Untitled

mail@pastecode.io avatar
unknown
plain_text
2 years ago
20 kB
5
Indexable
Never
#########  General vars #########
aws_profile_name = "prod-admin"
aws_region = "us-east-1"
lb_type = "ApplicationELB"  # one of: ApplicationELB / ELB
lb_name = "Spot-lb-hc-ITF"
alb_id = "app/Spot-lb-hc-ITF/2f86b53ef66a07c6" # for ALB: this should be the suffix of the ARN staring from "app"
tg_ids = ["targetgroup/sig-4ac4425a-4/48e9bc74d4d3a855"]   # this is an optional param, used in alarms-hosts.tf if lb_type = "ApplicationELB"


alarm_actions = ["arn:aws:sns:us-east-1:178579023202:terraform-cloudwatch-lb-alarms"]
ok_actions    = ["arn:aws:sns:us-east-1:178579023202:terraform-cloudwatch-lb-alarms"]
insufficient_data_actions = []

#########  Backend HTTP Codes Alarms ##########

# High HTTP 5xx - Sum, above 3 for 1 consecutive period of 1 min - P2
evaluation_periods_high_http_5XX        = "1"
period_high_http_5XX                    = "60"
threshold_high_http_5XX                 = "3"
actions_enabled_high_http_5XX           = true
treat_missing_data_high_http_5XX        = "notBreaching"
datapoints_to_alarm_high_http_5XX       = "1"
alarm_actions_high_http_5XX             = ["arn:aws:sns:us-east-1:178579023202:terraform-cloudwatch-lb-alarms"]
ok_actions_high_http_5XX                = ["arn:aws:sns:us-east-1:178579023202:terraform-cloudwatch-lb-alarms"]


# High healthy hosts - Average, above 2 for 110 consecutive periods of 1 min - P1
evaluation_periods_high_healthy_hosts         = "110"
period_high_healthy_hosts                     = "60"
threshold_high_healthy_hosts                  = "2"
actions_enabled_high_healthy_hosts            = true
treat_missing_data_high_healthy_hosts         = "breaching"
datapoints_to_alarm_high_healthy_hosts        = "110"

# Critical HTTP 5xx - Sum, above 10 for 2 consecutive period of 1 min - P1
evaluation_periods_critical_http_5XX        = "2"
period_critical_http_5XX                    = "60"
threshold_critical_http_5XX                 = "10"
actions_enabled_critical_http_5XX           = true
treat_missing_data_critical_http_5XX        = "notBreaching"
datapoints_to_alarm_critical_http_5XX       = "2"

# Fatal HTTP 5xx - Sum, above 100 for 2 consecutive period of 1 min - P1
evaluation_periods_fatal_http_5XX        = "2"
period_fatal_http_5XX                    = "60"
threshold_fatal_http_5XX                 = "100"
actions_enabled_fatal_http_5XX           = true
treat_missing_data_fatal_http_5XX        = "notBreaching"
datapoints_to_alarm_fatal_http_5XX       = "2"
alarm_actions_fatal_http_5XX             = ["arn:aws:sns:us-east-1:178579023202:terraform-cloudwatch-lb-alarms"]
ok_actions_fatal_http_5XX                = ["arn:aws:sns:us-east-1:178579023202:terraform-cloudwatch-lb-alarms"]

# High HTTP 4xx - Sum, above 5 for 2 consecutive period of 1 min - P2
evaluation_periods_high_http_4XX        = "2"
period_high_http_4XX                    = "60"
threshold_high_http_4XX                 = "5"
actions_enabled_high_http_4XX           = true
treat_missing_data_high_http_4XX        = "notBreaching"
datapoints_to_alarm_high_http_4XX       = "2"
alarm_actions_high_http_4XX             = ["arn:aws:sns:us-east-1:178579023202:terraform-cloudwatch-lb-alarms"]
ok_actions_high_http_4XX                = ["arn:aws:sns:us-east-1:178579023202:terraform-cloudwatch-lb-alarms"]

# Critical HTTP 4xx - Sum, above 10 for 3 consecutive period of 1 min - P1
evaluation_periods_critical_http_4XX        = "3"
period_critical_http_4XX                    = "60"
threshold_critical_http_4XX                 = "10"
actions_enabled_critical_http_4XX           = true
treat_missing_data_critical_http_4XX        = "notBreaching"
datapoints_to_alarm_critical_http_4XX       = "3"

