Date: Thu, 5 Jul 2018 18:06:11 -0700 (PDT) From: "Rodney W. Grimes" <freebsd-rwg@pdx.rh.CN85.dnsmgr.net> To: Alan Somers <asomers@freebsd.org> Cc: Wojciech Puchar <wojtek@puchar.net>, FreeBSD Hackers <freebsd-hackers@freebsd.org>, Stefan Blachmann <sblachmann@gmail.com>, Lev Serebryakov <lev@freebsd.org>, George Mitchell <george+freebsd@m5p.com> Subject: Re: Confusing smartd messages Message-ID: <201807060106.w6616Bs4049980@pdx.rh.CN85.dnsmgr.net> In-Reply-To: <CAOtMX2ijjJ5jdSU_effzY-rF9Pyg%2Bb09dmNcOZprN=dx7Sy-ww@mail.gmail.com>
next in thread | previous in thread | raw e-mail | index | archive | help
[ Charset UTF-8 unsupported, converting... ] > On Thu, Jul 5, 2018 at 12:15 PM, Rodney W. Grimes < > freebsd-rwg@pdx.rh.cn85.dnsmgr.net> wrote: > > > > On Thu, Jul 5, 2018 at 11:03 AM, Wojciech Puchar <wojtek@puchar.net> > > wrote: > > > > > > > > > > >> Rewriting suspicious sectors is useless in this day and age. HDDs and > > > >> SSDs > > > >> already do it internally and have for years. Even healthy sectors get > > > >> > > > > > > > > unreadable sectors cannot be rewritten by drive electronics as it > > doesn't > > > > know what to rewrite. it may possibly remap it but still report read > > error > > > > until some data will be written - unless giving no error and returning > > > > meaningless data is an accepted behaviour. > > > > > > > > > > But if that disk is already managed by ZFS, the pool is redundant, and > > the > > > bad sector is allocated by ZFS, then ZFS will immediately rewrite the > > > unreadable sector. > > > > ZFS, if it gets a re error, will rewrite the unreadable sector > > to a DIFFERENT block, not over the top of the bad spot. > > > > Are you sure? For read errors, I think ZFS rewrites the data in-place, so > it doesn't have to rewrite it on all other members of the same mirror/raid > group. For persistent write errors of course, it would have to move it to > a different LBA as you describe. Your right, I am not sure exactly what happens during a scrub that finds a checksum error, or encounters a low level device I/O error. I was wrongly assuming that given the COW nature of the whole system that it would never overwrite anything. I wonder if you can send ZFS into a loop with a hard write failing sector. > > > > > > > only on write it can be done properly. > > > > > > > > that the HDD/SSD won't fix itself would be a checksum error. Those are > > > >> > > > > > > > > yes and this will happen if you powerdown your disk on write. or get > > some > > > > power spike or other source of noise that would affect electronic > > > > components. > > > > > > > > > > It happens surprisingly rarely. Even on a sudden power loss, the drive > > is > > > usually able to finish its current write operation. When you run into > > > problems would be if the power loss were coincident with a mechanical > > shock > > > that knocks the head off-track, or something like that. > > > > I agree that "power failure" are rare causes of write errors, and an > > idea of how often this might of happened is look at the emergency > > retract counter, if your gettng lots of those you should try to find > > out why and stop that. Vibration has become a serious problem though, > > at todays head flight hight drives are sensitive to this, you can > > even cause a drive to do retires by yelling at it with a loud > > voice :-) Look at the "high fly" counter to see if your getting > > this issue. > > > > > > performing full disk rewrite (so not zfs rebuilds) and THEN looking at > > > > smart stats and THEN performing regular smartctl -t long will tell the > > > > truth. > > > > > > > > which usually is "drive is fine" in my practice. really faulty drive > > will > > > > QUICKLY develop new problems. > > > > > > > > > > Yeah, that should make the error go away. It takes a long time, though. > > > With a SCSI drive, you can get the exact LBAs affected with a "READ > > > DEFECTS" command. But there isn't a vendor-independent equivalent for > > > SATA, unfortunately. > > > > My bitch exactly about ATA missing this. Though there are vendor specific > > commands to get it. > > > > -- > > Rod Grimes > > rgrimes@freebsd.org > > -- Rod Grimes rgrimes@freebsd.org
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?201807060106.w6616Bs4049980>