Amazon Managed Service for Prometheus pricing

In this example you are monitoring 1 Kubernetes cluster with 10 nodes with 1000 Prometheus metrics per node collected every 30 seconds for the entire month (744 hours). This example does not factor in your AWS Free Tier benefits. The following usage and costs will be calculated for 1) metric sample ingestion, 2) storage, and 3) query.

Metric sample ingestion usage and costs
First, we will calculate the number of metric samples ingested and the costs for these samples. The monthly number of samples ingested is 892.8 million samples (10 nodes * 1000 metrics per node / 30 collection interval in seconds * 3,600 seconds in an hour * 744 hours in a month = 10 * 1000/30 * 3,600 * 744 = 892.8 million samples).

The monthly metrics ingested costs is $80.93 ($0.90/10,000,000 for the first 2 billion metric samples * 892.8 million samples).

Storage usage and costs
Second, we will calculate the storage in GB and the costs for the metric samples and metric metadata stored. For this example, we will assume that you have 20 labels per metrics with 100 unique label values per label. The metadata including the metric name, labels, and label values are stored daily. We will also assume an average of 30 bytes per label and label value and 2 bytes per metric sample ingested.

The monthly GB of storage is 0.25 GB (metric metadata * # of days per month) + (2 bytes * number of metric samples = (1000 metrics * 20 labels * 100 label values * 30 bytes * ~30 days per month) + (2 bytes * 892.8 million samples) = 1800 million bytes + 1,785.6 million bytes = 3,585.6 million bytes = 3.34 GB).

The storage charge is $0.10 ($0.03/GB * 3.34 GB).

Query usage and costs
We will now calculate the query samples processed and query costs for this example. We will assume you have 1 end user monitoring a dashboard for an average of 2 hours per day refreshing it every 60 seconds with 20 chart widgets per dashboard (assuming 1 PromQL query per widget). We will assume for this example that the samples processed per query is 100,000.

First, start by calculating the count of queries per month which is 72,000 queries (1 end user * 20 charts * 2 hours of viewing a dashboard * 3,600 seconds per hour / 60 seconds per dashboard refresh * ~30 days in a month).

Second, we calculate the query samples processed per month which is 7.2 billion (72,000 queries * 100,000/query)

We can now calculate the query charge of $0.72 ($0.10/billion samples processed * 7.2 billion samples processed).

We can now calculate your monthly costs for this example:

• Metric ingestion costs: $80.93 ($0.90/10M * 892.8 million samples)
  
• Storage costs: $0.10 ($0.03/GB * 3.34 GB)
• Query costs: $0.72 ($0.10/billion samples processed * 7.2 billion samples processed)

Total Monthly charges = $80.93 (metric ingestion) + $0.10 (storage) + $0.72 (query) = $81.75

In this example you are monitoring 1 Kubernetes cluster with 10 nodes with 1000 Prometheus metrics per node collected every 30 seconds by Amazon Managed Service for Prometheus collector for the entire month (744 hours). The following usage and costs will be calculated 1) metric sample ingestion, 2) storage, and 3) Amazon Managed Service for Prometheus collector samples collected and collector hours. 

Metric sample ingestion usage and costs
First, we will calculate the number of metric samples ingested and the costs for these samples. The monthly number of samples ingested is 892.8 million samples (10 nodes * 1000 metrics per node / 30 collection interval in seconds * 3,600 seconds in an hour * 744 hours in a month = 10 * 100/60 * 3,600 * 744 = 892.8 million samples).

The monthly metrics ingested costs is $80.93 ($0.90/10,000,000 for the first 2 billion metric samples * 892.8 million samples).

Storage usage and costs
Second, we will calculate the storage in GB and the costs for the metric samples and metric metadata stored. For this example, we will assume that you have 20 labels per metrics with 100 unique label values per label. The metadata including the metric name, labels, and label values are stored daily. We will also assume an average of 30 bytes per label and label value and 2 bytes per metric sample ingested.

The monthly GB of storage is 0.25 GB (metric metadata * # of days per month) + (2 bytes * number of metric samples = (1000 metrics * 20 labels * 100 label values * 30 bytes * ~30 days per month) + (2 bytes * 892.8 million samples) = 1800 million bytes + 1,785.6 million bytes = 3,585.6 million bytes = 3.34 GB).

The storage charge is $0.10 ($0.03/GB * 3.34 GB).

Samples collected and collector hour costs
Third, we will calculate the collector hour costs. The agentless collector costs $0.04 per collector hour and $0.03 per 10M samples collected, and we assume a single collector for a single EKS cluster. Assuming 744 hours in a month, the monthly cost is: 1 * $0.04 * 744 + $0.03 * 892.8 million samples / 10M = $32.44.

The agentless collector charge is $32.44. 

We can now calculate your monthly costs for this example:

• Metric ingestion costs: $80.93 ($0.90/10M * 892.8 million samples)
  
• Storage costs: $0.10 ($0.03/GB * 3.34 GB)
• The agentless collector costs: $32.44 ($0.04 per collector-hour * 744 hours + $0.03 * 892.8 million samples / 10M)
  

Total costs: $113.47

In this example, you are running a query to calculate the CPU of your 10,000 nodes in the last hour and an alerting rule running every minute to notify you when one of those nodes exceeds 80% CPU capacity. This example does not factor in your AWS Free Tier benefits. We will calculate the usage and costs of running the query and the alerting rule.

Query usage and cost
First, we calculate the number of samples a single query processes by identifying the number of metric time series and the provided time range. A metric time series is identified by its metric name and key-value pairs. In this example, each node has a single CPU metric resulting in a query that processes 10,000 metric time series for a 1 hour time range. Prometheus queries process one sample per resolution step, which is the minimum amount of time between any two data points. For example, we’ll assume a step of 1 minute, making it 1 sample per minute in the time range. This is 10,000 time series x 1 hour x 1 sample/min x 60 min/hour = 600,000 samples/query.

Assuming you ran the query 1,000 times, the cost would be $0.06 ($0.10/billion samples processed * 0.0006 billion samples * 1,000 queries).

Alerting rule usage and cost
We calculate the number of samples the rule processes per evaluation by identifying the number of selected metric time series. Since the rule identifies one CPU metric time series per node, it processes 10,000 time series per evaluation. Alerting rules in Prometheus run as instant queries, which are queries that are evaluated at a single point in time. Therefore, this alerting rule evaluates 10,000 time series x 1 sample/time series = 10,000 samples. Since this alarm runs every minute, this alerting rule processes 10,000 samples x 60 min/hour x 730 hours/month = 438,000,000 samples/month.

The cost of this alerting rule is $0.04 per month ($0.10/billion samples processed * 0.438 billion samples).