# Fatal HTTP 4xx - Sum, above 100 for 3 consecutive period of 1 min - P1
evaluation_periods_fatal_http_4XX        = "3"
period_fatal_http_4XX                    = "60"
threshold_fatal_http_4XX                 = "100"
actions_enabled_fatal_http_4XX           = true
treat_missing_data_fatal_http_4XX        = "notBreaching"
datapoints_to_alarm_fatal_http_4XX       = "3"
alarm_actions_fatal_http_4XX             = ["arn:aws:sns:us-east-1:178579023202:terraform-cloudwatch-lb-alarms"]
ok_actions_fatal_http_4XX                = ["arn:aws:sns:us-east-1:178579023202:terraform-cloudwatch-lb-alarms"]

# High HTTP 3xx - Sum, above 2 for 1 consecutive period of 1 min - P2
configure_alarm_high_http_3XX           = false
evaluation_periods_high_http_3XX        = "1"
period_high_http_3XX                    = "60"
threshold_high_http_3XX                 = "2"
actions_enabled_high_http_3XX           = true
treat_missing_data_high_http_3XX        = "notBreaching"
datapoints_to_alarm_high_http_3XX       = "1"
alarm_actions_high_http_3XX             = ["arn:aws:sns:us-east-1:178579023202:terraform-cloudwatch-lb-alarms"]
ok_actions_high_http_3XX                = ["arn:aws:sns:us-east-1:178579023202:terraform-cloudwatch-lb-alarms"]

#########  ELB HTTP Codes Alarms #########
# ELB 5xx Anomaly - P2
configure_alarm_anomaly_elb_5XX             = false
evaluation_periods_anomaly_elb_5XX          = "5"
actions_enabled_anomaly_elb_5XX             = true
treat_missing_data_anomaly_elb_5XX          = "notBreaching"
period_anomaly_elb_5XX                      = "60"
band_anomaly_elb_5XX                        = "5"
alarm_actions_anomaly_elb_5XX               = ["arn:aws:sns:us-east-1:178579023202:terraform-cloudwatch-lb-alarms"]
ok_actions_anomaly_elb_5XX                  = ["arn:aws:sns:us-east-1:178579023202:terraform-cloudwatch-lb-alarms"]

# High ELB 5xx 1m - Sum, above 5 for 2 consecutive period of 1 min - P2
evaluation_periods_high_elb_5XX_short        = "2"
period_high_elb_5XX_short                    = "60"
threshold_high_elb_5XX_short                 = "5"
actions_enabled_high_elb_5XX_short           = true
treat_missing_data_high_elb_5XX_short        = "notBreaching"
datapoints_to_alarm_high_elb_5XX_short       = "2"
alarm_actions_high_elb_5XX_short             = ["arn:aws:sns:us-east-1:178579023202:terraform-cloudwatch-lb-alarms"]
ok_actions_high_elb_5XX_short                = ["arn:aws:sns:us-east-1:178579023202:terraform-cloudwatch-lb-alarms"]

# ELB 5xx 15m - Sum, above 15 for 2 consecutive period of 5 min - P1
evaluation_periods_high_elb_5XX_long        = "3"
period_high_elb_5XX_long                    = "300"
threshold_high_elb_5XX_long                 = "15"
actions_enabled_high_elb_5XX_long           = true
treat_missing_data_high_elb_5XX_long        = "notBreaching"
datapoints_to_alarm_high_elb_5XX_long       = "2"

# High ELB 4xx - Sum, above 20 for 2 consecutive period of 1 min - P2
configure_alarm_high_elb_4XX           = false
evaluation_periods_high_elb_4XX        = "2"
period_high_elb_4XX                    = "300"
threshold_high_elb_4XX                 = "20"
actions_enabled_high_elb_4XX           = true
treat_missing_data_high_elb_4XX        = "notBreaching"
datapoints_to_alarm_high_elb_4XX       = "2"
alarm_actions_high_elb_4XX             = ["arn:aws:sns:us-east-1:178579023202:terraform-cloudwatch-lb-alarms"]
ok_actions_high_elb_4XX                = ["arn:aws:sns:us-east-1:178579023202:terraform-cloudwatch-lb-alarms"]

#########  Healthy/UnHealthy Hosts Alarms #########

# High unhealthy hosts - Average, above 1 for 35 consecutive periods of 35 min - P1
evaluation_periods_high_unhealthy_hosts         = "35"
period_high_unhealthy_hosts                     = "60"
threshold_high_unhealthy_hosts                  = "1"
actions_enabled_high_unhealthy_hosts            = true
treat_missing_data_high_unhealthy_hosts         = "missing"
datapoints_to_alarm_high_unhealthy_hosts        = "35"

