Jeff Vance <[log in to unmask]> asked:
<snip>
>This too has been the majority response. I will show PROGEN as an
accessor.
>Processes that sm_open then file (#2 type of access) are difficult to work
>with. They have no file number opened, no access method, no sharing mode.
>
>Here is an IMPORTANT consideration: if a process sm_opens a file and does
>not have any file system locks held I cannot detect details about this
accessor!
>That's life. The # of accessors in the Accessor summary line will count
the
>sm_open access and the '*' flag in formats 1 and 2 will show an sm_open
>access. However I cannot tell you its job/ses id, username, PIN, etc. If
the
>sm_open'er has one or more file system locks held then I can find out who
it
>is. Also the # of sm_open'ers can change while I am collecting data (the
number
>of normal openers, eg. fopen cannot change).
>
>Here are some conclusions based upon this fact:
>1) Some accessors will go undetected except by count. This is sort of ok
since
> they do not have the file locked, but they do have it pseudo opened
> and thus the file cannot be purged (although you may be able to rm it).
>
>2) The difference between the total # of accessors and the "O:n" count in
> the Accessor summary line is the # of sm_open'ers (at the time the
sm_open
> count was grabbed).
>
>3) STORE access can be detected and the Accessor summary line will contain
> "Storing" to indicate this. However, I cannot tell you which job/sess
or
> process is doing the store. You will have to use :showjob to figure
this
> out. (Yes, this makes one of my examples wrong.)
>
>4) I cannot distinguish a :listf's sm_open from anyone elses. Also, for
> formats -5,-3,1,2,3,4,5 the OPEN lock is held and hence info regarding
> listf's access can be shown. The remaining formats do not hold any
locks
> and thus no info other than a count is available. (Formats 0 and 6 do
not
> even sm_open the file and thus are not accessing it at all.)
OK. I suggest that this be documented *VERY, VERY* clearly.
>p.s. It has been proposed to show the domain name (or ns nodename) for
remote
> accesors rather than the dotted IP address. The reason cited is that
the
> domain name is much more useful to the system admin than then IP
> address. I had two reasons for showing the IP address:
> 1) it is faster to access than the domain name (which, in the worse
case,
> can require ~ 1 minute),
> 2) it is fixed length and fits in the display.
> I am interested in your comments on this choice.
I concur. If I want address to name translation there's probabyl a utility
I can use -- and there's no utility I suppose I can write a program that
calls gethostbyaddr().
-- Evan
--
|