@cryptotaxi247 / netdata-1 / commits / 5e0212919

health: remove "families" from alarms config (#15086)

Ilya Mashchenko committed Jun 1, 2023 at 20:36 UTC 5e0212919cd60493bf2d3716739fdf08ad0830f9
15 files changed +7 -79
health/health.d/boinc.conf
-4
@@ -8,7 +8,6 @@
8 component: BOINC
9 os: *
10 hosts: *
11 - families: *
11 lookup: average -10m unaligned of comperror
12 units: tasks
13 every: 1m
@@ -26,7 +25,6 @@ component: BOINC
25 component: BOINC
26 os: *
27 hosts: *
29 - families: *
28 lookup: average -10m unaligned of upload_failed
29 units: tasks
30 every: 1m
@@ -44,7 +42,6 @@ component: BOINC
42 component: BOINC
43 os: *
44 hosts: *
47 - families: *
45 lookup: average -10m unaligned of total
46 units: tasks
47 every: 1m
@@ -62,7 +59,6 @@ component: BOINC
59 component: BOINC
60 os: *
61 hosts: *
65 - families: *
62 lookup: average -10m unaligned of active
63 calc: ($boinc_total_tasks >= 1) ? ($this) : (inf)
64 units: tasks
health/health.d/btrfs.conf
-9
@@ -6,7 +6,6 @@
6 component: File system
7 os: *
8 hosts: *
9 - families: *
9 calc: 100 - ($unallocated * 100 / ($unallocated + $data_used + $data_free + $meta_used + $meta_free + $sys_used + $sys_free))
10 units: %
11 every: 10s
@@ -23,7 +22,6 @@ component: File system
22 component: File system
23 os: *
24 hosts: *
26 - families: *
25 calc: $used * 100 / ($used + $free)
26 units: %
27 every: 10s
@@ -40,7 +38,6 @@ component: File system
38 component: File system
39 os: *
40 hosts: *
43 - families: *
41 calc: ($used + $reserved) * 100 / ($used + $free + $reserved)
42 units: %
43 every: 10s
@@ -57,7 +54,6 @@ component: File system
54 component: File system
55 os: *
56 hosts: *
60 - families: *
57 calc: $used * 100 / ($used + $free)
58 units: %
59 every: 10s
@@ -74,7 +70,6 @@ component: File system
70 component: File system
71 os: *
72 hosts: *
77 - families: *
73 units: errors
74 lookup: max -10m every 1m of read_errs
75 warn: $this > 0
@@ -89,7 +84,6 @@ component: File system
84 component: File system
85 os: *
86 hosts: *
92 - families: *
87 units: errors
88 lookup: max -10m every 1m of write_errs
89 warn: $this > 0
@@ -104,7 +98,6 @@ component: File system
98 component: File system
99 os: *
100 hosts: *
107 - families: *
101 units: errors
102 lookup: max -10m every 1m of flush_errs
103 warn: $this > 0
@@ -119,7 +112,6 @@ component: File system
112 component: File system
113 os: *
114 hosts: *
122 - families: *
115 units: errors
116 lookup: max -10m every 1m of corruption_errs
117 warn: $this > 0
@@ -134,7 +126,6 @@ component: File system
126 component: File system
127 os: *
128 hosts: *
137 - families: *
129 units: errors
130 lookup: max -10m every 1m of generation_errs
131 warn: $this > 0
health/health.d/disks.conf
+2 -8
@@ -16,7 +16,7 @@
16 component: Disk
17 os: linux freebsd
18 hosts: *
19 - families: !/dev !/dev/* !/run !/run/* *
19 +chart labels: mount_point=!/dev !/dev/* !/run !/run/* *
20 calc: $used * 100 / ($avail + $used)
21 units: %
22 every: 1m
@@ -33,7 +33,7 @@ component: Disk
33 component: Disk
34 os: linux freebsd
35 hosts: *
36 - families: !/dev !/dev/* !/run !/run/* *
36 +chart labels: mount_point=!/dev !/dev/* !/run !/run/* *
37 calc: $used * 100 / ($avail + $used)
38 units: %
39 every: 1m
@@ -59,7 +59,6 @@ component: Disk
59 # on: disk.space
60 # os: linux freebsd
61 # hosts: *
62 -# families: *
62 # lookup: min -10m at -50m unaligned of avail
63 # calc: ($this - $avail) / (($now - $after) / 3600)
