6146ef2b4 · cryptotaxi247/netdata-1

@@ -1,218 +1,4 @@

1

2 -# -----------------------------------------------------------------------------

3 -# high level response code alarms

4 -

5 -# the following alarms trigger only when there are enough data.

6 -# we assume there are enough data when:

7 -#

8 -# $1m_requests > 120

9 -#

10 -# i.e. when there are at least 120 requests during the last minute

11 -

12 - template: 1m_requests

13 - on: web_log.response_statuses

14 - class: Workload

15 - type: Web Server

16 -component: Web log

17 - families: *

18 - lookup: sum -1m unaligned

19 - calc: ($this == 0)?(1):($this)

20 - units: requests

21 - every: 10s

22 - info: number of HTTP requests in the last minute

23 -

24 - template: 1m_successful

25 - on: web_log.response_statuses

26 - class: Workload

27 - type: Web Server

28 -component: Web log

29 - families: *

30 - lookup: sum -1m unaligned of successful_requests

31 - calc: $this * 100 / $1m_requests

32 - units: %

33 - every: 10s

34 - warn: ($1m_requests > 120) ? ($this < (($status >= $WARNING ) ? ( 95 ) : ( 85 )) ) : ( 0 )

35 - crit: ($1m_requests > 120) ? ($this < (($status == $CRITICAL) ? ( 85 ) : ( 75 )) ) : ( 0 )

36 - delay: up 2m down 15m multiplier 1.5 max 1h

37 - info: ratio of successful HTTP requests over the last minute (1xx, 2xx, 304, 401)

38 - to: webmaster

39 -

40 - template: 1m_redirects

41 - on: web_log.response_statuses

42 - class: Workload

43 - type: Web Server

44 -component: Web log

45 - families: *

46 - lookup: sum -1m unaligned of redirects

47 - calc: $this * 100 / $1m_requests

48 - units: %

49 - every: 10s

50 - warn: ($1m_requests > 120) ? ($this > (($status >= $WARNING ) ? ( 1 ) : ( 20 )) ) : ( 0 )

51 - delay: up 2m down 15m multiplier 1.5 max 1h

52 - info: ratio of redirection HTTP requests over the last minute (3xx except 304)

53 - to: webmaster

54 -

55 - template: 1m_bad_requests

56 - on: web_log.response_statuses

57 - class: Errors

58 - type: Web Server

59 -component: Web log

60 - families: *

61 - lookup: sum -1m unaligned of bad_requests

62 - calc: $this * 100 / $1m_requests

63 - units: %

64 - every: 10s

65 - warn: ($1m_requests > 120) ? ($this > (($status >= $WARNING) ? ( 10 ) : ( 30 )) ) : ( 0 )

66 - delay: up 2m down 15m multiplier 1.5 max 1h

67 - info: ratio of client error HTTP requests over the last minute (4xx except 401)

68 - to: webmaster

69 -

70 - template: 1m_internal_errors

71 - on: web_log.response_statuses

72 - class: Errors

73 - type: Web Server

74 -component: Web log

75 - families: *

76 - lookup: sum -1m unaligned of server_errors

77 - calc: $this * 100 / $1m_requests

78 - units: %

79 - every: 10s

80 - warn: ($1m_requests > 120) ? ($this > (($status >= $WARNING) ? ( 1 ) : ( 2 )) ) : ( 0 )

81 - crit: ($1m_requests > 120) ? ($this > (($status == $CRITICAL) ? ( 2 ) : ( 5 )) ) : ( 0 )

82 - delay: up 2m down 15m multiplier 1.5 max 1h

83 - info: ratio of server error HTTP requests over the last minute (5xx)

84 - to: webmaster

85 -

86 -# unmatched lines

87 -

88 -# the following alarms trigger only when there are enough data.

