| 1 |
Sebastian Pipping (sping@g.o) wrote on 3 May 2010 22:25: |
| 2 |
>even after reading [1] I keep wondering what happens if a user syncs |
| 3 |
>against our Gentoo e.V. rsync1 mirror while that one is syncing itself |
| 4 |
>against the master server. With bad luck a user may end up with mixed |
| 5 |
>files from before and after the sync (i.e. an inconsistent, possibly |
| 6 |
>broken tree), or not? |
| 7 |
|
| 8 |
Yes. |
| 9 |
|
| 10 |
>How bad would shutting rsyncd down for the moment and bringing it back |
| 11 |
>up after be? |
| 12 |
|
| 13 |
VERY BAD :-) |
| 14 |
|
| 15 |
>Is there a cleaner solution to this? |
| 16 |
|
| 17 |
Mirrors should use --delay-updates --delete-after when they update. |
| 18 |
It's not perfect but good enough. |
| 19 |
|
| 20 |
It seems the portage tree is very small (we only mirror distfiles so |
| 21 |
I'm not sure). If so, there's a truly atomic way to update using |
| 22 |
fancy rsync options. The atomic-rsync perl script, in the support |
| 23 |
directory of the rsync source, implements it. |
Archives