i guess my thinking is more along the lines of implementing a lustre interface
http://en.wikipedia.org/wiki/Lustre_%28file_system%29#Architecture where the chronology *may* follow: the File's Metadata attributes are written the particulars of where the data is written would be handled by OSS which delegates to 1..n target nodes (which then passes the information to that target's LVM / RAID device) Both MDS and OSS can implement either ext3 or ZFS/DMU Storage algos..in either case the metadata(MDT) metadata object record is written by the MDS .. the data will be written to one or more LOV formatted ext3 or ZFS/DMU nodes specifying byte-offset and size.. When the requesting client issues a write request for the OST the governing OSS issues the write request to the target that can fulfill it..if the requested target cannot complete the request that target passes ENOSPC back to OSS which then looks up the next target The admin of the target node will be notified of the failed attempt by alert or log but the OSS will hunt for the next target that can fulfill the write request of the OST this is my (albeit cursory) interpretation of Object Oriented Disk Architectures does this conform to your understanding? Martin Gainty ______________________________________________ Please do not alter/modify or disrupt this transmission. Thank You > Date: Fri, 21 May 2010 07:21:22 -0700 > From: t...@soe.ucsc.edu > To: mgai...@hotmail.com > CC: mysql@lists.mysql.com > Subject: Re: Database Quotas > > > if MYSQL attempts to insert more bytes than what is available > > on disk you will get 28 ENOSPC No space left on device > > http://dev.mysql.com/doc/refman/5.0/en/operating-system-error-codes.html > > Does it figured that out before it tries to write a record? So, if I have 2KB > left on the device and I write a 4KB record, does the first 2KB get written > and then the error occurs, or does the error occur before the write is > attempted? > > I guess what I'm asking is will the tables be marked as "crashed" when an > ENOSPC happens, or will the tables still be in good health? > > If they're still in good health, then I suppose that I could use ZFS file > systems to allocate space for databases...it just seems that this ought to be > a feature of the database. :) > > Tim Gustafson > Baskin School of Engineering > UC Santa Cruz > t...@soe.ucsc.edu > 831-459-5354 _________________________________________________________________ Hotmail is redefining busy with tools for the New Busy. Get more from your inbox. http://www.windowslive.com/campaign/thenewbusy?ocid=PID28326::T:WLMTAGL:ON:WL:en-US:WM_HMP:042010_2