wbyc/terraform-datadog-old-monitors
2
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

MON-73 Raise all thresholds involving erroneous application behaviors to only handle possible infrastructure failures

Browse Source
This commit is contained in:
Laurent Piroelle 2018-02-13 15:34:31 +01:00 committed by Quentin Manfroi
parent 97192755c4
commit fed0d592e9
14 changed files with 247 additions and 191 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

96
cloud/azure/README.md
View File

@ -31,46 +31,50 @@ Inputs
| Name | Description | Type | Default | Required |
|------|-------------|:----:|:-----:|:-----:|
| apimanagement_failed_requests_threshold_critical | Maximum acceptable percent of failed requests | string | `5` | no |
| apimanagement_other_requests_threshold_critical | Maximum acceptable percent of other requests | string | `5` | no |
| apimanagement_successful_requests_threshold_critical | Minimum acceptable percent of successful requests | string | `90` | no |
| apimanagement_unauthorized_requests_threshold_critical | Maximum acceptable percent of unauthorized requests | string | `5` | no |
| appservices_http_4xx_requests_threshold_critical | Maximum critical acceptable percent of 4xx errors | string | `30` | no |
| appservices_http_4xx_requests_threshold_warning | Maximum warning acceptable percent of 4xx errors | string | `15` | no |
| appservices_http_5xx_requests_threshold_critical | Maximum critical acceptable percent of 5xx errors | string | `20` | no |
| appservices_http_5xx_requests_threshold_warning | Maximum warning acceptable percent of 5xx errors | string | `10` | no |
| appservices_http_successful_requests_threshold_critical | Minimum critical acceptable percent of 2xx & 3xx requests | string | `90` | no |
| appservices_http_successful_requests_threshold_warning | Minimum warning acceptable percent of 2xx & 3xx requests | string | `95` | no |
| appservices_memory_usage_threshold_critical | Alerting threshold in Mib | string | `52430000` | no |
| appservices_memory_usage_threshold_warning | Warning threshold in MiB | string | `33550000` | no |
| appservices_response_time_threshold_critical | Alerting threshold in seconds | string | `0.8` | no |
| appservices_response_time_threshold_warning | Warning threshold in seconds | string | `0.4` | no |
| apimanagement_failed_requests_threshold_critical | Maximum acceptable percent of failed requests | string | `90` | no |
| apimanagement_failed_requests_threshold_warning | Warning regarding acceptable percent of failed requests | string | `50` | no |
| apimanagement_other_requests_threshold_critical | Maximum acceptable percent of other requests | string | `90` | no |
| apimanagement_other_requests_threshold_warning | Warning regarding acceptable percent of other requests | string | `50` | no |
| apimanagement_successful_requests_threshold_critical | Minimum acceptable percent of successful requests | string | `10` | no |
| apimanagement_successful_requests_threshold_warning | Warning regarding acceptable percent of successful requests | string | `30` | no |
| apimanagement_unauthorized_requests_threshold_critical | Maximum acceptable percent of unauthorized requests | string | `90` | no |
| apimanagement_unauthorized_requests_threshold_warning | Warning regarding acceptable percent of unauthorized requests | string | `50` | no |
| appservices_http_4xx_requests_threshold_critical | Maximum critical acceptable percent of 4xx errors | string | `90` | no |
| appservices_http_4xx_requests_threshold_warning | Warning regarding acceptable percent of 4xx errors | string | `50` | no |
| appservices_http_5xx_requests_threshold_critical | Maximum critical acceptable percent of 5xx errors | string | `90` | no |
| appservices_http_5xx_requests_threshold_warning | Warning regarding acceptable percent of 5xx errors | string | `50` | no |
| appservices_http_successful_requests_threshold_critical | Minimum critical acceptable percent of 2xx & 3xx requests | string | `10` | no |
| appservices_http_successful_requests_threshold_warning | Warning regarding acceptable percent of 2xx & 3xx requests | string | `30` | no |
| appservices_memory_usage_threshold_critical | Alerting threshold in Mib | string | `1073741824` | no |
| appservices_memory_usage_threshold_warning | Warning threshold in MiB | string | `536870912` | no |
| appservices_response_time_threshold_critical | Alerting threshold in seconds | string | `10` | no |
| appservices_response_time_threshold_warning | Warning threshold in seconds | string | `5` | no |
| delay | Delay in seconds for the metric evaluation | string | `600` | no |
| environment | Architecture environment | string | - | yes |
| eventhub_errors_rate_thresold_critical | Errors ratio (percentage) to trigger the critical alert | string | `3` | no |
| eventhub_errors_rate_thresold_warning | Errors ratio (percentage) to trigger a warning alert | string | `1` | no |
| eventhub_failed_requests_rate_thresold_critical | Failed requests ratio (percentage) to trigger the critical alert | string | `3` | no |
| eventhub_failed_requests_rate_thresold_warning | Failed requests ratio (percentage) to trigger a warning alert | string | `1` | no |
| eventhub_errors_rate_thresold_critical | Errors ratio (percentage) to trigger the critical alert | string | `90` | no |
