Experimental Physics and
Industrial Control System

Jim Thomas <[email protected]> · Thu, 21 Mar 2002 08:51:35 -1000 (HST)

>>>>> "Steve" == Steven Hartman <[email protected]> writes:

 Steve> On Thu, 21 Mar 2002, Deb Kerstiens wrote:
 >> I'm puzzled that "mv"ing the file results in having the logServer
 >> continue to write to the file in the new location, though. Maybe there's
 >> some system magic that keeps track of the file as long as it remains
 >> open for writing??

 Steve> That's proper unix behaviour. The iocLogServer process has the open
 Steve> file handle for writing to the log file and so will continue to
 Steve> write to that file even after it is mv'ed. Give the iocLogServer
 Steve> process a 'kill -HUP' (or 'stop' and 'start' via the logServer.init
 Steve> script included in base) after you mv the file.

and

>>>>> "Brian" == Brian McAllister <[email protected]> writes:

 Brian> Not magic, just standard Unix-style file I/O.  The file descriptor
 Brian> points to an "inode", not a physical disk location.  When you "mv"
 Brian> the file, the contents of the inode are modified to point to the
 Brian> new location, but it's still the same inode.  The "mv" is invisible
 Brian> to anyone who already has the file open.

but, just so no one gets confused, "mv" of a file from one file system to
another actually does a "cp".  In that case, hmmm.... :-)  I tried a test
shell loop

    atlas ~ > while true;do
    > echo here
    > sleep 1
    > done >test.log

, moved the file to a different file system after it started, and looked at
the result.  The moved file did not change after the move.  Any further
output from the shell loop disappeared and never showed on disk.  So though
the shell had the file open, the inode no longer existed and the output was
tossed.

Jim

Experimental Physics and Industrial Control System

Experimental Physics and
Industrial Control System