Merge pull request #1 from skyscrapers/init

Add Dynamodb, Kinesis and Lambda modules
skyscrapers · Mar 20, 2019 · 368d0dd · 368d0dd
2 parents 26b4ff1 + b2bec14
commit 368d0dd
Show file tree

Hide file tree

Showing 7 changed files with 308 additions and 0 deletions.
diff --git a/README.md b/README.md
@@ -0,0 +1,61 @@
+# terraform-cloudwatch
+
+Terraform module to setup cloudwatch alerts and push them to SNS. This repository contains the following modules:
+
+* `kinesis`: Creates alerts for a kinesis stream. This is used in the `skyscrapers/terraform-kinesis` module.
+* `dynamodb`: Creates the alerts needed for a dynamodb table.
+* `lambda_function`: Creates the alerts for lambda functions.
+
+## kinesis
+Creates alerts for a kinesis stream. This is used in the `skyscrapers/terraform-kinesis` module.
+
+The following resources are created:
+
+* Cloudwatch alerts for the kinesis stream that was passed as variable
+
+### Available variables
+
+| Name | Description | Type | Default | Required |
+|------|-------------|:----:|:-----:|:-----:|
+| kinesis\_iterator\_age\_error\_evaluation\_periods | The number of periods over which data is compared to the specified threshold. | string | `"1"` | no |
+| kinesis\_iterator\_age\_error\_period | The period in seconds over which the specified stat is applied. | string | `"300"` | no |
+| kinesis\_iterator\_age\_error\_threshold | The value against which the specified statistic is compared. | string | `"1000000"` | no |
+| kinesis\_stream\_name | Name of the kinesis stream to monitor | string | n/a | yes |
+| kinesis\_write\_throughput\_exceeded\_evaluation\_periods | The number of periods over which data is compared to the specified threshold. | string | `"6"` | no |
+| kinesis\_write\_throughput\_exceeded\_period | The period in seconds over which the specified stat is applied. | string | `"300"` | no |
+| kinesis\_write\_throughput\_exceeded\_threshold | The value against which the specified statistic is compared. | string | `"10"` | no |
+| sns\_topic\_arn | ARN of the SNS topic you want the alerts to be sent to | string | n/a | yes |
+
+## dynamodb
+
+Creates the alerts needed for a dynamodb table.
+
+### Available variables
+
+| Name | Description | Type | Default | Required |
+|------|-------------|:----:|:-----:|:-----:|
+| dynamodb\_table\_name | Name of the dynamodb table to monitor | string | n/a | yes |
+| dynamodb\_throttle\_evaluation\_periods | The period in seconds over which the specified stat is applied. | string | `"1"` | no |
+| dynamodb\_throttle\_period | The number of periods over which data is compared to the specified threshold. | string | `"60"` | no |
+| dynamodb\_throttle\_threshold | The value against which the specified statistic is compared. | string | `"0"` | no |
+| sns\_topic\_arn | ARN of the SNS topic you want the alerts to be sent to | string | n/a | yes |
+
+## lambda_function
+
+ Creates the alerts for lambda functions.
+
+### Available variables
+
+| Name | Description | Type | Default | Required |
+|------|-------------|:----:|:-----:|:-----:|
+| lambda\_function | Name of the lambda function to monitor | string | n/a | yes |
+| lambda\_invocation\_error\_evaluation\_periods | The number of periods over which data is compared to the specified threshold. | string | `"1"` | no |
+| lambda\_invocation\_error\_period | The period in seconds over which the specified stat is applied. | string | `"60"` | no |
+| lambda\_invocation\_error\_threshold | The value against which the specified statistic is compared. | string | `"5"` | no |
+| lambda\_iterator\_age\_error\_evaluation\_periods | The number of periods over which data is compared to the specified threshold. | string | `"1"` | no |
+| lambda\_iterator\_age\_error\_period | The period in seconds over which the specified stat is applied. | string | `"60"` | no |
+| lambda\_iterator\_age\_error\_threshold | The value against which the specified statistic is compared. | string | `"1000000"` | no |
+| lambda\_throttle\_error\_evaluation\_periods | The number of periods over which data is compared to the specified threshold. | string | `"1"` | no |
+| lambda\_throttle\_error\_period | The period in seconds over which the specified stat is applied. | string | `"60"` | no |
+| lambda\_throttle\_error\_threshold | The value against which the specified statistic is compared. | string | `"0"` | no |
+| sns\_topic\_arn | ARN of the SNS topic you want the alerts to be sent to | string | n/a | yes |
diff --git a/dynamodb/main.tf b/dynamodb/main.tf
@@ -0,0 +1,19 @@
+// Dynamo requests are being throttled.