| eventhub_errors_rate_thresold_warning | Errors ratio (percentage) to trigger a warning alert | string | `50` | no |
| eventhub_failed_requests_rate_thresold_critical | Failed requests ratio (percentage) to trigger the critical alert | string | `90` | no |
| eventhub_failed_requests_rate_thresold_warning | Failed requests ratio (percentage) to trigger a warning alert | string | `50` | no |
| filter_tags_custom | Tags used for custom filtering when filter_tags_use_defaults is false | string | `*` | no |
| filter_tags_use_defaults | Use default filter tags convention | string | `true` | no |
| iothub_dropped_d2c_telemetry_egress_threshold_critical | D2C Telemetry Dropped limit (critical threshold) | string | `1000` | no |
| iothub_dropped_d2c_telemetry_egress_threshold_warning | D2C Telemetry Dropped limit (warning threshold) | string | `500` | no |
| iothub_failed_c2d_methods_rate_threshold_critical | C2D Methods Failed rate limit (critical threshold) | string | `10` | no |
| iothub_failed_c2d_methods_rate_threshold_warning | C2D Methods Failed rate limit (warning threshold) | string | `0` | no |
| iothub_failed_c2d_twin_read_rate_threshold_critical | C2D Twin Read Failed rate limit (critical threshold) | string | `10` | no |
| iothub_failed_c2d_twin_read_rate_threshold_warning | C2D Twin Read Failed rate limit (warning threshold) | string | `0` | no |
| iothub_failed_c2d_twin_update_rate_threshold_critical | C2D Twin Update Failed rate limit (critical threshold) | string | `10` | no |
| iothub_failed_c2d_twin_update_rate_threshold_warning | C2D Twin Update Failed rate limit (warning threshold) | string | `0` | no |
| iothub_failed_d2c_twin_read_rate_threshold_critical | D2C Twin Read Failed rate limit (critical threshold) | string | `10` | no |
| iothub_failed_d2c_twin_read_rate_threshold_warning | D2C Twin Read Failed rate limit (warning threshold) | string | `0` | no |
| iothub_failed_d2c_twin_update_rate_threshold_critical | D2C Twin Update Failed rate limit (critical threshold) | string | `10` | no |
| iothub_failed_d2c_twin_update_rate_threshold_warning | D2C Twin Update Failed rate limit (warning threshold) | string | `0` | no |
| iothub_failed_jobs_rate_threshold_critical | Jobs Failed rate limit (critical threshold) | string | `10` | no |
| iothub_failed_jobs_rate_threshold_warning | Jobs Failed rate limit (warning threshold) | string | `0` | no |
| iothub_failed_listjobs_rate_threshold_critical | ListJobs Failed rate limit (critical threshold) | string | `10` | no |
| iothub_failed_listjobs_rate_threshold_warning | ListJobs Failed rate limit (warning threshold) | string | `0` | no |
| iothub_failed_queryjobs_rate_threshold_critical | QueryJobs Failed rate limit (critical threshold) | string | `10` | no |
| iothub_failed_queryjobs_rate_threshold_warning | QueryJobs Failed rate limit (warning threshold) | string | `0` | no |
| iothub_failed_c2d_methods_rate_threshold_critical | C2D Methods Failed rate limit (critical threshold) | string | `90` | no |
| iothub_failed_c2d_methods_rate_threshold_warning | C2D Methods Failed rate limit (warning threshold) | string | `50` | no |
| iothub_failed_c2d_twin_read_rate_threshold_critical | C2D Twin Read Failed rate limit (critical threshold) | string | `90` | no |
| iothub_failed_c2d_twin_read_rate_threshold_warning | C2D Twin Read Failed rate limit (warning threshold) | string | `50` | no |
| iothub_failed_c2d_twin_update_rate_threshold_critical | C2D Twin Update Failed rate limit (critical threshold) | string | `90` | no |
| iothub_failed_c2d_twin_update_rate_threshold_warning | C2D Twin Update Failed rate limit (warning threshold) | string | `50` | no |
| iothub_failed_d2c_twin_read_rate_threshold_critical | D2C Twin Read Failed rate limit (critical threshold) | string | `90` | no |
| iothub_failed_d2c_twin_read_rate_threshold_warning | D2C Twin Read Failed rate limit (warning threshold) | string | `50` | no |
| iothub_failed_d2c_twin_update_rate_threshold_critical | D2C Twin Update Failed rate limit (critical threshold) | string | `90` | no |
| iothub_failed_d2c_twin_update_rate_threshold_warning | D2C Twin Update Failed rate limit (warning threshold) | string | `50` | no |
| iothub_failed_jobs_rate_threshold_critical | Jobs Failed rate limit (critical threshold) | string | `90` | no |
| iothub_failed_jobs_rate_threshold_warning | Jobs Failed rate limit (warning threshold) | string | `50` | no |
| iothub_failed_listjobs_rate_threshold_critical | ListJobs Failed rate limit (critical threshold) | string | `90` | no |
| iothub_failed_listjobs_rate_threshold_warning | ListJobs Failed rate limit (warning threshold) | string | `50` | no |
| iothub_failed_queryjobs_rate_threshold_critical | QueryJobs Failed rate limit (critical threshold) | string | `90` | no |
| iothub_failed_queryjobs_rate_threshold_warning | QueryJobs Failed rate limit (warning threshold) | string | `50` | no |
| iothub_fallback_d2c_telemetry_egress_threshold_critical | D2C Telemetry Fallback limit (critical threshold) | string | `1000` | no |
| iothub_fallback_d2c_telemetry_egress_threshold_warning | D2C Telemetry Fallback limit (warning threshold) | string | `500` | no |
| iothub_invalid_d2c_telemetry_egress_threshold_critical | D2C Telemetry Invalid limit (critical threshold) | string | `1000` | no |
