Remove --conf argument
This commit is contained in:
1 parent
33882d5455
commit
1024b3264b
5 files changed
+16
-16
No files matched your search
+2
-2
@@ -11,7 +11,7 @@ if [[ -e /sys/devices/system/edac ]]; then
|
|||||||
uncorrectable_errors=$(($uncorrectable_errors + $(cat $mc/ue_count)))
|
uncorrectable_errors=$(($uncorrectable_errors + $(cat $mc/ue_count)))
|
||||||
done
|
done
|
||||||
|
|
||||||
gmetric --name="ecc_correctable_errors" --title="ECC correctable errors" --value="$correctable_errors" --type="uint32" --group="memory" --conf=$conf
|
gmetric --name="ecc_correctable_errors" --title="ECC correctable errors" --value="$correctable_errors" --type="uint32" --group="memory"
|
||||||
gmetric --name="ecc_uncorrectable_errors" --title="ECC uncorrectable errors" --value="$uncorrectable_errors" --type="uint32" --group="memory" --conf=$conf
|
gmetric --name="ecc_uncorrectable_errors" --title="ECC uncorrectable errors" --value="$uncorrectable_errors" --type="uint32" --group="memory"
|
||||||
|
|
||||||
fi
|
fi
|
||||||
+4
-4
@@ -13,9 +13,9 @@ if [[ ( -n "`which ibstat`" ) && ( -n "`ibstat`" ) ]]; then
|
|||||||
|
|
||||||
# use counter * 4 / 60 because running once a minute, and perfquery returns 'octets divided by 4' for some reason
|
# use counter * 4 / 60 because running once a minute, and perfquery returns 'octets divided by 4' for some reason
|
||||||
# we also use a float because gmetric only supports up to uint32, and IB can do more than 4GB/s = 16 Gbit
|
# we also use a float because gmetric only supports up to uint32, and IB can do more than 4GB/s = 16 Gbit
|
||||||
gmetric --name="ib_bytes_out" --title="InfiniBand Bytes Sent" --value="$(($ib_bytes_out * 4 / 60))" --units="bytes/sec" --type="float" --group="network" --conf=$conf
|
gmetric --name="ib_bytes_out" --title="InfiniBand Bytes Sent" --value="$(($ib_bytes_out * 4 / 60))" --units="bytes/sec" --type="float" --group="network"
|
||||||
gmetric --name="ib_bytes_in" --title="InfiniBand Bytes Received" --value="$(($ib_bytes_in * 4 / 60))" --units="bytes/sec" --type="float" --group="network" --conf=$conf
|
gmetric --name="ib_bytes_in" --title="InfiniBand Bytes Received" --value="$(($ib_bytes_in * 4 / 60))" --units="bytes/sec" --type="float" --group="network"
|
||||||
gmetric --name="ib_pkts_out" --title="InfiniBand Packets Sent" --value="$(($ib_pkts_out / 60))" --units="packets/sec" --type="float" --group="network" --conf=$conf
|
gmetric --name="ib_pkts_out" --title="InfiniBand Packets Sent" --value="$(($ib_pkts_out / 60))" --units="packets/sec" --type="float" --group="network"
|
||||||
gmetric --name="ib_pkts_in" --title="InfiniBand Packets Received" --value="$(($ib_pkts_out / 60))" --units="packets/sec" --type="float" --group="network" --conf=$conf
|
gmetric --name="ib_pkts_in" --title="InfiniBand Packets Received" --value="$(($ib_pkts_out / 60))" --units="packets/sec" --type="float" --group="network"
|
||||||
fi
|
fi
|
||||||
fi
|
fi
|
||||||
+2
-2
@@ -43,9 +43,9 @@ while read line; do
|
|||||||
reading_event=${reading_event//$'\n'/}
|
reading_event=${reading_event//$'\n'/}
|
||||||
reading_event=${reading_event//"' '"/, }
|
reading_event=${reading_event//"' '"/, }
|
||||||
reading_event=${reading_event//"'"/}
|
reading_event=${reading_event//"'"/}
|
||||||
gmetric --name="$reading_name" --value="$(xml_escape "$reading_event")" --type=string --group=physical --conf=$conf
|
gmetric --name="$reading_name" --value="$(xml_escape "$reading_event")" --type=string --group=physical
|
||||||
else
|
else
|
||||||
gmetric --name="$reading_name" --value="$reading_value" --units=$reading_units --type=float --group=physical --conf=$conf
|
gmetric --name="$reading_name" --value="$reading_value" --units=$reading_units --type=float --group=physical
|
||||||
fi
|
fi
|
||||||
|
|
||||||
# echo "$reading_name $reading_value $reading_units"
|
# echo "$reading_name $reading_value $reading_units"
|
||||||
|
|||||||
+7
-7
@@ -18,13 +18,13 @@ if [ $(which sinfo) ]; then
|
|||||||
# Slurm node state
|
# Slurm node state
|
||||||
slurm_node_state=$(sinfo --nodes=$hn --Node --noheader --format='%t' | head -n1)
