octopus: update slurm services

2 files changed, 112 insertions, 125 deletions

diff --git a/gn/deploy/octopus.scm b/gn/deploy/octopus.scm
index 9d1e7d3..ce480e2 100644
--- a/gn/deploy/octopus.scm
+++ b/gn/deploy/octopus.scm
@@ -2,42 +2,53 @@
 
 (use-modules (gnu)
              (gn services science)
-             (gn packages parallel))
+             (gn packages parallel)     ; for slurm-18.08
+             (srfi srfi-26))
 (use-service-modules networking ssh web)
-(use-package-modules certs tmux screen vim)
+(use-package-modules shells)
+
+(define %efraimf-ssh-pubkey
+  (plain-file "id_rsa.pub"
+              "ssh-rsa AAAAB3NzaC1yc2EAAAADAQABAAABAQDUCDY8ZKFF/ln0yzDt3CNmKz3cT4wzNv9bzCKvOBXcL0O7JtPWwqgLlZgmMHfzhzgReAkHcrt+Gdsyduzm/s9Y8c6QpyfaH6uoDwjfoOs6GrAjZaOXmAdncf+9HZEAy/IrygQ1YFRu6BvYogsdhhtN+O6IXBuvQQDRzldHs53Y53DK06Nrs19vAPwELXcDxcx1FvO+/L9nT8RHkI1Z0ucgTS+F/BWXl8+mh89r4j+4IRpZXOuCD0DrW5rgEE1EygF2dVdWZQESi23gU5Mt6vnmysXzwixB7j6I+xTih8LH4pz7hewEx6754e/cs9Gm7ZtfXKfXUt6+GtsBSBF3ULKl efraimf@octopus01"))
 
 (define %slurm.conf
   (plain-file "slurm.conf"
               (string-append
-"ClusterName=linux
-ControlMachine=octopus
-
-SlurmUser=slurm
-#SlurmdUser=root
-SlurmctldPort=6817
-SlurmdPort=6818
-AuthType=auth/munge
-StateSaveLocation=/var/spool/slurmd/ctld
-SlurmdSpoolDir=/var/spool/slurmd
-SwitchType=switch/none
-MpiDefault=none
-SlurmctldPidFile=/var/run/slurm/slurmctld.pid
-SlurmdPidFile=/var/run/slurm/slurmd.pid
-ProctrackType=proctrack/pgid
-ReturnToService=1
-
+"# Defaults are commented out, otherwise noted at the end of the line
+# Values are from example in the man page or from Debian
+ClusterName=linux       # no default, suggests lowercase
+#ControlMachine=octopus  # defunct, use SlurmctldHost
+SlurmctldHost=octopus   # no default, falls back to next SlurmctldHost in list
+
+SlurmUser=slurm         # default root, not recommended
+#SlurmctldPort=6817
+#SlurmdPort=6818
+#AuthType=auth/munge
+StateSaveLocation=/var/spool/slurmd/ctld    # default /var/spool
+#SlurmdSpoolDir=/var/spool/slurmd
+#SwitchType=switch/none
+#MpiDefault=none
+#SlurmctldPidFile=/var/run/slurmctld.pid
+#SlurmdPidFile=/var/run/slurmd.pid
+ProctrackType=proctrack/pgid    # default proctrack/cgroup
+ReturnToService=1       # default 0
+
+DebugFlags=NO_CONF_HASH # default empty
 # TIMERS
-InactiveLimit=0
-MinJobAge=300
-KillWait=30
+SlurmctldTimeout=300    # default 120
+#SlurmdTimeout=300
+#InactiveLimit=0
+#MinJobAge=300
+#KillWait=30
+#WaitTime=0
 #
 
 # LOGGING
-SlurmctldDebug=3
-SlurmctldLogFile=/var/log/slurm/slurmctld.log
-SlurmdDebug=3
-SlurmdLogFile=/var/log/slurm/slurmd.log
-JobCompType=jobcomp/none
+#SlurmctldDebug=3
+SlurmctldLogFile=/var/log/slurmctld.log     # default none, syslog
+#SlurmdDebug=3
+SlurmdLogFile=/var/log/slurmd.log           # default none, syslog
+#JobCompType=jobcomp/none
 
