1. Prometheus Client Libraries
Before Prometheus can monitor your application, you need to add instrumentation code using client libraries. These libraries support different metric types that Prometheus understands.
Following table summarizes the official Prometheus libraries.
|
Language |
Library Name |
Reference Link |
|
Go |
client_golang |
https://github.com/prometheus/client_golang |
|
Java / Scala |
client_java |
https://github.com/prometheus/client_java
|
|
Python |
client_python |
https://github.com/prometheus/client_python
|
|
Ruby |
client_ruby |
https://github.com/prometheus/client_ruby |
|
Rust |
client_rust |
https://github.com/prometheus/client_rust |
Here's the table for the Unofficial (Third-party) Prometheus Client Libraries:
|
Language |
Library Name |
Reference Link |
|
Bash |
client_bash |
https://github.com/aecolley/client_bash |
|
C |
prometheus-client-c |
https://github.com/digitalocean/prometheus-client-c |
|
C++ |
prometheus-cpp |
https://github.com/jupp0r/prometheus-cpp
|
|
Common Lisp |
prometheus.cl |
https://github.com/deadtrickster/prometheus.cl |
|
Dart |
prometheus_client |
https://github.com/tentaclelabs/prometheus_client
|
|
Delphi |
prometheus-client-delphi |
https://github.com/marcobreveglieri/prometheus-client-delphi
|
|
Elixir |
prometheus.ex |
https://github.com/deadtrickster/prometheus.ex |
|
Erlang |
prometheus.erl |
https://github.com/deadtrickster/prometheus.erl |
|
Haskell |
prometheus-haskell |
https://github.com/fimad/prometheus-haskell |
|
Julia |
Prometheus.jl |
https://github.com/fredrikekre/Prometheus.jl |
|
Lua (Nginx) |
nginx-lua-prometheus |
https://github.com/knyar/nginx-lua-prometheus |
|
Lua (Tarantool) |
metrics |
https://github.com/tarantool/metrics
|
|
.NET / C# |
prometheus-net |
https://github.com/prometheus-net/prometheus-net |
|
Node.js |
prom-client |
https://github.com/siimon/prom-client |
|
OCaml |
prometheus |
https://github.com/mirage/prometheus |
|
Perl |
Net::Prometheus |
https://metacpan.org/pod/Net::Prometheus |
|
PHP |
prometheus_client_php |
https://github.com/promphp/prometheus_client_php |
|
R |
pRometheus |
https://github.com/cfmack/pRometheus |
2. Prometheus Metric Types Exposed by these libraries
Before Prometheus can collect data from your applications, you need to instrument your code using one of its client libraries (based on your programming language). These libraries expose four primary metric types, each suited to a specific kind of measurement.
2.1 Counter
A Counter is a metric that only increases. Counters start at 0 and can only go up (or reset to 0 when the application restarts). Use counters to track things that only increase, never decrease.
Don't use counters for things that can decrease (like current memory or thread count).
Common Methods:
· inc(): Increments by 1
· inc(amount): Increments by a specific value
Examples:
· Number of HTTP requests received
· Number of errors occurred
· Total tasks processed
· Number of files uploaded
Sample Java Code
Counter requestCounter = Counter.build() .name("http_requests_total") .help("Total HTTP requests received") .register(); requestCounter.inc(); // Increment by 1 requestCounter.inc(5); // Increment by 5
2.2 Gauge
A Gauge is used for values that can go up and down. Use gauges to track things that fluctuate.
Common Methods:
· set(value): Set to a specific value
· inc(): Increment by 1
· dec(): Decrement by 1
Examples:
· Current number of active threads
· Memory usage
· Temperature readings
· Current queue size
· CPU usage
Sample Java Code
Gauge memoryUsage = Gauge.build() .name("current_memory_usage_bytes") .help("Tracks the current memory usage") .register(); memoryUsage.set(1024); // Set to specific value memoryUsage.inc(); // Increment by 1 memoryUsage.dec(); // Decrement by 1
2.3 Summary
A Summary is used to track the distribution of events over time, especially useful for measuring things like request durations or payload sizes. It provides total sum, count, and percentiles.
Primary Method
observe(value): Records the size/duration of the event
Examples:
· Response time of a web request
· Size of incoming requests
· Latency of database operations
Sample Java Code
Summary requestLatency = Summary.build() .name("request_latency_seconds") .help("Tracks request durations in seconds") .register(); requestLatency.observe(0.35); // Observe a request that took 350ms
2.4 Histogram
A Histogram tracks the distribution of events, but instead of percentiles, it places observations into predefined buckets. You can count how many requests fell into which time ranges.
Primary Method:
observe(value): Records an event’s value into a bucket
Examples:
· Request duration distribution (e.g., how many requests took <100ms, <500ms, etc.)
· File sizes
· DB query execution time
Key Difference from Summary:
· Histogram buckets are configurable
· Summary gives quantiles (client-side), Histogram gives buckets (server-side)
Sample Java Code
Histogram requestDuration = Histogram.build() .name("http_request_duration_seconds") .help("HTTP request latency buckets") .buckets(0.1, 0.3, 0.5, 1, 2, 5) // Custom buckets in seconds .register(); requestDuration.observe(0.25); // 250ms requestDuration.observe(0.42); // 420ms requestDuration.observe(1.3); // 1300ms requestDuration.observe(0.05); // 50ms requestDuration.observe(2.1); // 2100ms
Each call to .observe() adds a data point to:
· A counter that tracks the total number of observations (count)
· A sum that accumulates total observed value
· The appropriate bucket (like < 0.3, < 1, etc.)
In summary,
· Use Counter for totals like “how many files processed”.
· Use Gauge for real-time metrics like “how many users are online now”.
· Use Summary for approximate quantiles like “95% of responses completed in under X ms”.
· Use Histogram when you want bucketed distribution data (ideal for alerts and visualizations).
Previous Next Home
No comments:
Post a Comment