1 |
On 11/27/2018 09:37 AM, Sam Pfeiffer wrote: |
2 |
> On Tue, Nov 27, 2018 at 7:20 PM Fabian Groffen <grobian@g.o <mailto:grobian@g.o>> wrote: |
3 |
> |
4 |
> > I don't want to depress this entire discussion, but it would be really |
5 |
> > nice if we could somehow interact with special machines people have at |
6 |
> > their company or at home. Prefix needs testing on many different |
7 |
> > machines (non-Linux) which usually don't exist in docker images. |
8 |
|
9 |
I second this - and let me add a further aspect here: |
10 |
What I know from buildbot setup is that the master does provide (mostly shell) |
11 |
commands to be executed on the slave. This is fine as long as there is limited |
12 |
visibility for the master. But when a public buildbot master is being hijacked, |
13 |
it feels too easy to execute malicious commands even on the slave machines. |
14 |
|
15 |
So over a buildbot like setup, I would prefer a Jenkins like setup, where the |
16 |
master does provide only trigger information to slaves. And even more appealing |
17 |
would be a standalone slave setup, where the master does just receive the build |
18 |
logs for the public, without access to slave machines at all. |
19 |
|
20 |
> That's alright, we can use QEMU for some more esoteric hardware platforms,> if it's an OS that runs on a normal amd64/x86 computer a Docker image can be |
21 |
> built (I'm not an expert but there are images to learn how to do it).> Or in the worst case we can create an old-school VM for those weird OS |
22 |
> and automate the interaction with it (I did it for a robot by dumping all |
23 |
> disk once and creating a VM from it, it worked ok). |
24 |
|
25 |
Well... there's a bunch of OSs I fail to imagine the use of cloud driven |
26 |
hardware for them, like hppa-hpux or ia64-hpux for past ones, and ppc-aix, |
27 |
ppc-macos, sparc-solaris, arm-linux or m68k-mint for current ones. |
28 |
|
29 |
> > That said, focussing on the (usually fast) boxes like this to catch |
30 |
> > dependency problems and more is useful. In the case below it looks like |
31 |
> > the ld-wrapper is having issues. Would it be possible to enter the |
32 |
> > environment for that failed run? |
33 |
> |
34 |
> Glad you see the use of it :) Yeah as I mentioned in the previous mail, |
35 |
> having docker installed in your machine, to access that exact environment |
36 |
> after the failed bootstrap just do: |
37 |
> |
38 |
> # This will download the image to your machine (it may take a bit of time if its the first time you use docker its around 1GB of data I think) |
39 |
> docker pull awesomebytes/gentoo_prefix_latest_image |
40 |
> # This will drop you into a shell in that environment, ready to play! |
41 |
> docker run -it awesomebytes/gentoo_prefix_latest_image /bin/bash |
42 |
> |
43 |
> When you are done you can just type exit. |
44 |
|
45 |
Nevertheless, having the breaking environment as a docker image where |
46 |
possible (true for the major OSs we support) really is awesome! |
47 |
|
48 |
/haubi/ |