 # COMPUTE NODES
 NodeName=octopus CPUs=1 Boards=1 SocketsPerBoard=1 CoresPerSocket=1 ThreadsPerCore=1 RealMemory=1024
@@ -49,22 +60,22 @@ PartitionName=debug Nodes=ALL Default=YES MaxTime=INFINITE State=UP")))
 "###
 # Slurm cgroup support configuration file
 ###
-CgroupAutomount=yes
-ConstrainCores=yes
+CgroupAutomount=yes     # default no
+ConstrainCores=yes      # default no
 #")))
 
 (define %slurmdbd.conf
   (plain-file "slurmdbd.conf"
               (string-append
-"AuthType=auth/munge
-AuthInfo=/var/run/munge/munge.socket.2
-DbdHost=localhost
-StorageHost=localhost
-StorageType=accounting_storage/none
-StorageUser=slurm
-PidFile=/var/run/slurm/slurmdbd.pid
-LogFile=/var/log/slurm/slurmdbd.log
-SlurmUser=slurm")))
+"#AuthType=auth/munge
+#AuthInfo=/var/run/munge/munge.socket.2
+DbdHost=localhost       # must be specified
+StorageHost=localhost   # unclear, must be specified?
+StorageType=accounting_storage/none     # must be specified
+StorageUser=slurm       # unclear
+#PidFile=/var/run/slurmdbd.pid
+LogFile=/var/log/slurmdbd.log       # default none, syslog
+SlurmUser=slurm         # default root, not recommended")))
 
 (operating-system
   (host-name "octopus")
@@ -106,25 +117,37 @@ SlurmUser=slurm")))
 
   (users (cons*
            (user-account
+             (name "wrk")
+             (comment "Pjotr Prins")
+             (uid 502)
+             (group "users")
+             (supplementary-groups '("wheel" "kvm")))
+           (user-account
              (name "efraimf")
              (comment "Efraim Flashner")
              (uid 1000)
              (group "users")
              (supplementary-groups '("wheel" "kvm")))
            (user-account
-             (name "wrk")
-             (comment "Pjotr Prins")
-             (uid 502)
+             (name "erikg")
+             (comment "Erik Garrison")
+             (uid 1001)
              (group "users")
-             (supplementary-groups '("wheel" "kvm")))
+             (shell (file-append zsh "/bin/zsh")))
+           (user-account
+             (name "hchen")
+             (comment "Hao Chen")
+             (uid 1002)
+             (group "users"))
            %base-user-accounts))
 
 
-  (packages (cons*
-              nss-certs
-              screen
-              tmux
-              vim
+  (packages (append
+              (map (cut specification->package <>)
+                     '("nss-certs"
+                       "screen" "tmux"
+                       "vim"
+                       "htop"))
               %base-packages))
 
   (services
@@ -140,19 +163,15 @@ SlurmUser=slurm")))
               (service openssh-service-type
                        (openssh-configuration
                          (authorized-keys
-                           `(("efraimf" ,(local-file "/home/efraimf/.ssh/id_rsa.pub"))))))
+                           `(("efraimf" ,%efraimf-ssh-pubkey)))))
 
               (service munge-service-type)
-              (service slurmd-service-type
-                       (slurm-configuration
-                         (package slurm-18.08)))
-              (service slurmdbd-service-type
-                       (slurm-configuration
-                         (package slurm-18.08)
-                         (run-slurmdbd? #t)))
-              (service slurmctld-service-type
+              (service slurm-service-type
                        (slurm-configuration
                          (package slurm-18.08)
+                         (slurmd-log-file "/var/log/slurmd.log")
+                         (slurmctld-log-file "/var/log/slurmctld.log")
+                         (run-slurmdbd? #t)
                          (run-slurmctld? #t)))
 
