Save to kill rpcd ?

[Ralph Grothe]

Hi,

I need to extend the /opt filesystem on 10.20 system that is also SG/MC cluster with webserver as package.
I cannot bring down the webserver, nor switch it over to its peer node.
After having extended the underlying LV of /opt I wanted to run the usual fsadm -F vfxfs -b ... OnlineJFS filesystem extending,
when it moaned that there is no vxfs.
True, a fstyp revealed /opt to be hfs,
why I have to unmount it to extendfs.
I downed all unacessary jobs that keep the FS busy such as the measure ware daemosn.
Now there is an isolated rpcd (why can't they do a cd / befor daemonizing?)


# fuser -cu /opt
/opt: 675mt(root)

# ps -fp 675
UID PID PPID C STIME TTY TIME COMMAND
root 675 1 0 Sep 12 ? 95:05 /opt/dce/sbin/rpcd


Now I'm not sure if I can safely kill this RPC daemon without interfering with the cluster package.
There are no NFS services used on this cluster.
So who else is using RPCs that can be intercepted by an undue kill?

[Pete Randall]

Ralph,

This is certainly not definitive, I'm just trying to provide some info for you to think about. A perusal of the rpcd man page indicates that this is primarily associated with DCE, though NCS is mentioned as well. I honestly don't know whether DCE, or Distributed Computing Environment, is involved in MC/SG or not, but it seems to me that it certainly could be. Hmmmmm . . .


Pete

[Dave Olker]

Hi Ralph,

NFS does not use the /opt/dce/sbin/rpcd daemon. ONC (Open Network Computing) products like NFS, NIS, AutoFS, CacheFS, etc. do use RPC but not DCE/RPC.

Is your system running DCE applications or participating in a DCE/DFS cell? If not then I wouldn't worry about this daemon. It gets started on all systems by default, including my 11.23 systems. It's startup behavior is controlled by the /etc/rc.config.d/Rpcd file, and specifically the START_RPCD variable.

I've personally never heard of anyone configuring DCE stuff with ServiceGuard, so unless you're doing this I don't think it would be harmful to ServiceGuard to stop it.

The safest way to stop the daemon would be to issue the command:

# /sbin/init.d/Rpcd stop

Regards,

Dave

[Fred Ruffet]

I did this a few time before and discovered that on my servers, RPC is invlolved in NFS and also in NIS. Note that NFS is used by CIFS and PFS.

When all dependent services are down, use
/sbin/init.d/Rpcd stop
/sbin/init.d/Rpcd start

Before starting the service cd to / to avoid the same problem to happen again.

More generally, to stop a service, always use the script in /sbin/init.d and never kill when it's possible.

Regards,

Fred

[Ralph Grothe]

Hello again,

when I said I'd kill the rpcd of course had I meant to bring it down in an ordered manner, like by an init script (if such exists), or slowly increasing force (SIGTERM => SIGKILL).

Funny thing is on this box there's no discrete init script pre-source file for rpcd in /etc/rc.config.d

# ll /etc/rc.config.d/[rR]*
/etc/rc.config.d/[rR]* not found
# grep -l -i rpc /etc/rc.config.d/*
/etc/rc.config.d/dfs
/etc/rc.config.d/ncs
/etc/rc.config.d/nfsconf

I mentioned already that this box isn't using any NFS services

# grep -E '^[^#]' /etc/rc.config.d/nfsconf
NFS_CLIENT=0
NFS_SERVER=0
NUM_NFSD=4
NUM_NFSIOD=4
PCNFS_SERVER=0
LOCKD_OPTIONS=""
STATD_OPTIONS=""
MOUNTD_OPTIONS=""
AUTOMOUNT=0
AUTO_MASTER="/etc/auto_master"
AUTO_OPTIONS="-f $AUTO_MASTER"


nor is dfs configured to be run (whatever that is, looks like it is in the vein of DCE)

# grep -E '^[^#]' /etc/rc.config.d/dfs

DCE_KRPC=0
DFS_CORE=0
DFS_CLIENT=0
DFS_SERVER=0
DFS_EPISODE=0
EPIINIT=0
DFSEXPORT=0
BOSSERVER=0
DFSBIND=0
FXD=0
MEMCACHE=0
DFSGWD=0
DISKCACHEFORDFS=0
MEMCACHE_BLOCKS_SIZE=


But here's some interesting service that seems to got started

# grep -E '^[^#]' /etc/rc.config.d/ncs
START_LLBD=1
START_GLBD=0

Let's look what it is.
Ah, there's a pretty good comment header in the config file.
Looks as if this is the screw which started the rpcd.

# sed -n 10,13p /etc/rc.config.d/ncs
# If START_LLBD is 1, the DCE RPC daemon (/opt/dce/sbin/rpcd) will
# be started on this host. Because the DCE rc file also examines this
# variable, its name should not be changed.


Since I've haven't heard of DCE (is it some sort of CORBA?) we probably won't need it.
Is this too audacious an assumption?

[Con O'Kelly]

Hi Ralph

DCE is a communications method used primarily by Openview products such as OVO(ITO) and I think Measureware (OVPA). It listens on port 135.

It is started by the /sbin/init.d/Rpcd script. There is also a config script /etc/rc.config.d/Rpcd.

The man page for rpcd gives more details.

Cheers
Con

[Ralph Grothe]

Hi Con,

as I wrote already, I don't have a separate init script for rpc (this is a 10.20 box)

# ll /sbin/init.d/[Rr]*
-r-xr-xr-x 1 bin bin 2184 Jun 10 1996 /sbin/init.d/rarpd
-r-xr-xr-x 1 bin bin 1522 Jun 7 1996 /sbin/init.d/rbootd
-r-xr-xr-x 1 bin bin 2354 Jun 10 1996 /sbin/init.d/rdpd
-r-xr-xr-x 1 bin bin 1422 Jun 7 1996 /sbin/init.d/rwhod


But the comments in the init script /sbin/init.d/ncs are clearly pointing out that the rpcd is started either as part of ncs as replacement for the local location broker (which had a separate daemon in earlier HP-UX releases), or as part of the dce init script.
Both scripts use the same shell variable START_LLBD to decide whether to run this service.
So by now I've found out what needs to be done.
My only left doubt is if it will do any harm to the ServiceGuard suite of daemons and monitors when I stop the rpcd aka llbd?

[Ralph Grothe]

Ok guys, I was daring enough, and ran
/sbin/init.d/dce stop
which finished the rpcd that held /opt busy.
Then unmounting and extendfs-ing was no problem.
However, now I struggle with the installation of the software that gave cause to this thread.
But that's another story...