64 # every: 1m
@@ -75,7 +74,6 @@ component: Disk
74 # on: disk.space
75 # os: linux freebsd
76 # hosts: *
78 -# families: *
77 # calc: ($disk_fill_rate > 0) ? ($avail / $disk_fill_rate) : (inf)
78 # units: hours
79 # every: 10s
@@ -101,7 +99,6 @@ component: Disk
99 # on: disk.inodes
100 # os: linux freebsd
101 # hosts: *
104 -# families: *
102 # lookup: min -10m at -50m unaligned of avail
103 # calc: ($this - $avail) / (($now - $after) / 3600)
104 # every: 1m
@@ -116,7 +113,6 @@ component: Disk
113 # on: disk.inodes
114 # os: linux freebsd
115 # hosts: *
119 -# families: *
116 # calc: ($disk_inode_rate > 0) ? ($avail / $disk_inode_rate) : (inf)
117 # units: hours
118 # every: 10s
@@ -141,7 +137,6 @@ component: Disk
137 component: Disk
138 os: linux freebsd
139 hosts: *
144 - families: *
140 lookup: average -10m unaligned
141 units: %
142 every: 1m
@@ -163,7 +158,6 @@ component: Disk
158 component: Disk
159 os: linux
160 hosts: *
166 - families: *
161 lookup: average -10m unaligned
162 units: ms
163 every: 1m
health/health.d/exporting.conf
-2
@@ -1,6 +1,5 @@
1
2 template: exporting_last_buffering
3 - families: *
3 on: exporting_data_size
4 class: Latency
5 type: Netdata
@@ -15,7 +14,6 @@ component: Exporting engine
14 to: dba
15
16 template: exporting_metrics_sent
18 - families: *
17 on: exporting_data_size
18 class: Workload
19 type: Netdata
health/health.d/httpcheck.conf
-5
@@ -1,7 +1,6 @@
1
2 # This is a fast-reacting no-notification alarm ideal for custom dashboards or badges
3 template: httpcheck_web_service_up
4 - families: *
4 on: httpcheck.status
5 class: Utilization
6 type: Web Server
@@ -14,7 +13,6 @@ component: HTTP endpoint
13 to: silent
14
15 template: httpcheck_web_service_bad_content
17 - families: *
16 on: httpcheck.status
17 class: Workload
18 type: Web Server
@@ -29,7 +27,6 @@ component: HTTP endpoint
27 to: webmaster
28
29 template: httpcheck_web_service_bad_status
32 - families: *
30 on: httpcheck.status
31 class: Workload
32 type: Web Server
@@ -44,7 +41,6 @@ component: HTTP endpoint
41 to: webmaster
42
43 template: httpcheck_web_service_timeouts
47 - families: *
44 on: httpcheck.status
45 class: Latency
46 type: Web Server
@@ -59,7 +55,6 @@ component: HTTP endpoint
55 to: webmaster
56
57 template: httpcheck_web_service_no_connection
62 - families: *
58 on: httpcheck.status
59 class: Errors
60 type: Other
health/health.d/ioping.conf
-1
@@ -1,5 +1,4 @@
1 template: ioping_disk_latency
2 - families: *
2 on: ioping.latency
3 class: Latency
4 type: System
health/health.d/mdstat.conf
+1 -1
@@ -29,7 +29,7 @@ component: RAID
29 class: Errors
30 type: System
31 component: RAID
32 - families: !*(raid1) !*(raid10) *
32 +chart labels: raid_level=!raid1 !raid10 *
33 units: unsynchronized blocks
34 calc: $count
35 every: 60s
health/health.d/net.conf
+4 -14
@@ -11,7 +11,6 @@
11 component: Network
12 os: *
13 hosts: *
14 - families: *
14 calc: ( $nic_speed_max > 0 ) ? ( $nic_speed_max) : ( nan )
15 units: Mbit
16 every: 10s
@@ -24,7 +23,6 @@ component: Network
23 component: Network
24 os: linux
25 hosts: *
27 - families: *
26 lookup: average -1m unaligned absolute of received
27 calc: ($interface_speed > 0) ? ($this * 100 / ($interface_speed)) : ( nan )
28 units: %
@@ -41,7 +39,6 @@ component: Network
39 component: Network
40 os: linux
41 hosts: *
44 - families: *
42 lookup: average -1m unaligned absolute of sent
43 calc: ($interface_speed > 0) ? ($this * 100 / ($interface_speed)) : ( nan )
