1 |
Anthony Gorecki wrote: |
2 |
|
3 |
>On Friday 05 November 2004 7:17 pm, Zhang Weiwu wrote: |
4 |
> |
5 |
> |
6 |
>>My administration rule of thumb is "to keep it simple and stupid", so I |
7 |
>>hate to have all different software with same function all around. |
8 |
>> |
9 |
>> |
10 |
> |
11 |
>Using the same analogy, configuring a system to be "stupid" is synonymous to |
12 |
>asking for the very behaviors that you're trying to avoid. |
13 |
> |
14 |
> |
15 |
My English is not as good as you expect. so truly I don't understand |
16 |
what you mean. |
17 |
|
18 |
>>I |
19 |
>>configured the openldap, cryus, postfix, webapplicaton and web analizer |
20 |
>>all use berkdb, in fact many of them could be configured to use gdbm or |
21 |
>>mysql, but I hate to add administration difficulty by using different |
22 |
>>software. |
23 |
>> |
24 |
>> |
25 |
> |
26 |
>Keep in mind that these are typically configure-and-forget administration |
27 |
>requirements. Portage will keep the system up-to-date on its own. |
28 |
> |
29 |
> |
30 |
> |
31 |
Still, I don't get you. What is configure-and-forget? I have to recored |
32 |
all my configuration and plan, what I did and so like, because I am on |
33 |
my job and should be responsible. Even 1 hours unplanned working-time |
34 |
server down is not acceptable for my customers. If I didn't remember |
35 |
what I configured when things go wrong, what do I tell my boss? |
36 |
|
37 |
>>Just wish to hear your suggestions. |
38 |
>> |
39 |
>> |
40 |
> |
41 |
>Use whichever applications make the most sense for the solutions you implement |
42 |
>on your server; this does not equate to the applications which are the most |
43 |
>prestine and create the smallest possible amount of work to install -- |
44 |
>continually cutting corners will only result in an eventual car accident. |
45 |
> |
46 |
|
47 |
In my case, I would rather spend 100+ hours of extra work just to keep |
48 |
it always on. My point is stability and functionality. First I am afraid |
49 |
of having more software with same function running would add complicity |
50 |
to the server administration, thus being unstable and harder to debug, |
51 |
and I am afraid of running having many software depending on one single |
52 |
db (in my case, everything use berkdb) would make it difficult to |
53 |
recover from crash and to debug reason of the crash. |
54 |
|
55 |
I don't understand what you are suggesting. Let's go straight to the |
56 |
point, just let me know if you are in my position what will you do. In |
57 |
my case openldap, postfix, cyrus, web-application, web-analyzer all use |
58 |
berkdb for its data, and today server crashed because of berkdb data |
59 |
corruption (and recovered with yesterday's backup). The reason of the |
60 |
crash I am not sure yet (looks like because of berkdb), but the company |
61 |
lost money from this crash because the web-service are not responding in |
62 |
crash time. Now the first 3 part ("openldap, postfix, cyrus") could be |
63 |
configured to use another db, say gdbm. I just wish to try to move the 3 |
64 |
part to gdbm. I think this could make it easier if it crash again, to |
65 |
separate parts of the problems. But before I do any thing, I wanna learn |
66 |
practical suggestions from the list. |