@ -92,24 +96,24 @@ Inputs
| sqldatabase_diskspace_threshold_warning | Disk space used in percent (warning threshold) | string | `80` | no |
| sqldatabase_dtu_threshold_critical | Amount of DTU used (critical threshold) | string | `90` | no |
| sqldatabase_dtu_threshold_warning | Amount of DTU used (warning threshold) | string | `85` | no |
| storage_authorization_error_requests_threshold_critical | Maximum acceptable percent of authorization error requests for a storage | string | `50` | no |
| storage_authorization_error_requests_threshold_warning | Warning regarding acceptable percent of authorization error requests for a storage | string | `15` | no |
| storage_authorization_error_requests_threshold_critical | Maximum acceptable percent of authorization error requests for a storage | string | `90` | no |
| storage_authorization_error_requests_threshold_warning | Warning regarding acceptable percent of authorization error requests for a storage | string | `50` | no |
| storage_availability_threshold_critical | Minimum acceptable percent of availability for a storage | string | `50` | no |
| storage_availability_threshold_warning | Warning regarding acceptable percent of availability for a storage | string | `90` | no |
| storage_client_other_error_requests_threshold_critical | Maximum acceptable percent of client other error requests for a storage | string | `50` | no |
| storage_client_other_error_requests_threshold_warning | Warning regarding acceptable percent of client other error requests for a storage | string | `15` | no |
| storage_client_other_error_requests_threshold_critical | Maximum acceptable percent of client other error requests for a storage | string | `90` | no |
| storage_client_other_error_requests_threshold_warning | Warning regarding acceptable percent of client other error requests for a storage | string | `50` | no |
| storage_latency_threshold_critical | Maximum acceptable end to end latency (ms) for a storage | string | `2000` | no |
| storage_latency_threshold_warning | Warning regarding acceptable end to end latency (ms) for a storage | string | `1000` | no |
| storage_network_error_requests_threshold_critical | Maximum acceptable percent of network error requests for a storage | string | `50` | no |
| storage_network_error_requests_threshold_warning | Warning regarding acceptable percent of network error requests for a storage | string | `5` | no |
| storage_server_other_error_requests_threshold_critical | Maximum acceptable percent of server other error requests for a storage | string | `50` | no |
| storage_server_other_error_requests_threshold_warning | Warning regarding acceptable percent of server other error requests for a storage | string | `10` | no |
| storage_network_error_requests_threshold_critical | Maximum acceptable percent of network error requests for a storage | string | `90` | no |
| storage_network_error_requests_threshold_warning | Warning regarding acceptable percent of network error requests for a storage | string | `50` | no |
| storage_server_other_error_requests_threshold_critical | Maximum acceptable percent of server other error requests for a storage | string | `90` | no |
| storage_server_other_error_requests_threshold_warning | Warning regarding acceptable percent of server other error requests for a storage | string | `50` | no |
| storage_successful_requests_threshold_critical | Minimum acceptable percent of successful requests for a storage | string | `10` | no |
| storage_successful_requests_threshold_warning | Warning regarding acceptable percent of successful requests for a storage | string | `30` | no |
| storage_throttling_error_requests_threshold_critical | Maximum acceptable percent of throttling error requests for a storage | string | `50` | no |
| storage_throttling_error_requests_threshold_warning | Warning regarding acceptable percent of throttling error requests for a storage | string | `10` | no |
| storage_timeout_error_requests_threshold_critical | Maximum acceptable percent of timeout error requests for a storage | string | `50` | no |
| storage_timeout_error_requests_threshold_warning | Warning regarding acceptable percent of timeout error requests for a storage | string | `5` | no |
| storage_throttling_error_requests_threshold_critical | Maximum acceptable percent of throttling error requests for a storage | string | `90` | no |
| storage_throttling_error_requests_threshold_warning | Warning regarding acceptable percent of throttling error requests for a storage | string | `50` | no |
| storage_timeout_error_requests_threshold_critical | Maximum acceptable percent of timeout error requests for a storage | string | `90` | no |
| storage_timeout_error_requests_threshold_warning | Warning regarding acceptable percent of timeout error requests for a storage | string | `50` | no |
| streamanalytics_conversion_errors_threshold_critical | Conversion errors limit (critical threshold) | string | `10` | no |
| streamanalytics_conversion_errors_threshold_warning | Conversion errors limit (warning threshold) | string | `0` | no |
| streamanalytics_failed_function_requests_threshold_critical | Failed Function Request rate limit (critical threshold) | string | `10` | no |