44 units: %
@@ -68,7 +65,6 @@ component: Network
65 component: Network
66 os: linux
67 hosts: *
71 - families: *
68 lookup: sum -10m unaligned absolute of inbound
69 units: packets
70 every: 1m
@@ -81,7 +77,6 @@ component: Network
77 component: Network
78 os: linux
79 hosts: *
84 - families: *
80 lookup: sum -10m unaligned absolute of outbound
81 units: packets
82 every: 1m
@@ -94,7 +89,7 @@ component: Network
89 component: Network
90 os: linux
91 hosts: *
97 - families: !wl* *
92 +chart labels: device=!wl* *
93 lookup: sum -10m unaligned absolute of received
94 calc: (($inbound_packets_dropped != nan AND $this > 10000) ? ($inbound_packets_dropped * 100 / $this) : (0))
95 units: %
@@ -111,7 +106,7 @@ component: Network
106 component: Network
107 os: linux
108 hosts: *
114 - families: !wl* *
109 +chart labels: device=!wl* *
110 lookup: sum -10m unaligned absolute of sent
111 calc: (($outbound_packets_dropped != nan AND $this > 1000) ? ($outbound_packets_dropped * 100 / $this) : (0))
112 units: %
@@ -128,7 +123,7 @@ component: Network
123 component: Network
124 os: linux
125 hosts: *
131 - families: wl*
126 +chart labels: device=wl*
127 lookup: sum -10m unaligned absolute of received
128 calc: (($inbound_packets_dropped != nan AND $this > 10000) ? ($inbound_packets_dropped * 100 / $this) : (0))
129 units: %
@@ -145,7 +140,7 @@ component: Network
140 component: Network
141 os: linux
142 hosts: *
148 - families: wl*
143 +chart labels: device=wl*
144 lookup: sum -10m unaligned absolute of sent
145 calc: (($outbound_packets_dropped != nan AND $this > 1000) ? ($outbound_packets_dropped * 100 / $this) : (0))
146 units: %
@@ -165,7 +160,6 @@ component: Network
160 component: Network
161 os: freebsd
162 hosts: *
168 - families: *
163 lookup: sum -10m unaligned absolute of inbound
164 units: errors
165 every: 1m
@@ -181,7 +175,6 @@ component: Network
175 component: Network
176 os: freebsd
177 hosts: *
184 - families: *
178 lookup: sum -10m unaligned absolute of outbound
179 units: errors
180 every: 1m
@@ -205,7 +198,6 @@ component: Network
198 component: Network
199 os: linux
200 hosts: *
208 - families: *
201 lookup: sum -10m unaligned absolute
202 units: errors
203 every: 1m
@@ -230,7 +222,6 @@ component: Network
222 component: Network
223 os: linux freebsd
224 hosts: *
233 - families: *
225 lookup: average -1m unaligned of received
226 units: packets
227 every: 10s
@@ -243,7 +234,6 @@ component: Network
234 component: Network
235 os: linux freebsd
236 hosts: *
246 - families: *
237 lookup: average -10s unaligned of received
238 calc: $this * 100 / (($1m_received_packets_rate < 1000)?(1000):($1m_received_packets_rate))
239 every: 10s
health/health.d/nvme.conf
-1
@@ -1,7 +1,6 @@
1 # you can disable an alarm notification by setting the 'to' line to: silent
2
3 template: nvme_device_critical_warnings_state
4 - families: *
4 on: nvme.device_critical_warnings_state
5 class: Errors
6 type: System
health/health.d/ping.conf
-3
@@ -1,7 +1,6 @@
1 # you can disable an alarm notification by setting the 'to' line to: silent
2
3 template: ping_host_reachable
4 - families: *
4 on: ping.host_packet_loss
5 class: Errors
6 type: Other
@@ -16,7 +15,6 @@ component: Network
15 to: sysadmin
16
17 template: ping_packet_loss
19 - families: *
18 on: ping.host_packet_loss
19 class: Errors
20 type: Other
@@ -33,7 +31,6 @@ component: Network
31 to: sysadmin
32
33 template: ping_host_latency
