Export Prometheus metrics from your web server's access log

Web servers only offer limited metrics. For example, nginx only offers stub_status unless you pay for NGINX Plus, which will only give a basic count of active connections and requests served across all virtual hosts.

However nearly all web servers are configured to log all requests to a file.

This tool attaches to the log file (like tail -f does) and exposes metrics for Prometheus.

Project status

This is ready for production use, and deployed to my personal server which hosts multiple services.

How to use

$ ./access-log-to-prometheus-metrics /var/log/nginx/access.log '$host $remote_addr - $remote_user [$time_local] "$request" $status $body_bytes_sent "$http_referer" "$http_user_agent"'
Starting server at 127.0.0.1:9898
$ curl http://127.0.0.1:9898/
# HELP requests The total number of requests per HTTP status code and virtual host name
# TYPE requests counter
requests{status="200",vhost="app.taguette.org"} 123
requests{status="304",vhost="app.taguette.org"} 27
requests{status="207",vhost="cloud.rampin.org"} 8

Supported log formats

You can pass your NGINX log_format configuration line. If you are using a different web server, you will have to convert your log format to the equivalent NGINX configuration.

The following variables are read:

• The HTTP status code ($status), used as a label on the requests metric
• The virtual host name ($host), used as a label on the requests metric
• The presence of a user name, used as label on the requests metric (yes/no/unk)
• The time to process the request ($request_time), used in the request_duration metric
• The size of the response's body ($response_body_size), used in the response_body_size metric

Metrics

• requests is a counter for requests (e.g. log lines), organized by status (the HTTP status code) and vhost (the virtual host name)
• request_duration is a histogram for request durations, organized by status (the HTTP status code) and vhost (the virtual host name)
• response_body_size if a histogram for responses' body sizes in bytes, organized by status (the HTTP status code) and vhost (the virtual host name)

Labels default to the string "unk" if they are unknown (e.g. those variables are not present in your log format).

Custom rules

If the default fields extracted from the log are not sufficient, you can specify custom filters and labels based on any field.

The syntax for filters is:

-m <field>:<regex>

If the regex does not match the value of the field, that log entry will be dropped.

Examples:

# Only process errors
-m 'status:^50[0-9]$'
# Only measure requests to API
-m 'request:^[A-Z]+ /api/'
# Ignore robots.txt
-m 'request:^GET (?!robots\.txt) '

The syntax for labels is:

-l <label-name>:<value>:<field>:<regex>

A label <label-name> will be added to all metrics with the value extracted from the field.

Examples:

# Add label for API version
-l 'api_version:$1:request:^[A-Z]+ /api/(v[0-9]+)/'
# Add label for client version
-l 'client_version:app-$1,go-$2:referer:MyApp/([0-9.]+) Go/([0-9.]+)'
# Add label for HTTP version
-l 'http_version:HTTP/$1:request:^[A-Z]+ /[^ ]* HTTP/([0-9]+)$'