+resource "aws_cloudwatch_metric_alarm" "dynamo_table_throttles" {
+  alarm_name          = "${var.dynamodb_table_name}_throttles"
+  alarm_description   = "Requests are being throttled to the ${var.dynamodb_table_name} table"
+  namespace           = "AWS/DynamoDB"
+  metric_name         = "ThrottledRequests"
+  statistic           = "Sum"
+  comparison_operator = "GreaterThanThreshold"
+  threshold           = "${var.dynamodb_throttle_threshold}"
+  period              = "${var.dynamodb_throttle_period}"
+  evaluation_periods  = "${var.dynamodb_throttle_evaluation_periods}"
+
+  alarm_actions = ["${var.sns_topic_arn}"]
+  ok_actions    = ["${var.sns_topic_arn}"]
+
+  dimensions = {
+    TableName = "${var.dynamodb_table_name}"
+  }
+}
diff --git a/dynamodb/variables.tf b/dynamodb/variables.tf
@@ -0,0 +1,25 @@
+variable "sns_topic_arn" {
+  description = "ARN of the SNS topic you want the alerts to be sent to"
+}
+
+// Dynamodb table name
+variable "dynamodb_table_name" {
+  type        = "string"
+  description = "Name of the dynamodb table to monitor"
+}
+
+//Alarm settings
+variable "dynamodb_throttle_threshold" {
+  default     = 0
+  description = "The value against which the specified statistic is compared."
+}
+
+variable "dynamodb_throttle_period" {
+  default     = 60
+  description = "The number of periods over which data is compared to the specified threshold."
+}
+
+variable "dynamodb_throttle_evaluation_periods" {
+  default     = 1
+  description = "The period in seconds over which the specified stat is applied."
+}
diff --git a/kinesis/main.tf b/kinesis/main.tf
@@ -0,0 +1,43 @@
+// Kinesis: Iterator Age
+resource "aws_cloudwatch_metric_alarm" "_kinesis_iterator_age" {
+  alarm_name          = "${var.kinesis_stream_name}_high_iterator_age"
+  alarm_description   = "The Get iterator age of ${var.kinesis_stream_name} is starting to lag behind"
+  namespace           = "AWS/Kinesis"
+  metric_name         = "GetRecords.IteratorAgeMilliseconds"
+  statistic           = "Maximum"
+  comparison_operator = "GreaterThanThreshold"
+  threshold           = "${var.kinesis_iterator_age_error_threshold}"
+  evaluation_periods  = "${var.kinesis_iterator_age_error_evaluation_periods}"
+  period              = "${var.kinesis_iterator_age_error_period}"
+  alarm_description   = " Kinesis High Iterator Age: ${var.kinesis_stream_name}"
+
+  alarm_actions             = ["${var.sns_topic_arn}"]
+  ok_actions                = ["${var.sns_topic_arn}"]
+  insufficient_data_actions = ["${var.sns_topic_arn}"]
+
+  dimensions {
+    StreamName = "${var.kinesis_stream_name}"
+  }
+}
+
+// Kinesis: Write Throughput Exceeded
+resource "aws_cloudwatch_metric_alarm" "_kinesis_write_exceeded" {
+  alarm_name          = "${var.kinesis_stream_name}_write_exceeded"
+  alarm_description   = "There are more writes going into ${var.kinesis_stream_name} than the chards can handle"
+  namespace           = "AWS/Kinesis"
+  metric_name         = "WriteProvisionedThroughputExceeded"
+  statistic           = "Sum"
+  comparison_operator = "GreaterThanThreshold"
+  threshold           = "${var.kinesis_write_throughput_exceeded_threshold}"
+  evaluation_periods  = "${var.kinesis_write_throughput_exceeded_evaluation_periods}"
+  period              = "${var.kinesis_write_throughput_exceeded_period}"
+  alarm_description   = " Kinesis Write Throughput Exceeded: ${var.kinesis_stream_name}"
+
+  alarm_actions             = ["${var.sns_topic_arn}"]
+  ok_actions                = ["${var.sns_topic_arn}"]
+  insufficient_data_actions = ["${var.sns_topic_arn}"]
+
+  dimensions {
+    StreamName = "${var.kinesis_stream_name}"
+  }
+}
diff --git a/kinesis/variables.tf b/kinesis/variables.tf
@@ -0,0 +1,41 @@
+variable "sns_topic_arn" {
+  description = "ARN of the SNS topic you want the alerts to be sent to"
+}
+
+// Kinesis stream name
+variable "kinesis_stream_name" {
+  description = "Name of the kinesis stream to monitor"
+  type        = "string"
+}
+
+// Kinesis Iterator Age Alarm Settings
+variable "kinesis_iterator_age_error_threshold" {
+  description = "The value against which the specified statistic is compared."
+  default     = "1000000"
+}
+
+variable "kinesis_iterator_age_error_evaluation_periods" {
+  description = "The number of periods over which data is compared to the specified threshold."
+  default     = "1"
+}
+
+variable "kinesis_iterator_age_error_period" {
+  description = "The period in seconds over which the specified stat is applied."
+  default     = "300"
+}
+
+// Kinesis Write Throughput Alarm Settings
+variable "kinesis_write_throughput_exceeded_threshold" {
+  description = "The value against which the specified statistic is compared."
+  default     = "10"
+}
+
+variable "kinesis_write_throughput_exceeded_evaluation_periods" {
+  description = "The number of periods over which data is compared to the specified threshold."
