Bug #14193
closedXIO:OSD down when write data at fuse-client use iozone
0%
Description
we recently work at ceph 9.0.3, and try to use XIO msg in infiniband environment.Firstly, we deploy 8 osds in our cluster, and mount use fuse client, then write data with iozone. It seems stably writing 24 hours. Then, we try to deploy cluster with 24 osds, and do above steps, but when write a peirod of time, some osds downed.
we use accelio 1.4, and we found some errors in osds log, which come from accelio. These errors confused us, because these will also report when cluster run normally.
We found these will appear every time and always accompany request blocked. Here is one log of downed osd in below file, core dump found in it.
If you need any other logs or have ideas of these question it would be very appreciated.
Files
Updated by Josh Durgin about 7 years ago
- Status changed from New to Won't Fix
The rdma backend for async messenger is the path forward for infiniband support at this point.
Updated by Greg Farnum about 5 years ago
- Project changed from Ceph to Messengers
- Category deleted (
msgr)