# Low healthy hosts - Average, below to 1 for 35 consecutive periods of 1 min - P1
evaluation_periods_low_healthy_hosts         = "35"
period_low_healthy_hosts                     = "60"
threshold_low_healthy_hosts                  = "1"
actions_enabled_low_healthy_hosts            = true
treat_missing_data_low_healthy_hosts         = "breaching"
datapoints_to_alarm_low_healthy_hosts        = "35"

# Critical Low healthy hosts - Average, below 1 for 10 consecutive periods of 1 min
evaluation_periods_critical_low_healthy_hosts         = "10"
period_critical_low_healthy_hosts                     = "60"
threshold_critical_low_healthy_hosts                  = "1"
actions_enabled_critical_low_healthy_hosts            = true
treat_missing_data_critical_low_healthy_hosts         = "breaching"
datapoints_to_alarm_critical_low_healthy_hosts        = "10"
alarm_actions_critical_low_healthy_hosts              = ["arn:aws:sns:us-east-1:178579023202:terraform-cloudwatch-lb-alarms"]
ok_actions_critical_low_healthy_hosts                 = ["arn:aws:sns:us-east-1:178579023202:terraform-cloudwatch-lb-alarms"]

# Fatal unhealthy hosts - Average, above 1 for 15 consecutive periods of 1 min - P1 Fatal
evaluation_periods_fatal_unhealthy_hosts         = "15"
period_fatal_unhealthy_hosts                     = "60"
threshold_fatal_unhealthy_hosts                  = "1"
actions_enabled_fatal_unhealthy_hosts            = true
treat_missing_data_fatal_unhealthy_hosts         = "missing"
datapoints_to_alarm_fatal_unhealthy_hosts        = "15"
alarm_actions_fatal_unhealthy_hosts              = ["arn:aws:sns:us-east-1:178579023202:terraform-cloudwatch-lb-alarms"]
ok_actions_fatal_unhealthy_hosts                 = ["arn:aws:sns:us-east-1:178579023202:terraform-cloudwatch-lb-alarms"]

# Backend Connection Errors - Sum, above 0 for 1 consecutive period of 1 min - P2
configure_alarm_high_backend_connection_errors             = false
evaluation_periods_high_backend_connection_errors         = "1"
period_high_backend_connection_errors                     = "60"
threshold_high_backend_connection_errors                  = "1"
actions_enabled_high_backend_connection_errors            = true
treat_missing_data_high_backend_connection_errors         = "notBreaching"
datapoints_to_alarm_high_backend_connection_errors        = "1"
alarm_actions_high_backend_connection_errors              = ["arn:aws:sns:us-east-1:178579023202:terraform-cloudwatch-lb-alarms"]
ok_actions_high_backend_connection_errors                 = ["arn:aws:sns:us-east-1:178579023202:terraform-cloudwatch-lb-alarms"]

#########  Latency Alarms #########
# Latency Anomaly - P2
configure_alarm_anomaly_latency             = false
evaluation_periods_anomaly_latency          = "2"
actions_enabled_anomaly_latency             = true
treat_missing_data_anomaly_latency          = "breaching"
period_anomaly_latency                      = "600"
band_anomaly_latency                        = "12"
alarm_actions_anomaly_latency               = ["arn:aws:sns:us-east-1:178579023202:terraform-cloudwatch-lb-alarms"]
ok_actions_anomaly_latency                  = ["arn:aws:sns:us-east-1:178579023202:terraform-cloudwatch-lb-alarms"]

# High/Warn Latency - Average, above 3s for 3 consecutive periods of 1 min - P2
evaluation_periods_high_latency         = "3"
period_high_latency                     = "60"
threshold_high_latency                  = "3"
actions_enabled_high_latency            = true
treat_missing_data_high_latency         = "notBreaching"
datapoints_to_alarm_high_latency        = "3"
alarm_actions_high_latency              = ["arn:aws:sns:us-east-1:178579023202:terraform-cloudwatch-lb-alarms"]
ok_actions_high_latency                 = ["arn:aws:sns:us-east-1:178579023202:terraform-cloudwatch-lb-alarms"]

# Critical High Latency - Average, above 5s for 5 consecutive periods of 1 min - P1
evaluation_periods_critical_latency         = "5"
period_critical_latency                     = "60"
threshold_critical_latency                  = "5"
actions_enabled_critical_latency            = true
treat_missing_data_critical_latency         = "notBreaching"
datapoints_to_alarm_critical_latency        = "5"

