1 |
On 11/02/2014 13:07, Walter Dnes wrote: |
2 |
> On Tue, Feb 11, 2014 at 07:32:46AM +0200, Alan McKinnon wrote |
3 |
> |
4 |
>> emerge --sync X number of times daily in a cron |
5 |
>> |
6 |
>> emerge -avuND world deliberately and manually as the sysadmin at your |
7 |
>> leisure. |
8 |
>> |
9 |
>> Two different actions in time. |
10 |
> |
11 |
> Assume you sync once a day, and update once a week, the first 6 syncs |
12 |
> would be mostly wasted. Yes, your final sync would be smaller. But |
13 |
> your machine and the server would have to go through the file-comparison |
14 |
> process 7 times, instead of once. |
15 |
> |
16 |
|
17 |
So what? rsync is cheap and doesn't stress the server unduly. |
18 |
|
19 |
It doesn't check every object in the directory tree and stat 179680 |
20 |
files/dirs every time, the whole thing is hashed and it's the hash |
21 |
values that are compared. To compare a directory, rsync only needs to |
22 |
look at the directory inode, if they match on both ends then it's a |
23 |
certainty the files match. It's a *very* efficient system, all done |
24 |
in-memory, your average server can deal with many connections and not |
25 |
even break a sweat. |
26 |
|
27 |
If you want to minimize load, concentrate on making emerge world more |
28 |
efficient so that it takes less than 3 minutes to depgraph on a fast cpu |
29 |
and up to 40 on a slow one. Coupled with overlays, more often than not |
30 |
the portage cache is invalidated so emerge ends up sourcing almost every |
31 |
ebuild file in the tree. |
32 |
|
33 |
--sync is not something worth optimizing if done once a day. At that |
34 |
frequency you are well below the noise floor |
35 |
|
36 |
|
37 |
-- |
38 |
Alan McKinnon |
39 |
alan.mckinnon@×××××.com |