Skip to content

Commit

Permalink
updating panels to reflect metric name change
Browse files Browse the repository at this point in the history 
slurmjob_info -> rmsjob_info

Signed-off-by: Karl W Schulz <[email protected]>
  • Loading branch information
@koomie
koomie committed Jul 15, 2024
1 parent 1b7a12a commit c6585e2
Show file tree
Hide file tree
Showing 2 changed files with 81 additions and 81 deletions.
52 changes: 26 additions & 26 deletions grafana/json-models/slurm-global.json
View file
Original file line number Diff line number Diff line change
Expand Up @@ -96,7 +96,7 @@
},
"editorMode": "code",
"exemplar": false,
"expr": "count(slurmjob_info{jobid=~\"^\\\\d+\"})",
"expr": "count(rmsjob_info{jobid=~\"^\\\\d+\"})",
"instant": true,
"legendFormat": "__auto",
"range": false,
Expand Down Expand Up @@ -166,7 +166,7 @@
},
"editorMode": "code",
"exemplar": false,
"expr": "count(slurmjob_info{jobid=~\"^\\\\d+\", batchflag=\"1\"})",
"expr": "count(rmsjob_info{jobid=~\"^\\\\d+\", batchflag=\"1\"})",
"instant": true,
"legendFormat": "__auto",
"range": false,
Expand Down Expand Up @@ -235,7 +235,7 @@
"uid": "${source}"
},
"editorMode": "code",
"expr": "count(count by (jobid) (slurmjob_info{jobid=~\"^\\\\d.+\"}))",
"expr": "count(count by (jobid) (rmsjob_info{jobid=~\"^\\\\d.+\"}))",
"hide": false,
"instant": false,
"legendFormat": "__auto",
Expand Down Expand Up @@ -369,7 +369,7 @@
"uid": "${source}"
},
"editorMode": "code",
"expr": "count(rocm_utilization_percentage * on (instance) group_left() (slurmjob_info{jobid=~\"^\\\\d+\"}))",
"expr": "count(rocm_utilization_percentage * on (instance) group_left() (rmsjob_info{jobid=~\"^\\\\d+\"}))",
"instant": false,
"legendFormat": "__auto",
"range": true,
Expand Down Expand Up @@ -637,7 +637,7 @@
"uid": "${source}"
},
"editorMode": "code",
"expr": "avg by (jobid,user) (rocm_utilization_percentage * on (instance) group_left(jobid,user) (count by (instance,jobid,user) (slurmjob_info{jobid=~\"^\\\\d+\"})))",
"expr": "avg by (jobid,user) (rocm_utilization_percentage * on (instance) group_left(jobid,user) (count by (instance,jobid,user) (rmsjob_info{jobid=~\"^\\\\d+\"})))",
"hide": false,
"instant": false,
"legendFormat": "__auto",
Expand Down Expand Up @@ -807,7 +807,7 @@
"uid": "${source}"
},
"editorMode": "code",
"expr": "avg by (jobid,user) (rocm_utilization_percentage * on (instance) group_left(jobid,user) (count by (instance,jobid,user) (slurmjob_info{jobid=~\"^\\\\d+\"})))",
"expr": "avg by (jobid,user) (rocm_utilization_percentage * on (instance) group_left(jobid,user) (count by (instance,jobid,user) (rmsjob_info{jobid=~\"^\\\\d+\"})))",
"hide": false,
"instant": false,
"legendFormat": "__auto",
Expand Down Expand Up @@ -975,7 +975,7 @@
"uid": "${source}"
},
"editorMode": "code",
"expr": "avg by (jobid,user) (rocm_vram_used_percentage * on (instance) group_left(jobid,user) (count by (instance,jobid,user) (slurmjob_info{jobid=~\"^\\\\d+\"}))) / count by (jobid,user) (rocm_vram_used_percentage * on (instance) group_left(jobid,user) (count by (instance,jobid,user) (slurmjob_info{jobid=~\"^\\\\d+\"})))",
"expr": "avg by (jobid,user) (rocm_vram_used_percentage * on (instance) group_left(jobid,user) (count by (instance,jobid,user) (rmsjob_info{jobid=~\"^\\\\d+\"}))) / count by (jobid,user) (rocm_vram_used_percentage * on (instance) group_left(jobid,user) (count by (instance,jobid,user) (rmsjob_info{jobid=~\"^\\\\d+\"})))",
"instant": false,
"legendFormat": "__auto",
"range": true,
Expand All @@ -987,7 +987,7 @@