#########  Requests Count Alarms #########
# Requests Anomaly - P2
configure_alarm_anomaly_request_count             = false
evaluation_periods_anomaly_request_count          = "5"
actions_enabled_anomaly_request_count             = true
treat_missing_data_anomaly_request_count          = "notBreaching"
period_anomaly_request_count                      = "60"
band_anomaly_request_count                        = "3"
alarm_actions_anomaly_request_count              = ["arn:aws:sns:us-east-1:178579023202:terraform-cloudwatch-lb-alarms"]
ok_actions_anomaly_request_count                 = ["arn:aws:sns:us-east-1:178579023202:terraform-cloudwatch-lb-alarms"]

# High requests 1m - Sum, above 1000 for 3 consecutive period of 1 min - P2
evaluation_periods_high_request_count_short         = "3"
period_high_request_count_short                     = "60"
threshold_high_request_count_short                  = "1000"
actions_enabled_high_request_count_short            = true
treat_missing_data_high_request_count_short         = "breaching"
datapoints_to_alarm_high_request_count_short        = "3"
alarm_actions_high_request_count_short              = ["arn:aws:sns:us-east-1:178579023202:terraform-cloudwatch-lb-alarms"]
ok_actions_high_request_count_short                 = ["arn:aws:sns:us-east-1:178579023202:terraform-cloudwatch-lb-alarms"]

# High requests 15m - Sum, above 3,000 for 3 consecutive period of 5 min - P1
evaluation_periods_high_request_count_long         = "3"
period_high_request_count_long                     = "300"
threshold_high_request_count_long                  = "3000"
actions_enabled_high_request_count_long            = true
treat_missing_data_high_request_count_long         = "breaching"
datapoints_to_alarm_high_request_count_long        = "3"

# Low requests 1m - Sum, below 1 for 3 consecutive period of 1 min - P2
configure_alarm_low_request_count_short            = false
evaluation_periods_low_request_count_short         = "3"
period_low_request_count_short                     = "60"
threshold_low_request_count_short                  = "1"
actions_enabled_low_request_count_short            = true
treat_missing_data_low_request_count_short         = "breaching"
datapoints_to_alarm_low_request_count_short        = "3"
alarm_actions_low_request_count_short              = ["arn:aws:sns:us-east-1:178579023202:terraform-cloudwatch-lb-alarms"]
ok_actions_low_request_count_short                 = ["arn:aws:sns:us-east-1:178579023202:terraform-cloudwatch-lb-alarms"]

# Low requests 15m - Sum, below 10 for 2 consecutive period of 5 min - P1
configure_alarm_low_request_count_long            = false
evaluation_periods_low_request_count_long         = "3"
period_low_request_count_long                     = "300"
threshold_low_request_count_long                  = "10"
actions_enabled_low_request_count_long            = true
treat_missing_data_low_request_count_long         = "breaching"
datapoints_to_alarm_low_request_count_long        = "2"

#########  SurgeQueue/Spillover Alarms #########

# Surge Queue Anomaly - P2
configure_alarm_anomaly_surge_queue             = false
evaluation_periods_anomaly_surge_queue          = "5"
actions_enabled_anomaly_surge_queue             = true
treat_missing_data_anomaly_surge_queue          = "notBreaching"
period_anomaly_surge_queue                      = "60"
band_anomaly_surge_queue                        = "4"
alarm_actions_anomaly_surge_queue              = ["arn:aws:sns:us-east-1:178579023202:terraform-cloudwatch-lb-alarms"]
ok_actions_anomaly_surge_queue                 = ["arn:aws:sns:us-east-1:178579023202:terraform-cloudwatch-lb-alarms"]

# Surge Queue - Max, above 10 for 3 consecutive period of 1 min - P2
configure_alarm_high_surge_queue            = false
evaluation_periods_high_surge_queue         = "3"
period_high_surge_queue                     = "60"
threshold_high_surge_queue                  = "11"
actions_enabled_high_surge_queue            = true
treat_missing_data_high_surge_queue         = "notBreaching"
datapoints_to_alarm_high_surge_queue        = "3"
alarm_actions_high_surge_queue              = ["arn:aws:sns:us-east-1:178579023202:terraform-cloudwatch-lb-alarms"]
ok_actions_high_surge_queue                 = ["arn:aws:sns:us-east-1:178579023202:terraform-cloudwatch-lb-alarms"]

