User Accounts¶
To add, update and delete user accounts on the Monolithe there are 3 dedicated
scripts presented in the following sections. These scripts are only available on
the frontend (monolithe.soc.lip6.fr) and located in the /nfs/sbin folder.
To execute them you need to be in the sudoers group (= having root privileges).
Description¶
adduser-nfs¶
To add a new user on the Monolithe you need to call the adduser-nfs script.
The latest will create an account on the frontend and on all the available
nodes. The users creation on the nodes is performed thanks to Slurm (based on
srun command).
For instance, if you want to add a new user named toto-nfs you can do:
The previous command will ask you to enter en password.
If you prefer to directly enter a password you can also call adduser-nfs like
this:
adduser-nfs will automatically create a home directory located here:
/nfs/users.
Note
By convention on the frontend, logins are postfixed by -nfs. This
clearly indicates that these accounts have a home directory on the NFS.
Again, by convention, the UIDs are set to a value greater than 1200.
It ensures that these UIDs are not already in use on the nodes.
deluser-nfs¶
To delete a user on the frontend and on the compute nodes, you can call the
deluser-nfs script:
This will erase the user from the system database but will keep the user home.
If you also need to delete the home directory you can do:
It will remove the user from the system database and it will delete the
/nfs/users/toto-nfs folder.
upuser-nfs¶
When adding new nodes to the cluster, the user accounts are not automatically created on the new machines. If you want to propagate an user account on a new node that has been setup on the Slurm cluster you can do:
This script also checks that the UIDs between the frontend and the compute nodes are the same. If not, the script will stop with a error telling you which UIDs differs from the master and on which Slurm partition.
Info
Most of the time you will want to add all the NFS users to the new node. For this, you can directly call the following command:
This will call the previousupuser-nfs (user without "s") command on all
the available logins automatically.
Tips
For each user, on each node, a space on the local drive is automatically
created here: /scratch/toto-nfs. It is useful when a large amount of files
is generated. In this particular case, the performance of the NFS is pretty
bad and it is advised to use the scratch space.
Source Codes¶
adduser-nfs¶
#!/bin/bash
#set -x
if [[ $UID != 0 ]]; then
echo "Please run this script with sudo:"
echo "sudo $0 $*"
exit 1
fi
if [[ $# == 0 ]]; then
echo "usage: $0 username [userpasswd]"
exit 1
fi
USERNAME=$1
REGEX=".*-nfs$"
USERNAME_CHECK=$(echo $USERNAME | grep "$REGEX")
if [ -z "$USERNAME_CHECK" ]
then
echo "The username should match the following regex \"$REGEX\""
exit 1
fi
PREVUSERUID=$(cat /etc/passwd | cut -d ':' -f 4 | grep "12[0-9][0-9]" | sort | tail -n 1)
let "USERUID=$PREVUSERUID+1"
if [ -z "$PREVUSERUID" ]
then
USERUID=1201
echo "This is the first NFS user and its id is: $USERUID"
else
echo "Using the following userid: $USERUID"
fi
# create the new account on the frontend
if [[ $# == 1 ]]; then
adduser $USERNAME --home /nfs/users/$USERNAME --shell /bin/bash --uid $USERUID
rc=$?; if [[ $rc != 0 ]]; then exit $rc; fi
else
USERPASSWD=$2
adduser $USERNAME --home /nfs/users/$USERNAME --shell /bin/bash --uid $USERUID --disabled-password --gecos ""
rc=$?; if [[ $rc != 0 ]]; then exit $rc; fi
echo $USERNAME:$USERPASSWD | chpasswd
fi
cd $(dirname $0)
# create the new account on the nodes by calling the update script
./upuser-nfs $USERNAME
deluser-nfs¶
#!/bin/bash
#set -x
if [[ $UID != 0 ]]; then
echo "Please run this script with sudo:"
echo "sudo $0 $*"
exit 1
fi
if [[ $# == 0 ]]; then
echo "usage: $0 username [--delete-home]"
exit 1
fi
if [[ $# -gt 1 ]]; then
if [ "$2" != "--delete-home" ]; then
echo "usage: $0 username [--delete-home]"
exit 1
fi
fi
USERNAME=$1
REGEX=".*-nfs$"
USERNAME_CHECK=$(echo $USERNAME | grep "$REGEX")
if [ -z "$USERNAME_CHECK" ]
then
echo "The username should match the following regex \"$REGEX\""
exit 1
fi
USERUID=$(cat /etc/passwd | grep "^$USERNAME:.*" | cut -d ':' -f 4)
echo "The username has be found, it has the following uid: $USERUID"
echo "Deleting the account on the master node..."
