how to stop tshark from writing to a temp file?

Try using dumpcap it works really well and can create multiple files for you, you can provide the time or volume after which a new file should be created as well as you can provide the max number of files that should be rotated

eg: dumpcap -i any -b filesize:153600 -b files:2000 -Z none -f port 2123 -w c:\Traces\tracesaug.snoop

Unfortunately, TShark currently doesn't support any mechanism other than temporary files to pass packets from dumpcap to it; that's bug 2743. That bug would require some work to fix.

You could, however, use flags such as -b filesize:size and -b files:nfiles with TShark as well; they're passed on to dumpcap. Pick a small value for nfiles and a not-too-large value for size, and that won't take up too much disk space - only nfiles will be kept around, each one having a size approximately as big as size.

1. I must use tshark as a service for long time 24/7.

Sadly, eventually that 24/7 will have to end, as TShark, in order to dissect packets, saves various bits of state in its address space, and runs the risk of eventually running out of "memory" (address space/swap space, in reality).

it never got stuck , but i got a very large temp file

Then you weren't using the -b flags in question, because it wasn't rotating the files.

1. I'm using the -R filter option

Then you're out of luck, as

1. in order to use the -b options, you need to be saving to a file;
2. the -R option currently isn't supported if you're not saving to a file.

1. the computer is very strong with few cpu's and lots of memory

No matter how much memory you have, with enough packets, TShark can eat up all your memory and swap space (unless you're on an OS where swap space is easily expandable, in which case it'll eat up all your disk space instead). If your computer is running a 32-bit version of TShark, TShark will probably eat up all its address space before it even gets a chance to eat up all the memory and swap space.

1. 1. I have taken the advice of using the -b filesize/num and added an output file with -w, however they don't work with -R so i'm back in/to square one :-(

And there's currently a 1km high wall around that square, unfortunately.

is there any way to make it work?

No, because...

the key is having the -b work with -R

...and you can't currently do that.

is there a way to make dumpcap output a filtered CSV text info a file by itself

No. dumpcap was explicitly designed to be incapable of doing that. It might have to run with extra privileges in order to capture traffic, so the ability to dissect packets was explicitly left out of it to reduce the amount of code in it that could cause problems if there's a bug in it that opens a security hole.

At best, you could run dumpcap with the -b options to have it produce a sequence of capture files, but not remove them. You could then try running a script that watches for new capture files to appear and, when a new file appears, runs TShark on each of the older capture files, removing each one after it's processed. Note, however, that this means that if, in order to dissect a packet in a particular file, information from a packet in an earlier file is required, the later packet won't be able to be fully dissected, so, if the fields you're using with -e require a full dissection, you might not get the information you want. (The information in question is part of the stuff that eventually will fill up TShark's address space.)

What information are you trying to get? There might be other programs that can get that information for you.

I'm capturing RADIUS traffic and i split it into CSV fields like username,disconnection reason,packets used, etc.

In that case, I recommend a radius sniffer, like one of these.

http://sourceforge.net/projects/raddump/
http://www.manpagez.com/man/8/radsniff/

Combined with some scripting, you should be able to achieve your goal.

Regards
Kurt