On Fri, Mar 13, 2015 at 8:31 AM, Ryan Thompson <agyant@xxxxxxxxx> wrote: > On Fri, Mar 13, 2015 at 7:01 AM, Scott Marlowe <scott.marlowe@xxxxxxxxx> > wrote: >> >> On Fri, Mar 13, 2015 at 7:55 AM, Achilleas Mantzios >> <achill@xxxxxxxxxxxxxxxxxxxxx> wrote: >> > On 13/03/2015 15:47, Scott Marlowe wrote: >> >> >> >> On Fri, Mar 13, 2015 at 7:16 AM, Achilleas Mantzios >> >> <achill@xxxxxxxxxxxxxxxxxxxxx> wrote: >> >>> >> >>> On 13/03/2015 13:40, Joshua D. Drake wrote: >> >>>> >> >>>> On 03/13/2015 04:27 AM, Achilleas Mantzios wrote: >> >>>>> >> >>>>> >> >>>>> Hello, >> >>>>> >> >>>>> we maintain a DB of nearly 500GB of data (and always getting >> >>>>> larger), >> >>>>> and we are currently thinking of moving to SSD. >> >>>>> I have read Greg Smith's book on PostgreSQL 9.0 High Performance and >> >>>>> his >> >>>>> considerations on SSD and the way that write back works. >> >>>>> >> >>>>> This particular model (Samsung SSD 850 PRO 1T) does not employ any >> >>>>> special circuitry, battery or capacitor >> >>>>> to enforce that the data are really flushed to the medium. >> >>>>> >> >>>>> What is your take on this? Is it dangerous to have PgSQL on this >> >>>>> disk >> >>>>> especially in cases of power outages? (we have full UPS support, >> >>>>> however >> >>>>> nothing can be overlooked, anything can happen) >> >>>> >> >>>> >> >>>> If it does not have power loss protection, don't use it. >> >>> >> >>> >> >>> Thanx Joshua. >> >>> >> >>> If theoretically somehow we eliminate the power loss factor, would it >> >>> make >> >>> sense to use such a disk? >> >> >> >> Sadly there's no way to eliminate power loss in real life. >> >> >> >> Now if you can live with some data loss corruption that's a different >> >> matter. >> > >> > >> > Thanx, >> > the question is if postgresql will recover, we can tolerate some data >> > loss, >> > but no total db corruption, equivalent to fsync=false, with the db >> > unable to >> > recover. >> > >> >> Sadly that is the exact thing that is likely to happen. >> >> P.s. I worked in a BIG data center that lost all power. Three power >> conditioners, three USPs and the switch for the diesel gen all blew >> out at once. Lots of dbs that relied on never losing power were >> corrupted and took days to recover most of their data and get back >> online. >> > Personally, I'd steer clear of Samsung desktop SSDs for anything > mission-critical. I've seen them get ripped apart by standard SATA > controllers as well as LSI MegaRAID. My rule of thumb is Samsung is okay as > long as we're using RAID 1. > > For data loss, pick up a RAID controller card with BBU or capacitor-based > cache protection. That way, in the event of a total power failure, the data > integrity is maintained and simply continues writing to disk once power is > restored. That won't save you from SSDs that aren't safe if they lose power. The problem is that the SSDs that don't have the ability to finish writes when they lost power act as though they do. They respond positively to fsync requests while not actually writing your data out to the memory cells. So if you lose power, the RAID controller thinks that data is already written when it's not. Simple rule, avoid SSDs that don't have super caps or some other method to write out your data safely in the event of a power failure. -- Sent via pgsql-admin mailing list (pgsql-admin@xxxxxxxxxxxxxx) To make changes to your subscription: http://www.postgresql.org/mailpref/pgsql-admin
