Removed unavailable alerts; upped cpu rate to 90

monitor/aws/prometheus/alert.rules.yml

@@ -30,14 +30,6 @@ groups:
       summary: "The frequent crawls are not running as fast as expected!"
       description: "The frequent crawls do not appear to be running as fast as it should be."
 
-  - alert: tidy-logs_no_new_crawl_logs
-    expr: delta(ukwa_crawler_log_size_bytes{log='crawl.log'}[1h]) == 0 or absent(ukwa_crawler_log_size_bytes{log='crawl.log'})
-    for: 1h
-    labels:
-      severity: severe
-    annotations:
-      summary: "No new crawl logs from tidy-logs"
-      description: "{{ $labels.instance }} of job {{ $labels.job }} failed to run."
 
 - name: Generic metrics
   rules:
@@ -53,13 +45,13 @@ groups:
       description: "{{ $labels.instance }} of job {{ $labels.job }} has been down for more than 5 minutes."
 
   - alert: host_cpu_rate
-    expr: 100 - 100*avg(irate(node_cpu_seconds_total{mode='idle'}[10m])) > 80
-    for: 10m
+    expr: 100 - 100*avg(irate(node_cpu_seconds_total{mode='idle'}[10m])) > 90
+    for: 15m
     labels:
       severity: severe
     annotations:
-      summary: "CPU rate on {{ $labels.instance }} > 80%"
-      description: "CPU rate on {{ $labels.instance }} > 80% for more than 10 minutes"
+      summary: "CPU rate on {{ $labels.instance }} > 90%"
+      description: "CPU rate on {{ $labels.instance }} > 90% for more than 15 minutes"
 
   - alert: host_free_disk_space
     expr: (node_filesystem_avail_bytes{fstype=~"ext.|xfs"}/node_filesystem_size_bytes{fstype=~"ext.|xfs"}) * 100 < 15
@@ -69,12 +61,3 @@ groups:
     annotations:
       summary: "Disk space free < 15% on {{ $labels.instance }}"
       description: "Disk space free on {{ $labels.instance }} has been less that 15% for more than 5 minutes"
-
-  - alert: cpu_running_too_hot
-    expr: max(node_hwmon_temp_celsius) by (instance) > 70
-    for: 30m
-    labels:
-      severity: severe
-    annotations:
-      summary: "CPU running too hot?"
-      description: "The CPU on {{ $labels.instance }} is running hot (>70C for 30mins)."