On 2010/01/03 16:40, Piavlo wrote: > 1) First as you can see below the "Use%" field is not calculated correctly > > fire-srv4 ~ # shepherd info -t sheep > Id Size Used Use% > 0 5G 9G 181% > 1 2G 12G 643% > 2 8G 6G 83% > fire-srv4 ~ # > > I think it is obvious from the output exactly how wrong the "Use%" gets > calculated :) Oops, the total size seems to be wrongly caluculated. The folowing patch will fix the problem. diff --git a/collie/store.c b/collie/store.c index fce71ff..582b724 100644 --- a/collie/store.c +++ b/collie/store.c @@ -60,8 +60,8 @@ static int stat_sheep(uint64_t *store_size, uint64_t *store_free) used += s.st_size; } - *store_size = vs.f_frsize * vs.f_bfree; - *store_free = vs.f_frsize * vs.f_bfree - used; + *store_size = vs.f_frsize * vs.f_bfree + used; + *store_free = vs.f_frsize * vs.f_bfree; return SD_RES_SUCCESS; } > 2) I've created a 15G btrfs on 3 nodes and --copies=2 for the test and > wanted to fully fill just one of the nodes and see how sheepdog > behaves then. > I hoped that once one of the nodes gets out of space it would continue > to allocate blocks on the other 2 remaining nodes, but it just > got stuck with "aio_read_response 869: I/O error" Thanks for log messages. Sheepdog lacks a lot of error handling, and ENOSPC is also one of them. I really think we should improve this soon. > Also sheepdog show that zopa4 vdi is in running state in spite of that > i've terminated the kvm-img process: > fire-srv4 ~ # shepherd info -t vm > Name |Vdi size |Allocated| Shared | Status > ----------------+---------+---------+---------+------------ > zopa | 5120 MB| 4068 MB| 0 MB| not running > zopa2 | 5120 MB| 4068 MB| 0 MB| not running > zopa3 | 5120 MB| 4068 MB| 0 MB| not running > zopa4 | 5120 MB| 2848 MB| 0 MB| running on 192.1.1.4 You can mannually release the image from shepherd. $ shepherd debug -o release_vdi zopa4 But, of cource we should check whether client is alive or not. Thanks, Kazutaka Morita |