               ;; Some slurm configuration files
diff --git a/gn/services/science.scm b/gn/services/science.scm
index d1f3190..f0f43d0 100644
--- a/gn/services/science.scm
+++ b/gn/services/science.scm
@@ -5,9 +5,7 @@
 
             slurm-configuration
             slurm-configuration?
-            slurmd-service-type
-            slurmdbd-service-type
-            slurmctld-service-type))
+            slurm-service-type))
 
 (use-modules (gnu)
              (guix records)
@@ -177,9 +175,9 @@
   (slurm-conf-file      slurm-configuration-slurm-conf-file
                         (default "/etc/slurm/slurm.conf"))
   (slurmd-log-file      slurm-configuration-slurmd-log-file
-                        (default "/var/log/slurm/slurmd.log"))
+                        (default #f))               ; #f for syslog
   (slurmd-pid-file      slurm-configuration-slurmd-pid-file
-                        (default "/var/run/slurm/slurmd.pid"))
+                        (default "/var/run/slurmd.pid"))
 
   (slurmd-spooldir      slurm-configuration-slurmd-spooldir
                         (default "/var/spool/slurmd"))
@@ -187,16 +185,16 @@
   (run-slurmctld?       slurm-configuration-run-slurmctld
                         (default #f))
   (slurmctld-log-file   slurm-configuration-slurmctld-log-file
-                        (default "/var/log/slurm/slurmctld.log"))
+                        (default #f))               ; #f for syslog
   (slurmctld-pid-file   slurm-configuration-slurmctld-pid-file
-                        (default "/var/run/slurm/slurmctld.pid"))
+                        (default "/var/run/slurmctld.pid"))
 
   (run-slurmdbd?        slurm-configuration-run-slurmdbd
                         (default #f))
   (slurmdbd-conf-file   slurm-configuration-slurmdbd-conf-file
                         (default "/etc/slurm/slurmdbd.conf"))
   (slurmdbd-pid-file    slurm-configuration-slurmdbd-pid-file
-                        (default "/var/run/slurm/slurmdbd.pid")))
+                        (default "/var/run/slurmdbd.pid")))
 
 
 (define (slurm-activation config)
@@ -204,16 +202,22 @@
   (with-imported-modules '((guix build utils))
     #~(begin
         (use-modules (guix build utils))
-        (define %user (getpw "slurm"))
-        (let ((homedir     (passwd:dir %user))
-              (spooldir    #$(slurm-configuration-slurmd-spooldir config))
-              (logdir      (dirname #$(slurm-configuration-slurmd-log-file config)))
-              (piddir      (dirname #$(slurm-configuration-slurmd-pid-file config))))
-          (for-each (lambda (dir)
-                      (unless (file-exists? dir)
-                        (mkdir-p dir))
-                      (chown dir (passwd:uid %user) (passwd:gid %user)))
-                    (list homedir spooldir piddir logdir)))
+        (let* ((%user       (getpw "slurm"))
+               (spooldir    #$(slurm-configuration-slurmd-spooldir config))
+               (logdir      (dirname (or #$(slurm-configuration-slurmd-log-file config)
+                                         #$(slurm-configuration-slurmctld-log-file config)
+                                         "/var/log/slurmd.log")))
+               (piddir      (dirname #$(slurm-configuration-slurmd-pid-file config))))
+          (unless (file-exists? spooldir)
+            (mkdir-p spooldir))
+          (chown spooldir (passwd:uid %user) (passwd:gid %user))
+          (when logdir
+            (unless (file-exists? logdir)
+              (mkdir-p logdir))
+            (when (> (string-length logdir) (string-length "/var/log"))
+              (chown logdir (passwd:uid %user) (passwd:gid %user))))
+          (unless (file-exists? piddir)
+            (mkdir-p piddir)))
         ;; /etc/slurm/slurm.conf needs to exist.
         (file-exists? #$(slurm-configuration-slurm-conf-file config)))))
 