# Spillover Count Anomaly - P2
configure_alarm_anomaly_spillover_count             = false
evaluation_periods_anomaly_spillover_count          = "5"
actions_enabled_anomaly_spillover_count             = true
treat_missing_data_anomaly_spillover_count          = "notBreaching"
period_anomaly_spillover_count                      = "60"
band_anomaly_spillover_count                        = "4"
alarm_actions_anomaly_spillover_count              = ["arn:aws:sns:us-east-1:178579023202:terraform-cloudwatch-lb-alarms"]
ok_actions_anomaly_spillover_count                 = ["arn:aws:sns:us-east-1:178579023202:terraform-cloudwatch-lb-alarms"]

# Spillover Count - Sum, above 0 for 1 consecutive period of 1 min - P2
configure_alarm_high_spillover_count            = false
evaluation_periods_high_spillover_count         = "3"
period_high_spillover_count                     = "60"
threshold_high_spillover_count                  = "1"
actions_enabled_high_spillover_count            = true
treat_missing_data_high_spillover_count         = "notBreaching"
datapoints_to_alarm_high_spillover_count        = "3"
alarm_actions_high_spillover_count              = ["arn:aws:sns:us-east-1:178579023202:terraform-cloudwatch-lb-alarms"]
ok_actions_high_spillover_count                 = ["arn:aws:sns:us-east-1:178579023202:terraform-cloudwatch-lb-alarms"]

#########  Estimated-xxx Alarms #########
# Estimated Active Connection Count - Anomaly - P2
configure_alarm_anomaly_estimated_active_connection_count             = false
evaluation_periods_anomaly_estimated_active_connection_count          = "5"
actions_enabled_anomaly_estimated_active_connection_count             = true
treat_missing_data_anomaly_estimated_active_connection_count          = "missing"
period_anomaly_estimated_active_connection_count                      = "60"
band_anomaly_estimated_active_connection_count                        = "5"
alarm_actions_anomaly_estimated_active_connection_count               = ["arn:aws:sns:us-east-1:178579023202:terraform-cloudwatch-lb-alarms"]
ok_actions_anomaly_estimated_active_connection_count                  = ["arn:aws:sns:us-east-1:178579023202:terraform-cloudwatch-lb-alarms"]

# Estimated Consumed LCU - Anomaly - P2
configure_alarm_anomaly_estimated_consumed_lcu             = false
evaluation_periods_anomaly_estimated_consumed_lcu          = "5"
actions_enabled_anomaly_estimated_consumed_lcu             = true
treat_missing_data_anomaly_estimated_consumed_lcu          = "missing"
period_anomaly_estimated_consumed_lcu                      = "60"
band_anomaly_estimated_consumed_lcu                        = "3"
alarm_actions_anomaly_estimated_consumed_lcu               = ["arn:aws:sns:us-east-1:178579023202:terraform-cloudwatch-lb-alarms"]
ok_actions_anomaly_estimated_consumed_lcu                  = ["arn:aws:sns:us-east-1:178579023202:terraform-cloudwatch-lb-alarms"]

# Estimated Active New Connection Count - Anomaly - P2
configure_alarm_anomaly_estimated_new_connection_count             = false
evaluation_periods_anomaly_estimated_new_connection_count          = "5"
actions_enabled_anomaly_estimated_new_connection_count             = true
treat_missing_data_anomaly_estimated_new_connection_count          = "missing"
period_anomaly_estimated_new_connection_count                      = "60"
band_anomaly_estimated_new_connection_count                        = "5"
alarm_actions_anomaly_estimated_new_connection_count               = ["arn:aws:sns:us-east-1:178579023202:terraform-cloudwatch-lb-alarms"]
ok_actions_anomaly_estimated_new_connection_count                  = ["arn:aws:sns:us-east-1:178579023202:terraform-cloudwatch-lb-alarms"]

# Estimated Processed Bytes - Anomaly - P2
configure_alarm_anomaly_estimated_processed_bytes             = false
evaluation_periods_anomaly_estimated_processed_bytes          = "5"
actions_enabled_anomaly_estimated_processed_bytes             = true
treat_missing_data_anomaly_estimated_processed_bytes          = "missing"
period_anomaly_estimated_processed_bytes                      = "60"
band_anomaly_estimated_processed_bytes                        = "3"
alarm_actions_anomaly_estimated_processed_bytes               = ["arn:aws:sns:us-east-1:178579023202:terraform-cloudwatch-lb-alarms"]
ok_actions_anomaly_estimated_processed_bytes                  = ["arn:aws:sns:us-east-1:178579023202:terraform-cloudwatch-lb-alarms"]

#########  Custom Alarms #########