deluser $USERNAME
rc=$?; if [[ $rc != 0 ]]; then exit $rc; fi
if [[ $# -gt 1 ]]; then # --delete-home is set
echo "Removing \"/nfs/users/$USERNAME\" ..."
rm -rf /nfs/users/$USERNAME
rc=$?; if [[ $rc != 0 ]]; then exit $rc; fi
fi
# delete the account on the other nodes
PARTITIONS=$(sinfo | grep "idle" | cut -f 1 -d " ")
cd /nfs
echo "List of the Slurm partition:"
for PARTITION in $PARTITIONS; do
# remove the extra * character at the end of the default partition
PARTITIONCHECK=$(echo $PARTITION | grep ".*\*$")
if [ -z "$PARTITIONCHECK" ]
then
PARTITION=$PARTITION
else
PARTITION=$(echo ${PARTITION:0:-1})
fi
# check if the user exists on the node
USERNAMEEXIST=$(srun -p $PARTITION cat /etc/passwd | cut -d ':' -f 1 | grep "^$USERNAME$")
if [ "$USERNAME" = "$USERNAMEEXIST" ]; then
# check if the user id is the same on the master and on the partition
USERIDCHECK=$(srun -p $PARTITION cat /etc/passwd | grep "^$USERNAME:.*" | cut -d ':' -f 4)
rc=$?; if [[ $rc != 0 ]]; then exit $rc; fi
if [ "$USERUID" = "$USERIDCHECK" ]; then
echo "Deletion of the user on the following Slurm partition: $PARTITION ..."
srun -p $PARTITION deluser $USERNAME
rc=$?; if [[ $rc != 0 ]]; then exit $rc; fi
else
echo "(ERROR) User id mismatch on the master (uid = \"$USERUID\") and the partition \"$PARTITION\" (uid = \"$USERIDCHECK\")!"
echo "It should never happend! Exiting..."
exit 1
fi
else
echo "Skip the deletion of the user on the following Slurm partition: $PARTITION ..."
fi
done
upuser-nfs¶
#!/bin/bash
#set -x
if [[ $UID != 0 ]]; then
echo "Please run this script with sudo:"
echo "sudo $0 $*"
exit 1
fi
if [[ $# == 0 ]]; then
echo "usage: $0 username"
exit 1
fi
USERNAME=$1
REGEX=".*-nfs$"
USERNAME_CHECK=$(echo $USERNAME | grep "$REGEX")
if [ -z "$USERNAME_CHECK" ]
then
echo "The username should match the following regex \"$REGEX\""
exit 1
fi
USERNAMEEXIST=$(cat /etc/passwd | cut -d ':' -f 1 | grep "^$USERNAME$")
if [ -z "$USERNAMEEXIST" ]
then
echo "The username cannot be found"
exit 1
fi
USERUID=$(cat /etc/passwd | grep "^$USERNAME:.*" | cut -d ':' -f 4)
echo "The username has be found, it has the following uid: $USERUID"
# create the account on the nodes
PARTITIONS=$(sinfo | grep "idle" | cut -f 1 -d " ")
cd /nfs
echo "List of the Slurm partition:"
for PARTITION in $PARTITIONS; do
# remove the extra * character at the end of the default partition
PARTITIONCHECK=$(echo $PARTITION | grep ".*\*$")
if [ -z "$PARTITIONCHECK" ]
then
PARTITION=$PARTITION
else
PARTITION=$(echo ${PARTITION:0:-1})
fi
# check if the user exists on the current node
USERNAMEEXIST=$(srun -p $PARTITION cat /etc/passwd | cut -d ':' -f 1 | grep "^$USERNAME$")
if [ "$USERNAME" = "$USERNAMEEXIST" ]; then
# check if the user id is the same on the master and on the partition
USERIDCHECK=$(srun -p $PARTITION cat /etc/passwd | grep "^$USERNAME:.*" | cut -d ':' -f 4)
rc=$?; if [[ $rc != 0 ]]; then exit $rc; fi
if [ "$USERUID" = "$USERIDCHECK" ]; then
echo "Skip the creation of the user on the following Slurm partition: $PARTITION ..."