"uid": "${source}"
},
"editorMode": "code",
"expr": "max by (jobid,user) (rocm_vram_used_percentage * on (instance) group_left(jobid,user) (count by (instance,jobid,user) (slurmjob_info{jobid=~\"^\\\\d+\"}))) / count by (jobid,user) (rocm_vram_used_percentage * on (instance) group_left(jobid,user) (count by (instance,jobid,user) (slurmjob_info{jobid=~\"^\\\\d+\"})))",
"expr": "max by (jobid,user) (rocm_vram_used_percentage * on (instance) group_left(jobid,user) (count by (instance,jobid,user) (rmsjob_info{jobid=~\"^\\\\d+\"}))) / count by (jobid,user) (rocm_vram_used_percentage * on (instance) group_left(jobid,user) (count by (instance,jobid,user) (rmsjob_info{jobid=~\"^\\\\d+\"})))",
"hide": false,
"instant": false,
"legendFormat": "__auto",
Expand Down Expand Up @@ -1180,7 +1180,7 @@
"uid": "${source}"
},
"editorMode": "code",
"expr": "avg by (jobid,user) (rocm_vram_used_percentage * on (instance) group_left(jobid,user) (count by (instance,jobid,user) (slurmjob_info{jobid=~\"^\\\\d+\"}))) / count by (jobid,user) (rocm_vram_used_percentage * on (instance) group_left(jobid,user) (count by (instance,jobid,user) (slurmjob_info{jobid=~\"^\\\\d+\"})))",
"expr": "avg by (jobid,user) (rocm_vram_used_percentage * on (instance) group_left(jobid,user) (count by (instance,jobid,user) (rmsjob_info{jobid=~\"^\\\\d+\"}))) / count by (jobid,user) (rocm_vram_used_percentage * on (instance) group_left(jobid,user) (count by (instance,jobid,user) (rmsjob_info{jobid=~\"^\\\\d+\"})))",
"instant": false,
"legendFormat": "__auto",
"range": true,
Expand All @@ -1192,7 +1192,7 @@
"uid": "${source}"
},
"editorMode": "code",
"expr": "max by (jobid,user) (rocm_vram_used_percentage * on (instance) group_left(jobid,user) (count by (instance,jobid,user) (slurmjob_info{jobid=~\"^\\\\d+\"}))) / count by (jobid,user) (rocm_vram_used_percentage * on (instance) group_left(jobid,user) (count by (instance,jobid,user) (slurmjob_info{jobid=~\"^\\\\d+\"})))",
"expr": "max by (jobid,user) (rocm_vram_used_percentage * on (instance) group_left(jobid,user) (count by (instance,jobid,user) (rmsjob_info{jobid=~\"^\\\\d+\"}))) / count by (jobid,user) (rocm_vram_used_percentage * on (instance) group_left(jobid,user) (count by (instance,jobid,user) (rmsjob_info{jobid=~\"^\\\\d+\"})))",
"hide": false,
"instant": false,
"legendFormat": "__auto",
Expand Down Expand Up @@ -1366,7 +1366,7 @@
"uid": "${source}"
},
"editorMode": "code",
"expr": "avg by (user,jobid) (node_load1 * on (instance) group_left(jobid,user) (label_replace(slurmjob_info{jobid=~\"^\\\\d+\"}, \"instance\", \"$1:$node_exporter_port\", \"instance\", \"(.*):(.*)\")))",
"expr": "avg by (user,jobid) (node_load1 * on (instance) group_left(jobid,user) (label_replace(rmsjob_info{jobid=~\"^\\\\d+\"}, \"instance\", \"$1:$node_exporter_port\", \"instance\", \"(.*):(.*)\")))",
"hide": false,
"instant": false,
"legendFormat": "__auto",
Expand Down Expand Up @@ -1617,7 +1617,7 @@
"uid": "${source}"
},
"editorMode": "code",
"expr": "timestamp(group by (jobid,user,batchflag,partition,nodes) (slurmjob_info{jobid=~\"^\\\\d+\"}))",
"expr": "timestamp(group by (jobid,user,batchflag,partition,nodes) (rmsjob_info{jobid=~\"^\\\\d+\"}))",
"format": "table",
"hide": false,
"instant": false,
Expand Down Expand Up @@ -1940,7 +1940,7 @@
"uid": "${source}"
},
"editorMode": "code",
"expr": "rocm_utilization_percentage * on (instance) group_left(jobid,user) (slurmjob_info{jobid=~\"^\\\\d+\"})",
"expr": "rocm_utilization_percentage * on (instance) group_left(jobid,user) (rmsjob_info{jobid=~\"^\\\\d+\"})",