12
cloud/azure/apimanagement/README.md
View File

@ -29,13 +29,17 @@ Inputs
|------|-------------|:----:|:-----:|:-----:|
| delay | Delay in seconds for the metric evaluation | string | `600` | no |
| environment | Architecture environment | string | - | yes |
| failed_requests_threshold_critical | Maximum acceptable percent of failed requests | string | `5` | no |
| failed_requests_threshold_critical | Maximum acceptable percent of failed requests | string | `90` | no |
| failed_requests_threshold_warning | Warning regarding acceptable percent of failed requests | string | `50` | no |
| filter_tags_custom | Tags used for custom filtering when filter_tags_use_defaults is false | string | `*` | no |
| filter_tags_use_defaults | Use default filter tags convention | string | `true` | no |
| message | Message sent when a Redis monitor is triggered | string | - | yes |
| other_requests_threshold_critical | Maximum acceptable percent of other requests | string | `5` | no |
| successful_requests_threshold_critical | Minimum acceptable percent of successful requests | string | `90` | no |
| unauthorized_requests_threshold_critical | Maximum acceptable percent of unauthorized requests | string | `5` | no |
| other_requests_threshold_critical | Maximum acceptable percent of other requests | string | `90` | no |
| other_requests_threshold_warning | Warning regarding acceptable percent of other requests | string | `50` | no |
| successful_requests_threshold_critical | Minimum acceptable percent of successful requests | string | `10` | no |
| successful_requests_threshold_warning | Warning regarding acceptable percent of successful requests | string | `30` | no |
| unauthorized_requests_threshold_critical | Maximum acceptable percent of unauthorized requests | string | `90` | no |
| unauthorized_requests_threshold_warning | Warning regarding acceptable percent of unauthorized requests | string | `50` | no |
Related documentation
---------------------

28
cloud/azure/apimanagement/inputs.tf
View File

@ -27,20 +27,40 @@ variable "filter_tags_custom" {
# Azure API Management specific
variable "failed_requests_threshold_critical" {
description = "Maximum acceptable percent of failed requests"
default = 5
default = 90
}
variable "failed_requests_threshold_warning" {
description = "Warning regarding acceptable percent of failed requests"
default = 50
}
variable "other_requests_threshold_critical" {
description = "Maximum acceptable percent of other requests"
default = 5
default = 90
}
variable "other_requests_threshold_warning" {
description = "Warning regarding acceptable percent of other requests"
default = 50
}
variable "unauthorized_requests_threshold_critical" {
description = "Maximum acceptable percent of unauthorized requests"
default = 5
default = 90
}
variable "unauthorized_requests_threshold_warning" {
description = "Warning regarding acceptable percent of unauthorized requests"
default = 50
}
variable "successful_requests_threshold_critical" {
description = "Minimum acceptable percent of successful requests"
default = 90
default = 10
}
variable "successful_requests_threshold_warning" {
description = "Warning regarding acceptable percent of successful requests"
default = 30
}

