Re: Mirroring

In article <3313D72E.77A70447@www.weideneder.de>, Stefan Weideneder <stefan@www.weideneder.de> writes >cburton@gte.net wrote: >> >> I am currently on 7.11.UC1 on HP-UX 9.04 >> >> To experiment (before implementing), I created a mirror of the >> root_dbs. Both the root and the mirror are seperate logical (and >> physical) devices. The links to the devices were created in >> /usr/informix/dev. t_root_dbs and t_root_mir >> >> After recovery of the root had taken place (watched the log), I did a >> level 0 to make sure everything was in sync. Then I deleted the >> symbolic link in /usr/informix/dev to t_root_dbs so it would be on the >> mirror. I then waited for onstat -d or the log to report the root_dbs >> was down. I then did update stats on a table. Dropped and recreated >> another table. Created a stored procedure. (All should of had >> activity on the root_dbs space). onstat -d showed reads and writes >> from the now (supposedly missing) root_dbs. Again I waited. Two >> check points; another hour. NO MESSAGES that I now have a (simulated) >> device error. I called INFORMIX help, and their answer was that I had >> to bounce the server (at least once a week, according to their >> guidelines) in order to find a problem. If I have a 24x7 system, & >> cannot bounce the server, the mirroring is a moot point. >> >> Someone please tell me this is wrong. >> >> TIA >> >> Chris > >Hi Chris, > >I think you should contact your UNIX support or just read the manual >entry of "rm". It's only removed from your directory, but it's still >there. A file will not be deleted as long as a process has >an open connection (read/write) to this file. It's very difficult >to "simulate" a device error. If you want to simulate a device error >you must force the UNIX calls (read/write/lseek/tell) to return the >value -1. Otherwise it's nearly impossible, because Informix will >open it's chunks only at startup time. >Maybe it's possible to switch off your disks, or make use of a floppy >disk to simulate the device error. > >Have a nice day, > >Stefan Correct. rm removes the directory entry note the file. UNIX top of the day: way to create temporary file which a) cannot be overwritten. b) is automatically deletes once the program dies e.g. if it dies with a core dump or kill -9:- fd1=open("/tmp/x","a"); /* create file */ unlink("/tmp/x"); /* remove directory entry */ /* From here on the file cannot be accessed by anything but this process i.e. other processes can create file /tmp/x but it will use a different inode and hence be a separate file!! */ The only way to do what you want to achieve is to pick a large file e.g. oninit and dd the first 100K onto the start of the mirror chunk i.e. dd if=oninit of=<chunk path> bs=1024 oseek=<offset of start of chunk in K> count=100 This is :- if=oninit : input file = oninit of=<chunk path> : output file = your chunk bs=1024 : input and output block size is 1024 bytes (1L) oseek : skip first n blocks of <chunk path> Note to get offset a) use onstat -b and look at end of output to get pagesize in bytes. b) use onstat -d to get start offset for chunk in pages calculate (offset in pages) x (pagesize/1024) to get the value to use here. count=1000 : copy 100 blocks (in this case 100K). I have done this before and it works (Online 5.07.UC1 on Solaris 2.5). Yes I got wrong once and used seek= rather that oseek=. Yes it skip block from the input rather than the output. Yes I overwrote the VTOC on the disk and it had used filesystems on it Yes I got virtually thrown into a car and driven on-site at stupid speeds by a salesman to fix the problem No I will never get it wrong again PS Check the above dd command before using it. -- David Williams