"format": "table",
"hide": false,
"instant": false,
Expand All @@ -1954,7 +1954,7 @@
"uid": "${source}"
},
"editorMode": "code",
"expr": "rocm_vram_used_percentage * on (instance) group_left(jobid,user) (slurmjob_info{jobid=~\"^\\\\d+\"})",
"expr": "rocm_vram_used_percentage * on (instance) group_left(jobid,user) (rmsjob_info{jobid=~\"^\\\\d+\"})",
"format": "table",
"hide": false,
"instant": false,
Expand All @@ -1968,7 +1968,7 @@
"uid": "${source}"
},
"editorMode": "code",
"expr": "timestamp(group by (jobid,user) (slurmjob_info{jobid=~\"^\\\\d+\", batchflag=\"1\"}))",
"expr": "timestamp(group by (jobid,user) (rmsjob_info{jobid=~\"^\\\\d+\", batchflag=\"1\"}))",
"format": "table",
"hide": false,
"instant": false,
Expand Down Expand Up @@ -2181,7 +2181,7 @@
},
"editorMode": "code",
"exemplar": false,
"expr": "timestamp(group by (jobid) (slurmjob_info{jobid=~\"^\\\\d+\", batchflag=\"1\"}))",
"expr": "timestamp(group by (jobid) (rmsjob_info{jobid=~\"^\\\\d+\", batchflag=\"1\"}))",
"format": "table",
"instant": false,
"legendFormat": "{{label_name}}",
Expand Down Expand Up @@ -2656,7 +2656,7 @@
"uid": "${source}"
},
"editorMode": "code",
"expr": "avg by (instance) (rocm_temperature_edge_celsius) * on (instance) (slurmjob_info{jobid=~\"^\\\\d+\"})",
"expr": "avg by (instance) (rocm_temperature_edge_celsius) * on (instance) (rmsjob_info{jobid=~\"^\\\\d+\"})",
"instant": false,
"legendFormat": "Node: {{instance}}",
"range": true,
Expand Down Expand Up @@ -2771,7 +2771,7 @@
"uid": "${source}"
},
"editorMode": "code",
"expr": "avg by (instance) (rocm_temperature_edge_celsius) * on (instance) (slurmjob_info{jobid=\"\"})",
"expr": "avg by (instance) (rocm_temperature_edge_celsius) * on (instance) (rmsjob_info{jobid=\"\"})",
"instant": false,
"legendFormat": "Node: {{instance}}",
"range": true,
Expand Down Expand Up @@ -3116,7 +3116,7 @@
"uid": "${source}"
},
"editorMode": "code",
"expr": "(node_load1 * on (instance) group_left(jobid,user) (label_replace(slurmjob_info{jobid=~\"^\\\\d+\"}, \"instance\", \"$1:$node_exporter_port\", \"instance\", \"(.*):(.*)\")))",
"expr": "(node_load1 * on (instance) group_left(jobid,user) (label_replace(rmsjob_info{jobid=~\"^\\\\d+\"}, \"instance\", \"$1:$node_exporter_port\", \"instance\", \"(.*):(.*)\")))",
"instant": false,
"legendFormat": "Node: {{instance}}",
"range": true,
Expand Down Expand Up @@ -3221,7 +3221,7 @@
"uid": "${source}"
},
"editorMode": "code",
"expr": "(node_load1 * on (instance) group_left(jobid,user) (label_replace(slurmjob_info{jobid=\"\"}, \"instance\", \"$1:$node_exporter_port\", \"instance\", \"(.*):(.*)\")))",
"expr": "(node_load1 * on (instance) group_left(jobid,user) (label_replace(rmsjob_info{jobid=\"\"}, \"instance\", \"$1:$node_exporter_port\", \"instance\", \"(.*):(.*)\")))",
"instant": false,
"legendFormat": "Node: {{instance}}",
"range": true,
Expand Down Expand Up @@ -3341,7 +3341,7 @@
"uid": "${source}"
},
"editorMode": "code",
"expr": "sum((sum by (instance) (rate(node_infiniband_port_data_received_bytes_total[$__rate_interval])) * on (instance) group_left(jobid,user) (label_replace(slurmjob_info{jobid=~\"^\\\\d+\"}, \"instance\", \"$1:$node_exporter_port\", \"instance\", \"(.*):(.*)\"))))",
"expr": "sum((sum by (instance) (rate(node_infiniband_port_data_received_bytes_total[$__rate_interval])) * on (instance) group_left(jobid,user) (label_replace(rmsjob_info{jobid=~\"^\\\\d+\"}, \"instance\", \"$1:$node_exporter_port\", \"instance\", \"(.*):(.*)\"))))",
"instant": false,