36 - families: *
34 on: ping.host_rtt
35 class: Latency
36 type: Other
health/health.d/portcheck.conf
-3
@@ -1,7 +1,6 @@
1
2 # This is a fast-reacting no-notification alarm ideal for custom dashboards or badges
3 template: portcheck_service_reachable
4 - families: *
4 on: portcheck.status
5 class: Workload
6 type: Other
@@ -14,7 +13,6 @@ component: TCP endpoint
13 to: silent
14
15 template: portcheck_connection_timeouts
17 - families: *
16 on: portcheck.status
17 class: Errors
18 type: Other
@@ -29,7 +27,6 @@ component: TCP endpoint
27 to: sysadmin
28
29 template: portcheck_connection_fails
32 - families: *
30 on: portcheck.status
31 class: Errors
32 type: Other
health/health.d/redis.conf
-4
@@ -1,7 +1,6 @@
1 # you can disable an alarm notification by setting the 'to' line to: silent
2
3 template: redis_connections_rejected
4 - families: *
4 on: redis.connections
5 class: Errors
6 type: KV Storage
@@ -15,7 +14,6 @@ component: Redis
14 to: dba
15
16 template: redis_bgsave_broken
18 - families: *
17 on: redis.bgsave_health
18 class: Errors
19 type: KV Storage
@@ -28,7 +26,6 @@ component: Redis
26 to: dba
27
28 template: redis_bgsave_slow
31 - families: *
29 on: redis.bgsave_now
30 class: Latency
31 type: KV Storage
@@ -43,7 +40,6 @@ component: Redis
40 to: dba
41
42 template: redis_master_link_down
46 - families: *
43 on: redis.master_link_down_since_time
44 class: Errors
45 type: KV Storage
health/health.d/vsphere.conf
-8
@@ -43,7 +43,6 @@ component: Memory
43 type: Virtual Machine
44 component: Network
45 hosts: *
46 - families: *
46 lookup: sum -10m unaligned absolute match-names of rx
47 units: packets
48 every: 1m
@@ -55,7 +54,6 @@ component: Network
54 type: Virtual Machine
55 component: Network
56 hosts: *
58 - families: *
57 lookup: sum -10m unaligned absolute match-names of tx
58 units: packets
59 every: 1m
@@ -69,7 +67,6 @@ component: Network
67 type: Virtual Machine
68 component: Network
69 hosts: *
72 - families: *
70 lookup: sum -10m unaligned absolute match-names of rx
71 calc: (($vsphere_inbound_packets_errors != nan AND $this > 1000) ? ($vsphere_inbound_packets_errors * 100 / $this) : (0))
72 units: %
@@ -85,7 +82,6 @@ component: Network
82 type: Virtual Machine
83 component: Network
84 hosts: *
88 - families: *
85 lookup: sum -10m unaligned absolute match-names of tx
86 calc: (($vsphere_outbound_packets_errors != nan AND $this > 1000) ? ($vsphere_outbound_packets_errors * 100 / $this) : (0))
87 units: %
@@ -121,7 +117,6 @@ component: CPU
117 type: Virtual Machine
118 component: Network
119 hosts: *
124 - families: *
120 lookup: sum -10m unaligned absolute match-names of rx
121 units: packets
122 every: 1m
@@ -133,7 +128,6 @@ component: Network
128 type: Virtual Machine
129 component: Network
130 hosts: *
136 - families: *
131 lookup: sum -10m unaligned absolute match-names of tx
132 units: packets
133 every: 1m
@@ -147,7 +141,6 @@ component: Network
141 type: Virtual Machine
142 component: Network
143 hosts: *
150 - families: *
144 lookup: sum -10m unaligned absolute match-names of rx
145 calc: (($vsphere_inbound_packets_dropped != nan AND $this > 1000) ? ($vsphere_inbound_packets_dropped * 100 / $this) : (0))
146 units: %
@@ -163,7 +156,6 @@ component: Network
156 type: Virtual Machine
157 component: Network
158 hosts: *
166 - families: *
159 lookup: sum -10m unaligned absolute match-names of tx