4
cloud/azure/apimanagement/monitors-azure-apimanagement.tf
View File

@ -49,6 +49,7 @@ resource "datadog_monitor" "apimgt_failed_requests" {
thresholds {
critical = "${var.failed_requests_threshold_critical}"
warning = "${var.failed_requests_threshold_warning}"
}
type = "metric alert"
@ -79,6 +80,7 @@ resource "datadog_monitor" "apimgt_other_requests" {
thresholds {
critical = "${var.other_requests_threshold_critical}"
warning = "${var.other_requests_threshold_warning}"
}
type = "metric alert"
@ -109,6 +111,7 @@ resource "datadog_monitor" "apimgt_unauthorized_requests" {
thresholds {
critical = "${var.unauthorized_requests_threshold_critical}"
warning = "${var.unauthorized_requests_threshold_warning}"
}
type = "metric alert"
@ -139,6 +142,7 @@ resource "datadog_monitor" "apimgt_successful_requests" {
thresholds {
critical = "${var.successful_requests_threshold_critical}"
warning = "${var.successful_requests_threshold_warning}"
}
type = "metric alert"

20
cloud/azure/app-services/README.md
View File

@ -32,17 +32,17 @@ Inputs
| environment | Architecture environment | string | - | yes |
| filter_tags_custom | Tags used for custom filtering when filter_tags_use_defaults is false | string | `*` | no |
| filter_tags_use_defaults | Use default filter tags convention | string | `true` | no |
| http_successful_requests_threshold_critical | Minimum critical acceptable percent of 2xx & 3xx requests | string | `90` | no |
| http_successful_requests_threshold_warning | Minimum warning acceptable percent of 2xx & 3xx requests | string | `95` | no |
| http_4xx_requests_threshold_critical | Maximum critical acceptable percent of 4xx errors | string | `30` | no |
| http_4xx_requests_threshold_warning | Maximum warning acceptable percent of 4xx errors | string | `15` | no |
| http_5xx_requests_threshold_critical | Maximum critical acceptable percent of 5xx errors | string | `20` | no |
| http_5xx_requests_threshold_warning | Maximum warning acceptable percent of 5xx errors | string | `10` | no |
| memory_usage_threshold_critical | Alerting threshold in Mib | string | `52430000` | no |
| memory_usage_threshold_warning | Warning threshold in MiB | string | `33550000` | no |
| http_4xx_requests_threshold_critical | Maximum critical acceptable percent of 4xx errors | string | `90` | no |
| http_4xx_requests_threshold_warning | Maximum warning acceptable percent of 4xx errors | string | `50` | no |
| http_5xx_requests_threshold_critical | Maximum critical acceptable percent of 5xx errors | string | `90` | no |
| http_5xx_requests_threshold_warning | Maximum warning acceptable percent of 5xx errors | string | `50` | no |
| http_successful_requests_threshold_critical | Minimum critical acceptable percent of 2xx & 3xx requests | string | `10` | no |
| http_successful_requests_threshold_warning | Minimum warning acceptable percent of 2xx & 3xx requests | string | `30` | no |
| memory_usage_threshold_critical | Alerting threshold in Mib | string | `1073741824` | no |
| memory_usage_threshold_warning | Warning threshold in MiB | string | `536870912` | no |
| message | Message sent when a monitor is triggered | string | - | yes |
| response_time_threshold_critical | Alerting threshold in seconds | string | `0.8` | no |
| response_time_threshold_warning | Warning threshold in seconds | string | `0.4` | no |
| response_time_threshold_critical | Alerting threshold in seconds | string | `10` | no |
| response_time_threshold_warning | Warning threshold in seconds | string | `5` | no |
Related documentation
---------------------

26
cloud/azure/app-services/inputs.tf
View File

@ -27,12 +27,12 @@ variable "delay" {
###################################
variable "response_time_threshold_critical" {
default = 0.8
default = 10
description = "Alerting threshold in seconds"
}
variable "response_time_threshold_warning" {
default = 0.4
default = 5
description = "Warning threshold in seconds"
}
@ -41,12 +41,12 @@ variable "response_time_threshold_warning" {
###################################
variable "memory_usage_threshold_critical" {
default = 52430000
default = 1073741824 # 1Gb
description = "Alerting threshold in Mib"
}
variable "memory_usage_threshold_warning" {
default = 33550000
default = 536870912 # 512Mb
description = "Warning threshold in MiB"
}
@ -55,13 +55,13 @@ variable "memory_usage_threshold_warning" {
#################################
variable "http_5xx_requests_threshold_critical" {
default = 20
default = 90
description = "Maximum critical acceptable percent of 5xx errors"
}
variable "http_5xx_requests_threshold_warning" {
default = 10
description = "Maximum warning acceptable percent of 5xx errors"
default = 50
description = "Warning regarding acceptable percent of 5xx errors"
}
#################################
@ -69,13 +69,13 @@ variable "http_5xx_requests_threshold_warning" {
#################################
variable "http_4xx_requests_threshold_critical" {
default = 30
default = 90
description = "Maximum critical acceptable percent of 4xx errors"
}
variable "http_4xx_requests_threshold_warning" {
default = 15
description = "Maximum warning acceptable percent of 4xx errors"
default = 50
description = "Warning regarding acceptable percent of 4xx errors"
}
#################################
@ -83,11 +83,11 @@ variable "http_4xx_requests_threshold_warning" {
#################################
variable "http_successful_requests_threshold_critical" {
default = 90
default = 10
description = "Minimum critical acceptable percent of 2xx & 3xx requests"
}
variable "http_successful_requests_threshold_warning" {
default = 95
description = "Minimum warning acceptable percent of 2xx & 3xx requests"
default = 30
description = "Warning regarding acceptable percent of 2xx & 3xx requests"
}

8
cloud/azure/eventhub/README.md
View File

@ -29,10 +29,10 @@ Inputs
|------|-------------|:----:|:-----:|:-----:|
| delay | Delay in seconds for the metric evaluation | string | `600` | no |
| environment | Architecture environment | string | - | yes |
| errors_rate_thresold_critical | Errors ratio (percentage) to trigger the critical alert | string | `3` | no |
| errors_rate_thresold_warning | Errors ratio (percentage) to trigger a warning alert | string | `1` | no |
| failed_requests_rate_thresold_critical | Failed requests ratio (percentage) to trigger the critical alert | string | `3` | no |
| failed_requests_rate_thresold_warning | Failed requests ratio (percentage) to trigger a warning alert | string | `1` | no |
| errors_rate_thresold_critical | Errors ratio (percentage) to trigger the critical alert | string | `90` | no |
| errors_rate_thresold_warning | Errors ratio (percentage) to trigger a warning alert | string | `50` | no |
| failed_requests_rate_thresold_critical | Failed requests ratio (percentage) to trigger the critical alert | string | `90` | no |
| failed_requests_rate_thresold_warning | Failed requests ratio (percentage) to trigger a warning alert | string | `50` | no |
| filter_tags_custom | Tags used for custom filtering when filter_tags_use_defaults is false | string | `*` | no |
| filter_tags_use_defaults | Use default filter tags convention | string | `true` | no |
| message | Message sent when an alert is triggered | string | - | yes |

