🚀 go-pugleaf

In article <1992Nov11.195233.20996@leland.Stanford.EDU>  
gcolello@biosphere.Stanford.EDU (Greg Colello) writes:
=>Paul:
=>
=>Thanks for the very detailed response to my detailed request. Below I  
=>answer your question about how we make the mail clients "FROM:" field have  
=>the server's address. I also have lots of questions in response to your  
=>answers:
=>
=>Greg
=>

Sure, my pleasure. And thanks for the tip on sendmail...

=>---------------------------------------------
=>In article <1992Nov11.152052.8171@menudo.uh.edu> sears@tree.egr.uh.edu  
=>(Paul S. Sears) writes:
=>
=>> =>8. Make the server the mail server.
=>> 
=>> Good.  export /usr/spool/mail.  Make sure sticky bit is set.  Chmod 1777  
=>> /usr/spool/mail.
=>
=>But you're exporting / as you explain below. I thought you couldn't export  
=>a child path if its parent was already exported (more below). What's this  
=>about setting the sticky bit?
=>

Ah, the rub.... You are indeed correct, in one sense.  I think I need to make  
a distiction between exporting and mounting.  We export our / partition (we  
have two on one disk / and /clients) as rw, but everyone mounts it as ro.  All  
the clients mount the various directories in various ways.  Our entry for the  
mail directory looks like this (imagine you were doing this by hand instead of  
in Netinfo..._

#mount mailhost:/usr/spool/mail /usr/spool/mail nfs rw,bg,intr,noquota 0 0

Our other single mountpoints (remember I suggested that you have as few  
mountpoints as you can get away with and just link everything to that  
mountpoint?) looks like this:

#tree:/ /NeXTMount nfs ro,bg,intr,hard,nosuid,noquota,mnttimeo=5 0 0

(note the HARD mount.  Soft mounts are not reliable...)

Then on all our clients links were made as such:

/LocalApps -> /NeXTMount/LocalApps/@
/LocalDeveloper -> /NeXTMount/LocalDeveloper/@
/LocalLibrary -> /NeXTMount/LocalLibrary/@

Everything goes through a single mount point.  We have reduced our total  
number of mount points to the following:
/usr/spool/mail (server 1) via hard mount
/Users (server 1) via /Net
/Users (server 2) via /Net
/Users2 (server 2) via /Net
/NeXTMount  (server 1) via hard mount
/NeXTMount2 (server 2) via hard mount
/usr/spool/NeXTFaxes (server 2) via hard mount
/CD-Rom (client) via /Net
/Misc (client2) via /Net

=>---------------------------------------------
=>> =>My first question is how many of the step 5 directories need  
=>Read/Write  
=>> =>access? For example I know that /usr/spool/mail-->/private/spool/mail  
 
 
 
=>> 
=>
=>Doesn't /usr/spool/mail need to be writable? I think so for users who  
=>prefer to use Unix command line mail (we have some).
=>
=>Whoa. Partition? Who said anything about a partition? We have no  
=>partitions on our disks. Is that how you are able to export the root as  
=>read only and yet also export children like "/usr/spool/mail" and "/Users"  
=>as read/write?
=>
=>BTW the 3.0 installer apparently doesn't give you the ability to create  
=>partitions when you use its "complete install" mode.
=>

Er, I have a bad habit of referring to everything as "partitions".  Ok.. so  
you don't have multiple partitions... but you still have 1 partition :-)

Yes, /usr/spool/mail needs to be writable.  Check above how I made an attempt  
to clarify exporting and mounting...

=>> be _hard_ mounts.  Having things in /Net via automounter proved to be  
=>way too  
=>> unreliable when the server gets bogged down with NFS requests.  Also, in  
=>your  
=>
=>What is it with this /Net thing anyway? I want to understand what Next  
=>intended before I assume it's not for me. I've been studying the way it  
=>works and it is slightly confusing. I can only surmise that it was  
=>intended for read only mounts by the following line in /etc/mtab for the  
=>client stoma (placed there as a consequence of Netinfo setup on the server  
=>biosphere):
=>
=>stoma:(autonfsmount[87]) "/Net" nfs ro,intr,port=686 0 0
=>
=>Now, when the exported path "biosphere:/Users" is mounted on stoma, it is  
=>done as follows (also from /etc/mtab on the client):
=>
=>biosphere:/Users "/private/Net/biosphere/Users" nfs rw,bg,intr,noquota,net  
=>0 0
=>
=>The server's name is slipped into the path. Netinfo set that up. As part  
=>of the mount a soft link /Net/biosphere to /private/Net/biosphere is  
=>created. What causes this to be done? Autonfsmount? Also how does  
=>/etc/mtab get updated with mounts that I can't find in /etc/fstab? How are  
=>these /Net mountings (such as /Users) being directed?
=>
=>Now while /Net is read only, which is also obvious from its permissions  
=>dr-xr-xr-x, I assume that read/write directories (like /Users) can still  
=>be mounted to it and accessed as read/write by making a soft link /Users  
=>to /Net/biosphere/Users for example? I get this idea from observing how  
=>/usr/spool/mail appears to give universal access to the restricted path  
=>/private/spool/mail.
=>

