Merge branch 'rnou/rootinstall_fixes' into 'main'

Rnou/rootinstall fixes

This MR solves some issues with the installation on root scenarios (plugin / systemd). This also solves a CPU hog (100%) for the scheduler.

See merge request !125

plugins/slurm/defaults.h.in

@@ -39,7 +39,7 @@
 #define SCORDCTL_TMPDIR_DEFAULT "/tmp"
 #define CARGO_PROG_DEFAULT     "@CARGO_PROGRAM@"
 #define CARGO_PROTO_DEFAULT    SCORD_PROTO_DEFAULT
-#define CARGO_PORT_DEFAULT     62000
+#define CARGO_PORT_DEFAULT     @CARGO_BIND_PORT@
 
 // clang-format on
 

plugins/slurm/scord_common.sh.in

@@ -183,4 +183,4 @@ export SCORD_QUERY_PROGRAM="@SCORD_QUERY_PROGRAM@"
 export SCORDCTL_PROGRAM="@SCORDCTL_PROGRAM@"
 export SCORDCTL_PROTO="@SCORD_TRANSPORT_PROTOCOL@"
 export SCORDCTL_PORT="@SCORD_CTL_BIND_PORT@"
-export CARGO_PORT="@CARGO_PORT@"
+export CARGO_PORT="@CARGO_BIND_PORT@"

plugins/slurm/scord_prolog.sh.in

@@ -143,7 +143,7 @@ fi
 # stored (note that $HOME is not set when this prolog script is being executed).
 # shellcheck disable=SC2016
 USER_HOME=$(run_as "$SLURM_JOB_USER" echo '$HOME')
-USER_CONFIG_DIRECTORY="${XDG_CONFIG_HOME:-$USER_HOME/.config}"
+USER_CONFIG_DIRECTORY="$USER_HOME/.config"
 CARGO_CONFIG_DIRECTORY="$USER_CONFIG_DIRECTORY/cargo"
 SYSTEMD_USER_DIRECTORY="$USER_CONFIG_DIRECTORY/systemd/user"
 
@@ -189,16 +189,16 @@ CARGO_HOSTS=$hostnames_csv
 CARGO_NUM_NODES=$CARGO_NUM_NODES
 CARGO_ADDRESS=$CARGO_MASTER_ADDRESS
 EOT
-
+CUID=$(id -u $SLURM_JOB_USER)
 chown "$SLURM_JOB_USER":"$SLURM_JOB_GROUP" "$CARGO_CONFIG_FILE"
 
-if ! run_as "$SLURM_JOB_USER" systemctl --user start "$CARGO_INSTANCE_NAME"; then
+if ! run_as "$SLURM_JOB_USER" DBUS_SESSION_BUS_ADDRESS=unix:path=/run/user/$CUID/bus systemctl --user start "$CARGO_INSTANCE_NAME"; then
   exit 1
 fi
 
 sleep 1s
 
-if ! run_as "$SLURM_JOB_USER" systemctl --user is-active --quiet "$CARGO_INSTANCE_NAME"; then
+if ! run_as "$SLURM_JOB_USER" DBUS_SESSION_BUS_ADDRESS=unix:path=/run/user/$CUID/bus systemctl --user is-active --quiet "$CARGO_INSTANCE_NAME"; then
   echo "Cargo data stager failed to start"
   exit 1
 fi

plugins/slurm/systemd/cargo@.service.in

@@ -3,7 +3,7 @@ Description=Cargo parallel data stager
 
 [Service]
 Type=simple
-EnvironmentFile=%S/cargo/%I.cfg
+EnvironmentFile=%h/.config/cargo/%I.cfg
 ExecStart=@CMAKE_INSTALL_FULL_DATADIR@/@PROJECT_NAME@/slurm/cargoctl start -s ${CARGO_ADDRESS} -H ${CARGO_HOSTS} -n ${CARGO_NUM_NODES}
 ExecStop=@CMAKE_INSTALL_FULL_DATADIR@/@PROJECT_NAME@/slurm/cargoctl stop -s ${CARGO_ADDRESS}
 Restart=no

src/scord/rpc_server.cpp

@@ -951,7 +951,8 @@ rpc_server::scheduler_update() {
     std::vector<std::pair<std::string, int>> return_set;
     const auto threshold = 0.1f;
     while(!m_shutting_down) {
-        thallium::thread::self().sleep(m_network_engine, 1000);
+        sleep(1);
+        //thallium::thread::self().sleep(m_network_engine, 500);
         m_transfer_manager.lock();
         const auto transfer = m_transfer_manager.transfer();
         std::vector<scord::transfer_id> v_ids;

src/scord/rpc_server.hpp

@@ -118,8 +118,8 @@ private:
     adhoc_storage_manager m_adhoc_manager;
     pfs_storage_manager m_pfs_manager;
     transfer_manager<cargo::transfer> m_transfer_manager;
-
-    // Dedicated execution stream for the MPI listener ULT
+    
+    // Dedicated execution stream for the Scheduler listener ULT
     thallium::managed<thallium::xstream> m_scheduler_ess;
     // ULT for the MPI listener
     thallium::managed<thallium::thread> m_scheduler_ult;