89 -# we assume there are enough data when:

90 -#

91 -# $1m_total_requests > 120

92 -#

93 -# i.e. when there are at least 120 requests during the last minute

94 -

95 - template: 1m_total_requests

96 - on: web_log.response_codes

97 - class: Workload

98 - type: Web Server

99 -component: Web log

100 - families: *

101 - lookup: sum -1m unaligned

102 - calc: ($this == 0)?(1):($this)

103 - units: requests

104 - every: 10s

105 - info: number of HTTP requests over the last minute

106 -

107 - template: 1m_unmatched

108 - on: web_log.response_codes

109 - class: Errors

110 - type: Web Server

111 -component: Web log

112 - families: *

113 - lookup: sum -1m unaligned of unmatched

114 - calc: $this * 100 / $1m_total_requests

115 - units: %

116 - every: 10s

117 - warn: ($1m_total_requests > 120) ? ($this > 1) : ( 0 )

118 - delay: up 1m down 5m multiplier 1.5 max 1h

119 - info: percentage of unparsed log lines over the last minute

120 - to: webmaster

121 -

122 -# -----------------------------------------------------------------------------

123 -# web slow

124 -

125 -# the following alarms trigger only when there are enough data.

126 -# we assume there are enough data when:

127 -#

128 -# $1m_requests > 120

129 -#

130 -# i.e. when there are at least 120 requests during the last minute

131 -

132 - template: 10m_response_time

133 - on: web_log.response_time

134 - class: Latency

135 - type: System

136 -component: Web log

137 - families: *

138 - lookup: average -10m unaligned of avg

139 - units: ms

140 - every: 30s

141 - info: average HTTP response time over the last 10 minutes

142 -

143 - template: web_slow

144 - on: web_log.response_time

145 - class: Latency

146 - type: Web Server

147 -component: Web log

148 - families: *

149 - lookup: average -1m unaligned of avg

150 - units: ms

151 - every: 10s

152 - green: 500

153 - red: 1000

154 - warn: ($1m_requests > 120) ? ($this > $green && $this > ($10m_response_time * 2) ) : ( 0 )

155 - crit: ($1m_requests > 120) ? ($this > $red && $this > ($10m_response_time * 4) ) : ( 0 )

156 - delay: down 15m multiplier 1.5 max 1h

157 - info: average HTTP response time over the last minute

158 - options: no-clear-notification

159 - to: webmaster

160 -

161 -# -----------------------------------------------------------------------------

162 -# web too many or too few requests

163 -

164 -# the following alarms trigger only when there are enough data.

165 -# we assume there are enough data when:

166 -#

167 -# $5m_successful_old > 120

168 -#

169 -# i.e. when there were at least 120 requests during the 5 minutes starting

170 -# at -10m and ending at -5m

171 -

172 - template: 5m_successful_old

173 - on: web_log.response_statuses

174 - class: Workload

175 - type: Web Server

176 -component: Web log

177 - families: *

178 - lookup: average -5m at -5m unaligned of successful_requests

179 - units: requests/s

180 - every: 30s

181 - info: average number of successful HTTP requests for the 5 minutes starting 10 minutes ago

182 -

183 - template: 5m_successful

184 - on: web_log.response_statuses

185 - class: Workload

186 - type: Web Server

187 -component: Web log

188 - families: *

189 - lookup: average -5m unaligned of successful_requests

190 - units: requests/s

191 - every: 30s

192 - info: average number of successful HTTP requests over the last 5 minutes

193 -

194 - template: 5m_requests_ratio

195 - on: web_log.response_codes

196 - class: Workload

197 - type: Web Server

198 -component: Web log

199 - families: *

200 - calc: ($5m_successful_old > 0)?($5m_successful * 100 / $5m_successful_old):(100)

201 - units: %

202 - every: 30s

203 - warn: ($5m_successful_old > 120) ? ($this > 200 OR $this < 50) : (0)

204 - crit: ($5m_successful_old > 120) ? ($this > 400 OR $this < 25) : (0)

205 - delay: down 15m multiplier 1.5 max 1h

206 - options: no-clear-notification

207 - info: ratio of successful HTTP requests over the last 5 minutes, \

208 - compared with the previous 5 minutes \

209 - (clear notification for this alarm will not be sent)

210 - to: webmaster

211 -

212 -

213 -

214 -# ---------------------------------------------------GO-VERSION---------------------------------------------------------

215 -

2 # unmatched lines

3

4 # the following alarms trigger only when there are enough data.

chore(health): remove py web_log alarms (#13404)