else
echo "ERROR: the user id mismatch on the master (\"$USERUID\") and the partition \"$PARTITION\" (\"$USERIDCHECK\")!"
echo "It should never happend! Exiting..."
exit 1
fi
else
echo "Creation of the user on the following Slurm partition: $PARTITION ..."
# adduser work differently on Fedora... It is a shame but what can we do?
if [ "$PARTITION" = "m1u" ]; then
srun -p $PARTITION adduser $USERNAME --home /nfs/users/$USERNAME --shell /bin/bash --uid $USERUID --no-create-home
rc=$?; if [[ $rc != 0 ]]; then exit $rc; fi
else
srun -p $PARTITION adduser $USERNAME --home /nfs/users/$USERNAME --shell /bin/bash --uid $USERUID --no-create-home --disabled-password --gecos ""
rc=$?; if [[ $rc != 0 ]]; then exit $rc; fi
fi
fi
# if the partition is "brub", add the user to the "sog" and "dialout" groups to be able to use the energy measurements
if [ "$PARTITION" = "brub" ]; then
srun -p $PARTITION usermod -a -G sog,dialout $USERNAME
rc=$?; if [[ $rc != 0 ]]; then exit $rc; fi
fi
# if the partition is "em780" or "x7ti", add the user to the "render" group to be able to use OpenCL
if [ "$PARTITION" = "em780" ] || [ "$PARTITION" = "x7ti" ]; then
srun -p $PARTITION usermod -a -G render $USERNAME
rc=$?; if [[ $rc != 0 ]]; then exit $rc; fi
fi
# add users to the `video` on Orin boards (to use PoCL)
if [ "$PARTITION" = "oagx" ] || [ "$PARTITION" = "onx" ] || [ "$PARTITION" = "onano" ]; then
srun -p $PARTITION usermod -a -G video $USERNAME
rc=$?; if [[ $rc != 0 ]]; then exit $rc; fi
fi
$(srun -p $PARTITION [ -d "/scratch" ])
rc=$?;
if [[ $rc == 1 ]]; then
echo "Creation of the scratch folder: $PARTITION ..."
$(srun -p $PARTITION mkdir /scratch)
rc=$?; if [[ $rc != 0 ]]; then exit $rc; fi
fi
$(srun -p $PARTITION [ -d "/scratch/$USERNAME" ])
rc=$?;
if [[ $rc == 1 ]]; then
echo "Creation of the scratch folder for '$USERNAME' user: $PARTITION ..."
$(srun -p $PARTITION mkdir /scratch/$USERNAME)
rc=$?; if [[ $rc != 0 ]]; then exit $rc; fi
$(srun -p $PARTITION chown -R $USERNAME:$USERNAME /scratch/$USERNAME)
rc=$?; if [[ $rc != 0 ]]; then exit $rc; fi
$(srun -p $PARTITION chmod -R 700 /scratch/$USERNAME)
fi
done
upusers-nfs¶
#!/bin/bash
#set -x
if [[ $UID != 0 ]]; then
echo "Please run this script with sudo:"
echo "sudo $0 $*"
exit 1
fi
USERNAMES=$(cat /etc/passwd | grep "\:12[0-9][0-9]\:" | cut -d ':' -f 1)
cd $(dirname $0)
for USERNAME in $USERNAMES; do
echo "Updating user: $USERNAME"
# create the current account on the nodes by calling the update script
./upuser-nfs $USERNAME
done