+  default     = "6"
+}
+
+variable "kinesis_write_throughput_exceeded_period" {
+  description = "The period in seconds over which the specified stat is applied."
+  default     = "300"
+}
diff --git a/lambda_function/main.tf b/lambda_function/main.tf
@@ -0,0 +1,61 @@
+// Setup Cloudwatch alarms for all Lambda function aliases (environments)
+// Send alerts to given SNS topics.
+
+// Lambda: Invocation Errors
+resource "aws_cloudwatch_metric_alarm" "streamalert_lambda_invocation_errors" {
+  alarm_name          = "${var.lambda_function}_invocation_errors"
+  alarm_description   = "The errors on ${var.lambda_function} are higher than ${var.lambda_invocation_error_threshold} for ${var.lambda_invocation_error_period}"
+  namespace           = "AWS/Lambda"
+  metric_name         = "Errors"
+  statistic           = "Sum"
+  comparison_operator = "GreaterThanThreshold"
+  threshold           = "${var.lambda_invocation_error_threshold}"
+  evaluation_periods  = "${var.lambda_invocation_error_evaluation_periods}"
+  period              = "${var.lambda_invocation_error_period}"
+
+  alarm_actions = ["${var.sns_topic_arn}"]
+
+  dimensions {
+    FunctionName = "${var.lambda_function}"
+  }
+}
+
+// Lambda: Throttles
+resource "aws_cloudwatch_metric_alarm" "streamalert_lambda_throttles" {
+  alarm_name          = "${var.lambda_function}_throttles"
+  alarm_description   = "Lambda function ${var.lambda_function} is being throttled"
+  namespace           = "AWS/Lambda"
+  metric_name         = "Throttles"
+  statistic           = "Sum"
+  comparison_operator = "GreaterThanThreshold"
+  threshold           = "${var.lambda_throttle_error_threshold}"
+  evaluation_periods  = "${var.lambda_throttle_error_evaluation_periods}"
+  period              = "${var.lambda_throttle_error_period}"
+
+  alarm_actions = ["${var.sns_topic_arn}"]
+  ok_actions    = ["${var.sns_topic_arn}"]
+
+  dimensions {
+    FunctionName = "${var.lambda_function}"
+  }
+}
+
+// Lambda: IteratorAge
+resource "aws_cloudwatch_metric_alarm" "streamalert_lambda_iterator_age" {
+  alarm_name          = "${var.lambda_function}_iterator_age"
+  alarm_description   = "Lambda High Iterator Age for ${var.lambda_function}"
+  namespace           = "AWS/Lambda"
+  metric_name         = "IteratorAge"
+  statistic           = "Maximum"
+  comparison_operator = "GreaterThanThreshold"
+  threshold           = "${var.lambda_iterator_age_error_threshold}"
+  evaluation_periods  = "${var.lambda_iterator_age_error_evaluation_periods}"
+  period              = "${var.lambda_iterator_age_error_period}"
+
+  alarm_actions = ["${var.sns_topic_arn}"]
+  ok_actions    = ["${var.sns_topic_arn}"]
+
+  dimensions {
+    FunctionName = "${var.lambda_function}"
+  }
+}
diff --git a/lambda_function/variables.tf b/lambda_function/variables.tf
@@ -0,0 +1,58 @@
+variable "sns_topic_arn" {
+  description = "ARN of the SNS topic you want the alerts to be sent to"
+}
+
+// Lambda Function name
+variable "lambda_function" {
+  type        = "string"
+  description = "Name of the lambda function to monitor"
+}
+
+// Lambda Invocation Error Alarm Settings
+
+variable "lambda_invocation_error_threshold" {
+  default     = "5"
+  description = "The value against which the specified statistic is compared."
+}
+
+variable "lambda_invocation_error_evaluation_periods" {
+  default     = "1"
+  description = "The number of periods over which data is compared to the specified threshold."
+}
+
+variable "lambda_invocation_error_period" {
+  default     = "60"
+  description = "The period in seconds over which the specified stat is applied."
+}
+
+// Lambda Throttling Alarm Settings
+variable "lambda_throttle_error_threshold" {
+  default     = "0"
+  description = "The value against which the specified statistic is compared."
+}
+
+variable "lambda_throttle_error_evaluation_periods" {
+  default     = "1"
+  description = "The number of periods over which data is compared to the specified threshold."
+}
+
+variable "lambda_throttle_error_period" {
+  default     = "60"
+  description = "The period in seconds over which the specified stat is applied."
+}
+
+// Lambda Iterator Age Alarm Settings
+variable "lambda_iterator_age_error_threshold" {
+  default     = "1000000"
+  description = "The value against which the specified statistic is compared."
+}
+
+variable "lambda_iterator_age_error_evaluation_periods" {
+  default     = "1"
+  description = "The number of periods over which data is compared to the specified threshold."
+}
+
+variable "lambda_iterator_age_error_period" {
+  default     = "60"
+  description = "The period in seconds over which the specified stat is applied."
+}