160 calc: (($vsphere_outbound_packets_dropped != nan AND $this > 1000) ? ($vsphere_outbound_packets_dropped * 100 / $this) : (0))
161 units: %
health/health.d/web_log.conf
-12
@@ -13,7 +13,6 @@
13 class: Workload
14 type: Web Server
15 component: Web log
16 - families: *
16 lookup: sum -1m unaligned
17 calc: ($this == 0)?(1):($this)
18 units: requests
@@ -25,7 +24,6 @@ component: Web log
24 class: Errors
25 type: Web Server
26 component: Web log
28 - families: *
27 lookup: sum -1m unaligned of unmatched
28 calc: $this * 100 / $web_log_1m_total_requests
29 units: %
@@ -50,7 +48,6 @@ component: Web log
48 class: Workload
49 type: Web Server
50 component: Web log
53 - families: *
51 lookup: sum -1m unaligned
52 calc: ($this == 0)?(1):($this)
53 units: requests
@@ -62,7 +59,6 @@ component: Web log
59 class: Workload
60 type: Web Server
61 component: Web log
65 - families: *
62 lookup: sum -1m unaligned of success
63 calc: $this * 100 / $web_log_1m_requests
64 units: %
@@ -78,7 +74,6 @@ component: Web log
74 class: Workload
75 type: Web Server
76 component: Web log
81 - families: *
77 lookup: sum -1m unaligned of redirect
78 calc: $this * 100 / $web_log_1m_requests
79 units: %
@@ -93,7 +88,6 @@ component: Web log
88 class: Errors
89 type: Web Server
90 component: Web log
96 - families: *
91 lookup: sum -1m unaligned of bad
92 calc: $this * 100 / $web_log_1m_requests
93 units: %
@@ -108,7 +102,6 @@ component: Web log
102 class: Errors
103 type: Web Server
104 component: Web log
111 - families: *
105 lookup: sum -1m unaligned of error
106 calc: $this * 100 / $web_log_1m_requests
107 units: %
@@ -134,7 +127,6 @@ component: Web log
127 class: Latency
128 type: System
129 component: Web log
137 - families: *
130 lookup: average -10m unaligned of avg
131 units: ms
132 every: 30s
@@ -145,7 +137,6 @@ component: Web log
137 class: Latency
138 type: Web Server
139 component: Web log
148 - families: *
140 lookup: average -1m unaligned of avg
141 units: ms
142 every: 10s
@@ -174,7 +165,6 @@ component: Web log
165 class: Workload
166 type: Web Server
167 component: Web log
177 - families: *
168 lookup: average -5m at -5m unaligned of success
169 units: requests/s
170 every: 30s
@@ -185,7 +175,6 @@ component: Web log
175 class: Workload
176 type: Web Server
177 component: Web log
188 - families: *
178 lookup: average -5m unaligned of success
179 units: requests/s
180 every: 30s
@@ -196,7 +185,6 @@ component: Web log
185 class: Workload
186 type: Web Server
187 component: Web log
199 - families: *
188 calc: ($web_log_5m_successful_old > 0)?($web_log_5m_successful * 100 / $web_log_5m_successful_old):(100)
189 units: %
190 every: 30s
health/health.d/windows.conf
-4
@@ -62,7 +62,6 @@ component: Memory
62 component: Network
63 os: linux
64 hosts: *
65 - families: *
65 lookup: sum -10m unaligned absolute match-names of inbound
66 units: packets
67 every: 1m
@@ -78,7 +77,6 @@ component: Network
77 component: Network
78 os: linux
79 hosts: *
81 - families: *
80 lookup: sum -10m unaligned absolute match-names of outbound
81 units: packets
82 every: 1m
@@ -94,7 +92,6 @@ component: Network
92 component: Network
93 os: linux
94 hosts: *
97 - families: *
95 lookup: sum -10m unaligned absolute match-names of inbound
96 units: packets
97 every: 1m
@@ -110,7 +107,6 @@ component: Network
107 component: Network
108 os: linux
109 hosts: *
113 - families: *
110 lookup: sum -10m unaligned absolute match-names of outbound
111 units: packets
112 every: 1m