"legendFormat": "Received",
"range": true,
Expand All @@ -3353,7 +3353,7 @@
"uid": "${source}"
},
"editorMode": "code",
"expr": "sum((sum by (instance) (rate(node_infiniband_port_data_transmitted_bytes_total[$__rate_interval])) * on (instance) group_left(jobid,user) (label_replace(slurmjob_info{jobid=~\"^\\\\d+\"}, \"instance\", \"$1:$node_exporter_port\", \"instance\", \"(.*):(.*)\"))))",
"expr": "sum((sum by (instance) (rate(node_infiniband_port_data_transmitted_bytes_total[$__rate_interval])) * on (instance) group_left(jobid,user) (label_replace(rmsjob_info{jobid=~\"^\\\\d+\"}, \"instance\", \"$1:$node_exporter_port\", \"instance\", \"(.*):(.*)\"))))",
"hide": false,
"instant": false,
"legendFormat": "Transmitted",
Expand Down Expand Up @@ -3460,7 +3460,7 @@
"uid": "${source}"
},
"editorMode": "code",
"expr": "(sum by (instance) (rate(node_infiniband_port_data_received_bytes_total[$__rate_interval])) * on (instance) group_left(jobid,user) (label_replace(slurmjob_info{jobid=~\"^\\\\d+\"}, \"instance\", \"$1:$node_exporter_port\", \"instance\", \"(.*):(.*)\")))",
"expr": "(sum by (instance) (rate(node_infiniband_port_data_received_bytes_total[$__rate_interval])) * on (instance) group_left(jobid,user) (label_replace(rmsjob_info{jobid=~\"^\\\\d+\"}, \"instance\", \"$1:$node_exporter_port\", \"instance\", \"(.*):(.*)\")))",
"instant": false,
"legendFormat": "Node: {{instance}}",
"range": true,
Expand Down Expand Up @@ -3566,7 +3566,7 @@
"uid": "${source}"
},
"editorMode": "code",
"expr": "(sum by (instance) (rate(node_infiniband_port_data_received_bytes_total[$__rate_interval])) * on (instance) group_left(jobid,user) (label_replace(slurmjob_info{jobid=\"\"}, \"instance\", \"$1:$node_exporter_port\", \"instance\", \"(.*):(.*)\")))",
"expr": "(sum by (instance) (rate(node_infiniband_port_data_received_bytes_total[$__rate_interval])) * on (instance) group_left(jobid,user) (label_replace(rmsjob_info{jobid=\"\"}, \"instance\", \"$1:$node_exporter_port\", \"instance\", \"(.*):(.*)\")))",
"instant": false,
"legendFormat": "Node: {{instance}}",
"range": true,
Expand Down Expand Up @@ -3672,7 +3672,7 @@
"uid": "${source}"
},
"editorMode": "code",
"expr": "(sum by (instance) (rate(node_infiniband_port_data_transmitted_bytes_total[$__rate_interval])) * on (instance) group_left(jobid,user) (label_replace(slurmjob_info{jobid=~\"^\\\\d+\"}, \"instance\", \"$1:$node_exporter_port\", \"instance\", \"(.*):(.*)\")))",
"expr": "(sum by (instance) (rate(node_infiniband_port_data_transmitted_bytes_total[$__rate_interval])) * on (instance) group_left(jobid,user) (label_replace(rmsjob_info{jobid=~\"^\\\\d+\"}, \"instance\", \"$1:$node_exporter_port\", \"instance\", \"(.*):(.*)\")))",
"instant": false,
"legendFormat": "Node: {{instance}}",
"range": true,
Expand Down Expand Up @@ -3778,7 +3778,7 @@
"uid": "${source}"
},
"editorMode": "code",
"expr": "(sum by (instance) (rate(node_infiniband_port_data_transmitted_bytes_total[$__rate_interval])) * on (instance) group_left(jobid,user) (label_replace(slurmjob_info{jobid=\"\"}, \"instance\", \"$1:$node_exporter_port\", \"instance\", \"(.*):(.*)\")))",
"expr": "(sum by (instance) (rate(node_infiniband_port_data_transmitted_bytes_total[$__rate_interval])) * on (instance) group_left(jobid,user) (label_replace(rmsjob_info{jobid=\"\"}, \"instance\", \"$1:$node_exporter_port\", \"instance\", \"(.*):(.*)\")))",
"instant": false,
"legendFormat": "Node: {{instance}}",
"range": true,
Expand Down
Loading

0 comments on commit c6585e2

Please sign in to comment.