|
slurm_node_state=$(sinfo --nodes=$hn --Node --noheader --format='%t' | head -n1)
|
||||||
if [ "$slurm_node_state" ]; then
|
if [ "$slurm_node_state" ]; then
|
||||||
gmetric --name="slurm_node_state" --title="Slurm node state" --value="$(xml_escape "$slurm_node_state")" --type="string" --group="slurm" --conf=$conf
|
gmetric --name="slurm_node_state" --title="Slurm node state" --value="$(xml_escape "$slurm_node_state")" --type="string" --group="slurm"
|
||||||
fi
|
fi
|
||||||
|
|
||||||
# Slurm reason for being down
|
# Slurm reason for being down
|
||||||
slurm_down_reason=$(sinfo --nodes=$hn --Node --noheader --format='%E' | head -n1)
|
slurm_down_reason=$(sinfo --nodes=$hn --Node --noheader --format='%E' | head -n1)
|
||||||
if [ "$slurm_down_reason" ]; then
|
if [ "$slurm_down_reason" ]; then
|
||||||
gmetric --name="slurm_down_reason" --title="Slurm node down reason" --value="$(xml_escape "$slurm_down_reason")" --type="string" --group="slurm" --conf=$conf
|
gmetric --name="slurm_down_reason" --title="Slurm node down reason" --value="$(xml_escape "$slurm_down_reason")" --type="string" --group="slurm"
|
||||||
fi
|
fi
|
||||||
|
|
||||||
# Slurm partition
|
# Slurm partition
|
||||||
@@ -33,7 +33,7 @@ if [ $(which sinfo) ]; then
|
|||||||
done <<< $(sinfo --nodes=$hn --Node --noheader --format='%R')
|
done <<< $(sinfo --nodes=$hn --Node --noheader --format='%R')
|
||||||
slurm_partition=${slurm_partition%,}
|
slurm_partition=${slurm_partition%,}
|
||||||
if [ $slurm_partition ]; then
|
if [ $slurm_partition ]; then
|
||||||
gmetric --name="slurm_partition" --title="Slurm partition" --value="$(xml_escape "$slurm_partition")" --type="string" --group="slurm" --conf=$conf
|
gmetric --name="slurm_partition" --title="Slurm partition" --value="$(xml_escape "$slurm_partition")" --type="string" --group="slurm"
|
||||||
fi
|
fi
|
||||||
|
|
||||||
# Slurm jobs on this node
|
# Slurm jobs on this node
|
||||||
@@ -51,10 +51,10 @@ if [ $(which sinfo) ]; then
|
|||||||
slurm_job_user=${slurm_job_user%,}
|
slurm_job_user=${slurm_job_user%,}
|
||||||
slurm_job_time_elapsed=${slurm_job_time_elapsed%,}
|
slurm_job_time_elapsed=${slurm_job_time_elapsed%,}
|
||||||
|
|
||||||
gmetric --name="slurm_job_id" --title="Slurm running job ID" --value="$(xml_escape "$slurm_job_id")" --type="string" --group="slurm" --conf=$conf
|
gmetric --name="slurm_job_id" --title="Slurm running job ID" --value="$(xml_escape "$slurm_job_id")" --type="string" --group="slurm"
|
||||||
gmetric --name="slurm_job_name" --title="Slurm running job name" --value="$(xml_escape "$slurm_job_name")" --type="string" --group="slurm" --conf=$conf
|
gmetric --name="slurm_job_name" --title="Slurm running job name" --value="$(xml_escape "$slurm_job_name")" --type="string" --group="slurm"
|
||||||
gmetric --name="slurm_job_user" --title="Slurm running job user" --value="$(xml_escape "$slurm_job_user")" --type="string" --group="slurm" --conf=$conf
|
gmetric --name="slurm_job_user" --title="Slurm running job user" --value="$(xml_escape "$slurm_job_user")" --type="string" --group="slurm"
|
||||||
gmetric --name="slurm_job_time_elapsed" --title="Slurm running job elapsed time" --value="$(xml_escape "$slurm_job_time_elapsed")" --type="string" --group="slurm" --conf=$conf
|
gmetric --name="slurm_job_time_elapsed" --title="Slurm running job elapsed time" --value="$(xml_escape "$slurm_job_time_elapsed")" --type="string" --group="slurm"
|
||||||
fi
|
fi
|
||||||
|
|
||||||
fi
|
fi
|
||||||
+1
-1
@@ -2,4 +2,4 @@
|
|||||||
|
|
||||||
# systemd statistics
|
# systemd statistics
|
||||||
systemd_units_failed=$(systemctl --state=failed|grep -Po '\d+(?= loaded units listed.)')
|
systemd_units_failed=$(systemctl --state=failed|grep -Po '\d+(?= loaded units listed.)')
|
||||||
gmetric --name="systemd_units_failed" --title="systemd failed units" --value="$systemd_units_failed" --type="uint16" --group="process" --conf=$conf
|
gmetric --name="systemd_units_failed" --title="systemd failed units" --value="$systemd_units_failed" --type="uint16" --group="process"
|
||||||
Reference in new issue
Block a user