8
cloud/azure/eventhub/inputs.tf
View File

@ -26,20 +26,20 @@ variable "filter_tags_custom" {
variable "failed_requests_rate_thresold_critical" {
description = "Failed requests ratio (percentage) to trigger the critical alert"
default = 3
default = 90
}
variable "failed_requests_rate_thresold_warning" {
description = "Failed requests ratio (percentage) to trigger a warning alert"
default = 1
default = 50
}
variable "errors_rate_thresold_critical" {
description = "Errors ratio (percentage) to trigger the critical alert"
default = 3
default = 90
}
variable "errors_rate_thresold_warning" {
description = "Errors ratio (percentage) to trigger a warning alert"
default = 1
default = 50
}

118
cloud/azure/inputs.tf
View File

@ -31,175 +31,195 @@ variable "non_taggable_filter_tags" {
# Azure API Management specific variables
variable "apimanagement_failed_requests_threshold_critical" {
description = "Maximum acceptable percent of failed requests"
default = 5
default = 90
}
variable "apimanagement_failed_requests_threshold_warning" {
description = "Warning regarding acceptable percent of failed requests"
default = 50
}
variable "apimanagement_other_requests_threshold_critical" {
description = "Maximum acceptable percent of other requests"
default = 5
default = 90
}
variable "apimanagement_other_requests_threshold_warning" {
description = "Warning regarding acceptable percent of other requests"
default = 50
}
variable "apimanagement_unauthorized_requests_threshold_critical" {
description = "Maximum acceptable percent of unauthorized requests"
default = 5
default = 90
}
variable "apimanagement_unauthorized_requests_threshold_warning" {
description = "Warning regarding acceptable percent of unauthorized requests"
default = 50
}
variable "apimanagement_successful_requests_threshold_critical" {
description = "Minimum acceptable percent of successful requests"
default = 90
default = 10
}
variable "apimanagement_successful_requests_threshold_warning" {
description = "Warning regarding acceptable percent of successful requests"
default = 30
}
# Azure App Services specific variables
variable "appservices_response_time_threshold_critical" {
default = 0.8
default = 10
description = "Alerting threshold in seconds"
}
variable "appservices_response_time_threshold_warning" {
default = 0.4
default = 5
description = "Warning threshold in seconds"
}
variable "appservices_memory_usage_threshold_critical" {
default = 52430000
default = 1073741824 # 1Gb
description = "Alerting threshold in Mib"
}
variable "appservices_memory_usage_threshold_warning" {
default = 33550000
default = 536870912 # 512Mb
description = "Warning threshold in MiB"
}
variable "appservices_http_4xx_requests_threshold_critical" {
default = 30
default = 90
description = "Maximum critical acceptable percent of 4xx errors"
}
variable "appservices_http_4xx_requests_threshold_warning" {
default = 15
description = "Maximum warning acceptable percent of 4xx errors"
default = 50
description = "Warning regarding acceptable percent of 4xx errors"
}
variable "appservices_http_5xx_requests_threshold_critical" {
default = 20
default = 90
description = "Maximum critical acceptable percent of 5xx errors"
}
variable "appservices_http_5xx_requests_threshold_warning" {
default = 10
description = "Maximum warning acceptable percent of 5xx errors"
default = 50
description = "Warning regarding acceptable percent of 5xx errors"
}
variable "appservices_http_successful_requests_threshold_critical" {
default = 90
default = 10
description = "Minimum critical acceptable percent of 2xx & 3xx requests"
}
variable "appservices_http_successful_requests_threshold_warning" {
default = 95
description = "Minimum warning acceptable percent of 2xx & 3xx requests"
default = 30
description = "Warning regarding acceptable percent of 2xx & 3xx requests"
}
# Azure Event Hub specific variables
variable "eventhub_failed_requests_rate_thresold_critical" {
description = "Failed requests ratio (percentage) to trigger the critical alert"
default = 3
default = 90
}
variable "eventhub_failed_requests_rate_thresold_warning" {
description = "Failed requests ratio (percentage) to trigger a warning alert"
default = 1
default = 50
}
variable "eventhub_errors_rate_thresold_critical" {
description = "Errors ratio (percentage) to trigger the critical alert"
default = 3
default = 90
}
variable "eventhub_errors_rate_thresold_warning" {
description = "Errors ratio (percentage) to trigger a warning alert"
default = 1
default = 50
}
# IOT Hub specific variables
variable "iothub_failed_jobs_rate_threshold_warning" {
description = "Jobs Failed rate limit (warning threshold)"
default = 0
default = 50
}
variable "iothub_failed_jobs_rate_threshold_critical" {
description = "Jobs Failed rate limit (critical threshold)"
default = 10
default = 90
}
variable "iothub_failed_listjobs_rate_threshold_warning" {
description = "ListJobs Failed rate limit (warning threshold)"
default = 0
default = 50
}
variable "iothub_failed_listjobs_rate_threshold_critical" {
description = "ListJobs Failed rate limit (critical threshold)"
default = 10
default = 90
}
variable "iothub_failed_queryjobs_rate_threshold_warning" {
description = "QueryJobs Failed rate limit (warning threshold)"
default = 0
default = 50
}
variable "iothub_failed_queryjobs_rate_threshold_critical" {
description = "QueryJobs Failed rate limit (critical threshold)"
default = 10
default = 90
}
variable "iothub_failed_c2d_methods_rate_threshold_warning" {
description = "C2D Methods Failed rate limit (warning threshold)"
default = 0
default = 50
}
variable "iothub_failed_c2d_methods_rate_threshold_critical" {
description = "C2D Methods Failed rate limit (critical threshold)"
default = 10
default = 90
}
variable "iothub_failed_c2d_twin_read_rate_threshold_warning" {
description = "C2D Twin Read Failed rate limit (warning threshold)"
default = 0
default = 50
}
variable "iothub_failed_c2d_twin_read_rate_threshold_critical" {
description = "C2D Twin Read Failed rate limit (critical threshold)"
default = 10
default = 90
}
variable "iothub_failed_c2d_twin_update_rate_threshold_warning" {
description = "C2D Twin Update Failed rate limit (warning threshold)"
default = 0
default = 50
}
variable "iothub_failed_c2d_twin_update_rate_threshold_critical" {
description = "C2D Twin Update Failed rate limit (critical threshold)"
default = 10
default = 90
}
variable "iothub_failed_d2c_twin_read_rate_threshold_warning" {
description = "D2C Twin Read Failed rate limit (warning threshold)"
default = 0
default = 50
}
variable "iothub_failed_d2c_twin_read_rate_threshold_critical" {
description = "D2C Twin Read Failed rate limit (critical threshold)"
default = 10
default = 90
}
variable "iothub_failed_d2c_twin_update_rate_threshold_warning" {
description = "D2C Twin Update Failed rate limit (warning threshold)"
default = 0
default = 50
}
variable "iothub_failed_d2c_twin_update_rate_threshold_critical" {
description = "D2C Twin Update Failed rate limit (critical threshold)"
default = 10
default = 90
}
variable "iothub_dropped_d2c_telemetry_egress_threshold_warning" {
@ -342,62 +362,62 @@ variable "storage_latency_threshold_warning" {
variable "storage_timeout_error_requests_threshold_critical" {
description = "Maximum acceptable percent of timeout error requests for a storage"
default = 50
default = 90
}
variable "storage_timeout_error_requests_threshold_warning" {
description = "Warning regarding acceptable percent of timeout error requests for a storage"
default = 5
default = 50
}
variable "storage_network_error_requests_threshold_critical" {
description = "Maximum acceptable percent of network error requests for a storage"
default = 50
default = 90
}
variable "storage_network_error_requests_threshold_warning" {
description = "Warning regarding acceptable percent of network error requests for a storage"
default = 5
default = 50
}
variable "storage_throttling_error_requests_threshold_critical" {
description = "Maximum acceptable percent of throttling error requests for a storage"
default = 50
default = 90
}
variable "storage_throttling_error_requests_threshold_warning" {
description = "Warning regarding acceptable percent of throttling error requests for a storage"
default = 10
default = 50
}
variable "storage_server_other_error_requests_threshold_critical" {
description = "Maximum acceptable percent of server other error requests for a storage"
default = 50
default = 90
}
variable "storage_server_other_error_requests_threshold_warning" {
description = "Warning regarding acceptable percent of server other error requests for a storage"
default = 10
default = 50
}
variable "storage_client_other_error_requests_threshold_critical" {
description = "Maximum acceptable percent of client other error requests for a storage"
default = 50
default = 90
}
variable "storage_client_other_error_requests_threshold_warning" {
description = "Warning regarding acceptable percent of client other error requests for a storage"
default = 15
default = 50
}
variable "storage_authorization_error_requests_threshold_critical" {
description = "Maximum acceptable percent of authorization error requests for a storage"
default = 50
default = 90
}
variable "storage_authorization_error_requests_threshold_warning" {
description = "Warning regarding acceptable percent of authorization error requests for a storage"
default = 15
default = 50
}
# Azure Stream Analytics specific variables

