Note: "permalinks" may not be as permanent as we would like,
direct links of old sources may well be a few messages off.
Lars Ellenberg a écrit : >> GAUTIER Hervé a écrit : >> >>> Hi, >>> >>> I Would like to understand well the following message which appears to >>> me during a resync: >>> local disk flush failed with status -95 >>> >>> errno.h seems to say: >>> #define EOPNOTSUPP 95 /* Operation not supported on >>> transport endpoint */ >>> >>> After some research, I have understood that this is in relation with >>> "I/O barriers". >>> >>> If I've got this message, there is something (OS, driver, disk, ...) >>> in my system which does not support "I/O barriers", does it ? >>> > > right. > drbd cannot use barriers where it would want to. so it falls back to > instead "let IO drain and hope for the best". drbd used to do that > anyways up to 8.0.9 or so, and assumed you have a "safe" device > (no volatile caches involved). > > my recommendation: > get a decent raid controller with a battery backed write cache, > switch off all volatile caches in the chain (on the single drives), > and forget about the whole issue. > > btw, drbd itself does not yet support barriers for upper layers either. > but we are working on that... > Thank you very much for the explanation and advices. How could I know from where comes the barrier problem ? I have identified the following different level of the chain that may be involved: - OS: I am currently running on Red Hat Enterprise Linux ES release 4 (Nahant Update 6) kernel 2.6.9-67.ELsmp. Is there a minimum required kernel version for the barriers implementation ? - DRBD: The current version I am using is drbd-8.2.5. - Drivers: I currently use disks connected on a HP Smart Array 642 Controller using the cciss driver 2.6.16 provided by the Red Hat Enterprise Linux ES release 4 (Nahant Update 6). I haven't seen any option on this driver. - Controller: I will check in there is any cache option in the HP Smart Array 642 Controller firmware. - Disks: I will try to find any information on the disks used in order to check if the used some cache. Anything else I have forgotten ? Thank in advance, Best regards. -- Hervé GAUTIER