missing DIST entry, bad eclass). |
47 |
|
48 |
> |
49 |
> So all in all emerge --sync && emerge -uND @system being down for >10% |
50 |
> of the time. |
51 |
> |
52 |
> Now, I don't know if you use Gentoo, but I do, and I use it at work, so |
53 |
> having this level of randomization happen is not really useful. |
54 |
> |
55 |
> Tell me then, please - what can I/we do so that this kind of breakage |
56 |
> stops, and we can actually aim at having a most excellent distro? In the |
57 |
> long run I am considering just creating my own clone of all |
58 |
> infrastructure bits so I can fix things, but it's an option that is |
59 |
> needlessly braindead, wasting effort, and not really useful to users |
60 |
> that are not me. |
61 |
Your own infra option would NOT have fixed [2]/[7]/[9]. |
62 |
|
63 |
-- |
64 |
Robin Hugh Johnson |
65 |
Gentoo Linux: Developer, Infrastructure Lead, Foundation Trustee |
66 |
E-Mail : robbat2@g.o |
67 |
GnuPG FP : 11ACBA4F 4778E3F6 E4EDF38E B27B944E 34884E85 |