/Net is what automounter uses to have dynamic mounts.  These are soft mounts  
that can come and go as the exporting host comes and goes.  /Net is the  
primary place you would mount the /User partitions/directories as it is a  
place that would _look_ the same no matter what client you were on.  For  
example, if a user's home resides on a server, it would be  
/Net/server/Users/user's_home on all the clients.  If you had other users on  
other servers, they would be /Net/serverxxx/Users/userhome.

It basically provides a consistant place to mount things.  However, you do not  
have much control on things that are mounted via automounter (ie., you can't  
unmount them manually) because automounter handles those details.  If a server  
crashes, the directories in /Net will disappear, but they will re-appear as  
soon as the server comes back up (a benefit of using /Net).  However, I only  
recommend using /Net for /User home directories.  Static, hard mounts are much  
more reliable for everything else.  

=>---------------------------------------------
=>> /etc/rc, up the # of nfsd processes from 6 to at least 8.
=>> 
=>
=>Why is this be unreliable? It isn't clear to me that /Users is  
=>automounted. Just /Net. Why this bump up in processes?
=>

/Users is exported from your home directory server and the clients mount  
/Users via /Net, so it looks like /Net/server/Users/xxxx on all the clients  
(it is a consistant location.  Home directory does not change regardless of  
client).

=>---------------------------------------------
=>> And to address a particular misconception:
=>> 
=>[deleted stuff]...
=>
=>> for the wrong reasons.  NetBoot clients are a _severe_ drain on a  
=>server's  
=>> resources.  Try to avoid NetBoot clients if possible.  We have 11 (cubes  
=>with  
=>> 40M accelerator drives) out of 50 clients that are net boot.  They are  
=>mostly  
=>> serviced by our secondary server which has mucho process cycles to burn,  
=>but  
=>> even that server has its performance degraded when the netboots are in  
=>heavy  
=>> use...  
=>> 
=>
=>I don't get it. Why should Netbooting (with local swap disks) be any more  
=>of a stain on the server than having clients with their own OS and  
=>exporting root to them? Are you saying that serving the OS kernal is a  
=>bigger task than serving root? I would think that virtual memory would  
=>solve that kernal serving problem if the local swap disks were actually  
=>working.
=>

Er, well, think about it.  First, a Netboot clients has _all_ of its  
directories on the server (like /private/etc...) which means alot of nfs calls  
to do anything basic (even stuff in /bin, /usr/bin).  On a machine with  
minimal disk (105M with 73M used for NS), /bin, /private/etc, /usr/bin, and so  
on are local.  Much faster, no need to bog the network down with nfs traffic.

Secondly, if set up incorrectly, netboots will swap to the _server_.  That  
means that it is swapping across the network... with limited bandwidth  
already, this is a serious bad thing!

=>-----------------------------------------------------------------
=>BTW The new NFS Manager app seems to work great. It took me a while to  
=>figure out what order it wanted me to do things in, but once I did it  
=>seemed to greatly simplify the task. I think it may have a great future.
=>

I need to warn you about a slight (nothing serious, but can be frustrating)  
bug in NFSManager.app.  There is a little check box under the export  
filesystem panel that has to do with treat all unknown users as uid xxx.  Do  
not uncheck this box.  If you do, all mount requests to your server will be  
denied...

=>-----------------------------------------------------------------
=>Greg Colello
=>Carnegie Institution, Department of Plant Biology
=>Stanford University
=>gcolello@biosphere.stanford.edu

--
Paul S. Sears                *  sears@uh.edu (NeXT Mail OK)
The University of Houston    *  suggestions@tree.egr.uh.edu (NeXT
Engineering Computing Center *  comments, complaints, questions)
NeXT System Administration   *  DoD#1967 '83 NightHawk 650SC 
          >>> SSI Diving Certification #755020059 <<<
"Programming is like sex: One mistake and you support it a lifetime."