@@ -227,7 +231,7 @@
          (requirement '(loopback munge))
          (start #~(make-forkexec-constructor
                     (list #$(file-append package "/sbin/slurmd")
-                          "-L" #$slurmd-log-file
+                          ;"-L" #$slurmd-log-file
                           "-f" #$slurm-conf-file)
                     #:pid-file #$slurmd-pid-file))
          (stop #~(make-kill-destructor)))))))
@@ -242,7 +246,7 @@
          (requirement '(loopback munge))
          (start #~(make-forkexec-constructor
                     (list #$(file-append package "/sbin/slurmctld")
-                          "-L" #$slurmctld-log-file
+                          ;"-L" #$slurmctld-log-file
                           "-f" #$slurm-conf-file)
                     #:pid-file #$slurmctld-pid-file))
          (stop #~(make-kill-destructor))
@@ -250,8 +254,10 @@
 
 (define (slurmdbd-activation config)
   "Test the Slurmdbd configration exists."
-  (file-exists?
-    (slurm-configuration-slurmdbd-conf-file config)))
+  (when (slurm-configuration-run-slurmdbd config)
+    (file-exists?
+      (slurm-configuration-slurmdbd-conf-file config)))
+  #t)
 
 (define slurmdbd-shepherd-service
   (match-lambda
@@ -269,67 +275,29 @@
 
 (define (slurm-services-to-run config)
   (append (slurmd-shepherd-service config)
-          (if (slurm-configuration-run-slurmctld? config)
+          (if (slurm-configuration-run-slurmctld config)
             (slurmctld-shepherd-service config)
             '())
-          (if (slurm-configuration-run-slurmdbd? config)
+          (if (slurm-configuration-run-slurmdbd config)
             (slurmdbd-shepherd-service config)
             '())))
 
-(define (slurm-activations-to-run config)
-  (append (slurm-activation config)
-          (if (slurm-configuration-run-slurmctld? config)
-            (slurmctld-activation config)
-            '())
-          (if (slurm-configuration-run-slurmdbd? config)
-            (slurmdbd-activation config)
-            '())))
-
-(define slurmd-service-type
+(define slurm-service-type
   (service-type
-    (name 'slurmd)
+    (name 'slurm)
     (extensions
       (list
         (service-extension shepherd-root-service-type
-                           ;(cons slurmd-shepherd-service
-                           ;      slurmdbd-shepherd-service))
-                           slurmd-shepherd-service)
-                           ;slurm-services-to-run)
+                           slurm-services-to-run)
         (service-extension activation-service-type
-                           ;(append slurm-activation
-                           ;      slurmdbd-activation))
                            slurm-activation)
-                           ;slurm-activations-to-run)
+        (service-extension activation-service-type
+                           slurmdbd-activation)
         (service-extension account-service-type
                            (const %slurm-accounts))
         (service-extension profile-service-type
                            (compose list slurm-configuration-package))))
     (default-value (slurm-configuration))
     (description
-     "Run @url{https://slurm.schedmd.com/slurm.html,Slurm}, a workflow manager service.")))
-
-(define slurmdbd-service-type
-  (service-type
-    (name 'slurmdbd)
-    (extensions
-      (list
-        (service-extension shepherd-root-service-type
-                           slurmdbd-shepherd-service)
-        (service-extension activation-service-type
-                           slurmdbd-activation)))
-    (default-value (slurm-configuration))
-    (description
-      ;; TODO: Fix for slurmdbd or integrate with slurm(d).
-     "Run @url{https://slurm.schedmd.com/slurm.html,Slurm}, a workflow manager service.")))
-
-(define slurmctld-service-type
-  (service-type
-    (name 'slurmctld)
-    (extensions
-      (list
-        (service-extension shepherd-root-service-type
-                           slurmctld-shepherd-service)))
-    (default-value (slurm-configuration))
-    (description
-      ;; TODO: Fix for slurmctld or integrate with slurm(d).
-     "Run @url{https://slurm.schedmd.com/slurm.html,Slurm}, a workflow manager service.")))
+     "Run @url{https://slurm.schedmd.com/slurm.html,Slurm}, a workflow manager
+service.  Optionally also run @code{slurmctld} and @code{slurmdbd}.")))