1 |
Apparently, though unproven, at 23:21 on Tuesday 02 November 2010, Volker |
2 |
Armin Hemmann did opine thusly: |
3 |
|
4 |
> On Tuesday 02 November 2010, Alan McKinnon wrote: |
5 |
> > Apparently, though unproven, at 20:19 on Tuesday 02 November 2010, Volker |
6 |
> > |
7 |
> > Armin Hemmann did opine thusly: |
8 |
> > > On Tuesday 02 November 2010, Stroller wrote: |
9 |
> > > > On 2/11/2010, at 10:46am, Alan McKinnon wrote: |
10 |
> > > > > ... |
11 |
> > > > > hard links will only work if /etc/portage and /var/lib/portage are |
12 |
> > > > > on the same filesystem. Frequently, they are not. |
13 |
> > > > |
14 |
> > > > For small values of frequently. |
15 |
> > > > |
16 |
> > > > Stroller. |
17 |
> > > |
18 |
> > > for every sane system out there. |
19 |
> > > |
20 |
> > > /var is a candidate for surprisingly filling up / to 100% so it is a |
21 |
> > > smart and sane choice to put it on its own partition where damage will |
22 |
> > > be reduced to some log files or an aborted emerge. |
23 |
> > |
24 |
> > You're both right, but for different reasons. It'd done less often on a |
25 |
> > laptop or personal machine than on a server for instance. And on embedded |
26 |
> > stuff, almost never. Example: Any junior of mine who doesn't make /var |
27 |
> > separate is liable to be served his own testicles for dinner, and they |
28 |
> > know it. But my laptop is one big filesystem. One case definitely needs |
29 |
> > it, the other one doesn't really. |
30 |
> > |
31 |
> > You're probably looking at the same question from entirely different |
32 |
> > needs and viewpoints. |
33 |
> |
34 |
> I am looking at the question from the viewpoint of a person who was hit |
35 |
> very hard in the past. Surprise / fillup thanks to /var or /tmp is no fun |
36 |
> at all. |
37 |
|
38 |
I feel your pain. I know it well. That's why I mentioned roasted testicles. |
39 |
|
40 |
Right now I sit with 60+ SLES 9 machines that cannot be taken offline for any |
41 |
reason, and EVERY SINGLE ONE has one giant filesystem except for the database |
42 |
partitions - those are /dev/sdb in a RAID. |
43 |
|
44 |
I cannot fix this and still maintain my SLA because |
45 |
|
46 |
a) you can't reduce a mounted fs |
47 |
b) you can't umount / |
48 |
c) all disk bays are full |
49 |
d) I don't have budget for bigger replacement drives |
50 |
e) there's no way I'm sitting in that freezing data centre for a week fiddling |
51 |
with disks, breaking RAID, putting bigger drives in, rebuilding RAID, fdisk, |
52 |
mkfs, blah, blah, blah. And with my luck, all of those machines will decide |
53 |
the stress of pulling drives will cause others to fail just at the exact point |
54 |
I don't have redundancy. |
55 |
|
56 |
How did this happen? The man in charge three managers ago thought this was a |
57 |
cool way to configure critical servers. Because "One filesystem mounted at /" |
58 |
was option #1 on the disk page of the SLES install wizard. |
59 |
|
60 |
And lets not talk about the abuses /tmp can be subject to... |
61 |
|
62 |
<sigh> |
63 |
|
64 |
<rant over> |
65 |
|
66 |
|
67 |
-- |
68 |
alan dot mckinnon at gmail dot com |