34
cloud/azure/iothubs/README.md
View File

@ -42,25 +42,25 @@ Inputs
| dropped_d2c_telemetry_egress_threshold_critical | D2C Telemetry Dropped limit (critical threshold) | string | `1000` | no |
| dropped_d2c_telemetry_egress_threshold_warning | D2C Telemetry Dropped limit (warning threshold) | string | `500` | no |
| environment | Architecture Environment | string | - | yes |
| failed_c2d_methods_rate_threshold_critical | C2D Methods Failed rate limit (critical threshold) | string | `10` | no |
| failed_c2d_methods_rate_threshold_warning | C2D Methods Failed rate limit (warning threshold) | string | `0` | no |
| failed_c2d_twin_read_rate_threshold_critical | C2D Twin Read Failed rate limit (critical threshold) | string | `10` | no |
| failed_c2d_twin_read_rate_threshold_warning | C2D Twin Read Failed rate limit (warning threshold) | string | `0` | no |
| failed_c2d_twin_update_rate_threshold_critical | C2D Twin Update Failed rate limit (critical threshold) | string | `10` | no |
| failed_c2d_twin_update_rate_threshold_warning | C2D Twin Update Failed rate limit (warning threshold) | string | `0` | no |
| failed_d2c_twin_read_rate_threshold_critical | D2C Twin Read Failed rate limit (critical threshold) | string | `10` | no |
| failed_d2c_twin_read_rate_threshold_warning | D2C Twin Read Failed rate limit (warning threshold) | string | `0` | no |
| failed_d2c_twin_update_rate_threshold_critical | D2C Twin Update Failed rate limit (critical threshold) | string | `10` | no |
| failed_d2c_twin_update_rate_threshold_warning | D2C Twin Update Failed rate limit (warning threshold) | string | `0` | no |
| failed_jobs_rate_threshold_critical | Jobs Failed rate limit (critical threshold) | string | `10` | no |
| failed_jobs_rate_threshold_warning | Jobs Failed rate limit (warning threshold) | string | `0` | no |
| failed_listjobs_rate_threshold_critical | ListJobs Failed rate limit (critical threshold) | string | `10` | no |
| failed_listjobs_rate_threshold_warning | ListJobs Failed rate limit (warning threshold) | string | `0` | no |
| failed_queryjobs_rate_threshold_critical | QueryJobs Failed rate limit (critical threshold) | string | `10` | no |
| failed_queryjobs_rate_threshold_warning | QueryJobs Failed rate limit (warning threshold) | string | `0` | no |
| failed_c2d_methods_rate_threshold_critical | C2D Methods Failed rate limit (critical threshold) | string | `90` | no |
| failed_c2d_methods_rate_threshold_warning | C2D Methods Failed rate limit (warning threshold) | string | `50` | no |
| failed_c2d_twin_read_rate_threshold_critical | C2D Twin Read Failed rate limit (critical threshold) | string | `90` | no |
| failed_c2d_twin_read_rate_threshold_warning | C2D Twin Read Failed rate limit (warning threshold) | string | `50` | no |
| failed_c2d_twin_update_rate_threshold_critical | C2D Twin Update Failed rate limit (critical threshold) | string | `90` | no |
| failed_c2d_twin_update_rate_threshold_warning | C2D Twin Update Failed rate limit (warning threshold) | string | `50` | no |
| failed_d2c_twin_read_rate_threshold_critical | D2C Twin Read Failed rate limit (critical threshold) | string | `90` | no |
| failed_d2c_twin_read_rate_threshold_warning | D2C Twin Read Failed rate limit (warning threshold) | string | `50` | no |
| failed_d2c_twin_update_rate_threshold_critical | D2C Twin Update Failed rate limit (critical threshold) | string | `90` | no |
| failed_d2c_twin_update_rate_threshold_warning | D2C Twin Update Failed rate limit (warning threshold) | string | `50` | no |
| failed_jobs_rate_threshold_critical | Jobs Failed rate limit (critical threshold) | string | `90` | no |
| failed_jobs_rate_threshold_warning | Jobs Failed rate limit (warning threshold) | string | `50` | no |
| failed_listjobs_rate_threshold_critical | ListJobs Failed rate limit (critical threshold) | string | `90` | no |
| failed_listjobs_rate_threshold_warning | ListJobs Failed rate limit (warning threshold) | string | `50` | no |
| failed_queryjobs_rate_threshold_critical | QueryJobs Failed rate limit (critical threshold) | string | `90` | no |
| failed_queryjobs_rate_threshold_warning | QueryJobs Failed rate limit (warning threshold) | string | `50` | no |
| fallback_d2c_telemetry_egress_threshold_critical | D2C Telemetry Fallback limit (critical threshold) | string | `1000` | no |
| fallback_d2c_telemetry_egress_threshold_warning | D2C Telemetry Fallback limit (warning threshold) | string | `500` | no |
| filter_tags | Tags used for custom filtering when filter_tags_use_defaults is false | string | `*` | no |
| filter_tags | Tags used for filtering | string | `*` | no |
| invalid_d2c_telemetry_egress_threshold_critical | D2C Telemetry Invalid limit (critical threshold) | string | `1000` | no |
| invalid_d2c_telemetry_egress_threshold_warning | D2C Telemetry Invalid limit (warning threshold) | string | `500` | no |
| message | Message sent when an alert is triggered | string | - | yes |

32
cloud/azure/iothubs/inputs.tf
View File

@ -22,82 +22,82 @@ variable "filter_tags" {
# Azure IOT hubs specific
variable "failed_jobs_rate_threshold_warning" {
description = "Jobs Failed rate limit (warning threshold)"
default = 0
default = 50
}
variable "failed_jobs_rate_threshold_critical" {
description = "Jobs Failed rate limit (critical threshold)"
default = 10
default = 90
}
variable "failed_listjobs_rate_threshold_warning" {
description = "ListJobs Failed rate limit (warning threshold)"
default = 0
default = 50
}
variable "failed_listjobs_rate_threshold_critical" {
description = "ListJobs Failed rate limit (critical threshold)"
default = 10
default = 90
}
variable "failed_queryjobs_rate_threshold_warning" {
description = "QueryJobs Failed rate limit (warning threshold)"
default = 0
default = 50
}
variable "failed_queryjobs_rate_threshold_critical" {
description = "QueryJobs Failed rate limit (critical threshold)"
default = 10
default = 90
}
variable "failed_c2d_methods_rate_threshold_warning" {
description = "C2D Methods Failed rate limit (warning threshold)"
default = 0
default = 50
}
variable "failed_c2d_methods_rate_threshold_critical" {
description = "C2D Methods Failed rate limit (critical threshold)"
default = 10
default = 90
}
variable "failed_c2d_twin_read_rate_threshold_warning" {
description = "C2D Twin Read Failed rate limit (warning threshold)"
default = 0
default = 50
}
variable "failed_c2d_twin_read_rate_threshold_critical" {
description = "C2D Twin Read Failed rate limit (critical threshold)"
default = 10
default = 90
}
variable "failed_c2d_twin_update_rate_threshold_warning" {
description = "C2D Twin Update Failed rate limit (warning threshold)"
default = 0
default = 50
}
variable "failed_c2d_twin_update_rate_threshold_critical" {
description = "C2D Twin Update Failed rate limit (critical threshold)"
default = 10
default = 90
}
variable "failed_d2c_twin_read_rate_threshold_warning" {
description = "D2C Twin Read Failed rate limit (warning threshold)"
default = 0
default = 50
}
variable "failed_d2c_twin_read_rate_threshold_critical" {
description = "D2C Twin Read Failed rate limit (critical threshold)"
default = 10
default = 90
}
variable "failed_d2c_twin_update_rate_threshold_warning" {
description = "D2C Twin Update Failed rate limit (warning threshold)"
default = 0
default = 50
}
variable "failed_d2c_twin_update_rate_threshold_critical" {
description = "D2C Twin Update Failed rate limit (critical threshold)"
default = 10
default = 90
}
variable "dropped_d2c_telemetry_egress_threshold_warning" {

4
cloud/azure/monitors.tf
View File

@ -9,9 +9,13 @@ module "apimanagement" {
filter_tags_custom = "${var.filter_tags_custom}"
failed_requests_threshold_critical = "${var.apimanagement_failed_requests_threshold_critical}"
failed_requests_threshold_warning = "${var.apimanagement_failed_requests_threshold_warning}"
other_requests_threshold_critical = "${var.apimanagement_other_requests_threshold_critical}"
other_requests_threshold_warning = "${var.apimanagement_other_requests_threshold_warning}"
successful_requests_threshold_critical = "${var.apimanagement_successful_requests_threshold_critical}"
successful_requests_threshold_warning = "${var.apimanagement_successful_requests_threshold_warning}"
unauthorized_requests_threshold_critical = "${var.apimanagement_unauthorized_requests_threshold_critical}"
unauthorized_requests_threshold_warning = "${var.apimanagement_unauthorized_requests_threshold_warning}"
}
module "appservices" {

24
cloud/azure/storage/README.md
View File

@ -32,12 +32,12 @@ Inputs
| Name | Description | Type | Default | Required |
|------|-------------|:----:|:-----:|:-----:|
| authorization_error_requests_threshold_critical | Maximum acceptable percent of authorization error requests for a storage | string | `50` | no |
| authorization_error_requests_threshold_warning | Warning regarding acceptable percent of authorization error requests for a storage | string | `15` | no |
| authorization_error_requests_threshold_critical | Maximum acceptable percent of authorization error requests for a storage | string | `90` | no |
| authorization_error_requests_threshold_warning | Warning regarding acceptable percent of authorization error requests for a storage | string | `50` | no |
| availability_threshold_critical | Minimum acceptable percent of availability for a storage | string | `50` | no |
| availability_threshold_warning | Warning regarding acceptable percent of availability for a storage | string | `90` | no |
| client_other_error_requests_threshold_critical | Maximum acceptable percent of client other error requests for a storage | string | `50` | no |
| client_other_error_requests_threshold_warning | Warning regarding acceptable percent of client other error requests for a storage | string | `15` | no |
| client_other_error_requests_threshold_critical | Maximum acceptable percent of client other error requests for a storage | string | `90` | no |
| client_other_error_requests_threshold_warning | Warning regarding acceptable percent of client other error requests for a storage | string | `50` | no |
| delay | Delay in seconds for the metric evaluation | string | `600` | no |
| environment | Architecture environment | string | - | yes |
| filter_tags_custom | Tags used for custom filtering when filter_tags_use_defaults is false | string | `*` | no |
@ -45,16 +45,16 @@ Inputs
| latency_threshold_critical | Maximum acceptable end to end latency (ms) for a storage | string | `2000` | no |
| latency_threshold_warning | Warning regarding acceptable end to end latency (ms) for a storage | string | `1000` | no |
| message | Message sent when a Redis monitor is triggered | string | - | yes |
| network_error_requests_threshold_critical | Maximum acceptable percent of network error requests for a storage | string | `50` | no |
| network_error_requests_threshold_warning | Warning regarding acceptable percent of network error requests for a storage | string | `5` | no |
| server_other_error_requests_threshold_critical | Maximum acceptable percent of server other error requests for a storage | string | `50` | no |
| server_other_error_requests_threshold_warning | Warning regarding acceptable percent of server other error requests for a storage | string | `10` | no |
| network_error_requests_threshold_critical | Maximum acceptable percent of network error requests for a storage | string | `90` | no |
| network_error_requests_threshold_warning | Warning regarding acceptable percent of network error requests for a storage | string | `50` | no |
| server_other_error_requests_threshold_critical | Maximum acceptable percent of server other error requests for a storage | string | `90` | no |
| server_other_error_requests_threshold_warning | Warning regarding acceptable percent of server other error requests for a storage | string | `50` | no |
| successful_requests_threshold_critical | Minimum acceptable percent of successful requests for a storage | string | `10` | no |
| successful_requests_threshold_warning | Warning regarding acceptable percent of successful requests for a storage | string | `30` | no |
| throttling_error_requests_threshold_critical | Maximum acceptable percent of throttling error requests for a storage | string | `50` | no |
| throttling_error_requests_threshold_warning | Warning regarding acceptable percent of throttling error requests for a storage | string | `10` | no |
| timeout_error_requests_threshold_critical | Maximum acceptable percent of timeout error requests for a storage | string | `50` | no |
| timeout_error_requests_threshold_warning | Warning regarding acceptable percent of timeout error requests for a storage | string | `5` | no |
| throttling_error_requests_threshold_critical | Maximum acceptable percent of throttling error requests for a storage | string | `90` | no |
| throttling_error_requests_threshold_warning | Warning regarding acceptable percent of throttling error requests for a storage | string | `50` | no |
| timeout_error_requests_threshold_critical | Maximum acceptable percent of timeout error requests for a storage | string | `90` | no |
| timeout_error_requests_threshold_warning | Warning regarding acceptable percent of timeout error requests for a storage | string | `50` | no |
Related documentation
---------------------

24
cloud/azure/storage/inputs.tf
View File

@ -57,60 +57,60 @@ variable "latency_threshold_warning" {
variable "timeout_error_requests_threshold_critical" {
description = "Maximum acceptable percent of timeout error requests for a storage"
default = 50
default = 90
}
variable "timeout_error_requests_threshold_warning" {
description = "Warning regarding acceptable percent of timeout error requests for a storage"
default = 5
default = 50
}
variable "network_error_requests_threshold_critical" {
description = "Maximum acceptable percent of network error requests for a storage"
default = 50
default = 90
}
variable "network_error_requests_threshold_warning" {
description = "Warning regarding acceptable percent of network error requests for a storage"
default = 5
default = 50
}
variable "throttling_error_requests_threshold_critical" {
description = "Maximum acceptable percent of throttling error requests for a storage"
default = 50
default = 90
}
variable "throttling_error_requests_threshold_warning" {
description = "Warning regarding acceptable percent of throttling error requests for a storage"
default = 10
default = 50
}
variable "server_other_error_requests_threshold_critical" {
description = "Maximum acceptable percent of server other error requests for a storage"
default = 50
default = 90
}
variable "server_other_error_requests_threshold_warning" {
description = "Warning regarding acceptable percent of server other error requests for a storage"
default = 10
default = 50
}
variable "client_other_error_requests_threshold_critical" {
description = "Maximum acceptable percent of client other error requests for a storage"
default = 50
default = 90
}
variable "client_other_error_requests_threshold_warning" {
description = "Warning regarding acceptable percent of client other error requests for a storage"
default = 15
default = 50
}
variable "authorization_error_requests_threshold_critical" {
description = "Maximum acceptable percent of authorization error requests for a storage"
default = 50
default = 90
}
variable "authorization_error_requests_threshold_warning" {
description = "Warning regarding acceptable percent of authorization error requests for a storage"
default = 15
default = 50
}