指标与监控集成
实现指标收集，并与 Prometheus 和 Grafana 等外部监控系统集成，以提升系统可观测性。
指标与监控集成 是 CoddyKit 上的免费 Erlang OTP: Distributed & Fault-Tolerant Systems Programming 课时。 这是第 3 节课，共 4 节。 你可以在下方免费阅读本课时的完整内容 — 然后在浏览器中使用内置代码编辑器和全天候 AI 导师进行实践。 这是 Erlang OTP: Distributed & Fault-Tolerant Systems Programming 学习路径的一部分，你的进度在网页和 CoddyKit 应用中同步。 Erlang OTP: Distributed & Fault-Tolerant Systems Programming 课程共包含 4 节课。
本课时的部分内容尚未翻译，以英文显示。
Observability with Metrics
Welcome to Metrics & Monitoring Integration! In modern systems, understanding what's happening inside your application is crucial. This is called observability.
Observability relies on three pillars:
- Logs: Records of discrete events.
- Traces: End-to-end request flows across services.
- Metrics: Aggregated numerical data points over time.
This lesson focuses on metrics, which give you a high-level view of system health and performance trends.
Why Metrics Matter
Metrics are numerical measurements collected at regular intervals. They help you answer questions like:
- How many requests per second is my service handling?
- What's the average response time?
- How much memory is my Erlang VM using?
- Are there any errors occurring?
By tracking these over time, you can spot trends, identify bottlenecks, and react to issues proactively.
Types of Metrics
There are several common types of metrics:
- Counters: A single value that only ever goes up (e.g., total requests, errors).
- Gauges: A single value that can go up or down (e.g., current active users, CPU usage).
- Histograms: Sample observations (like request durations) and count them in configurable buckets, often providing sum and count.
- Summaries: Similar to histograms but calculate configurable quantiles (e.g., 99th percentile latency) on the client side.
Erlang's Built-in Statistics
Erlang provides some basic system statistics out of the box through the erlang:statistics/1 function. These are useful for fundamental VM health checks.
You can get data like:
scheduler_wall_time: CPU time spent by schedulers.io: Total I/O operations.memory: Memory usage details.
While useful, these often aren't enough for application-specific insights. We need custom metrics.
Custom Metric Collection
For application-specific metrics, you'll often create your own. A common pattern is to use a gen_server process to manage and update metric values, ensuring atomic updates.
Try running this example of a simple counter managed by a gen_server:
-module(main).
-export([main/0]).
% --- simple_counter.erl (simplified for single file demo) ---
% A simple GenServer to maintain a counter
-behaviour(gen_server).
-export([init/1, handle_call/3, handle_cast/2, handle_info/2, terminate/2, code_change/3]).
% GenServer callbacks
init([]) -> {ok, 0}.
handle_call(get_value, _From, State) -> {reply, State, State}.
handle_cast(increment, State) -> {noreply, State + 1}.
handle_cast(stop, _State) -> {stop, normal, ok}.
handle_info(_Info, State) -> {noreply, State}.
terminate(_Reason, _State) -> ok.
code_change(_OldVsn, State, _Extra) -> {ok, State}.
% Client API for our counter
start_counter() -> gen_server:start({local, my_counter}, ?MODULE, [], []).
increment_counter() -> gen_server:cast(my_counter, increment).
get_counter_value() -> gen_server:call(my_counter, get_value).
stop_counter() -> gen_server:cast(my_counter, stop).
% --- End simple_counter.erl ---
main() ->
io:format("Starting a simple counter process...~n"),
{ok, _Pid} = start_counter(),
io:format("Initial counter value: ~p~n", [get_counter_value()]),
increment_counter(),
io:format("After 1 increment: ~p~n", [get_counter_value()]),
increment_counter(),
increment_counter(),
io:format("After 3 increments: ~p~n", [get_counter_value()]),
stop_counter(),
io:format("Counter process stopped.~n").The Telemetry Library
For more advanced and standardized metric collection in Erlang/Elixir, the Telemetry library is widely used. It provides a common way to emit events from your application.
Instead of manually updating a counter, you can:
- Define telemetry events (e.g.,
[:my_app, :request, :start],[:my_app, :request, :stop]). - Attach handlers to these events to process and aggregate metrics.
This decouples event emission from metric collection, making your system more flexible.
Exposing Metrics
Once you collect metrics, you need to make them available to external monitoring systems. This usually involves an exporter.
An exporter is a component (often an HTTP server) that:
- Gathers current metric values from your application.
- Formats them into a standard format (e.g., Prometheus text format).
- Serves them via an HTTP endpoint (e.g.,
/metrics).
Monitoring systems then 'scrape' this endpoint to collect the data.
Prometheus: A Pull-Based System
Prometheus is a popular open-source monitoring system. It works on a pull model: instead of your application pushing metrics, Prometheus regularly 'scrapes' (pulls) metrics from configured targets (your application's exporter endpoints).
It then stores these metrics as time-series data, allowing you to query and analyze trends over time. Prometheus is excellent for operational monitoring and alerting.
Prometheus Text Format
Erlang applications integrate with Prometheus by exposing an HTTP endpoint that serves metrics in the Prometheus text exposition format. This is a simple, human-readable format.
Here's an example of what Prometheus expects to scrape:
# HELP my_app_requests_total Total number of requests processed.
# TYPE my_app_requests_total counter
my_app_requests_total 1234
# HELP my_app_active_users Current number of active users.
# TYPE my_app_active_users gauge
my_app_active_users 50
# HELP my_app_request_duration_seconds Request duration in seconds.
# TYPE my_app_request_duration_seconds histogram
my_app_request_duration_seconds_bucket{le="0.1"} 100
my_app_request_duration_seconds_bucket{le="0.5"} 250
my_app_request_duration_seconds_bucket{le="1.0"} 350
my_app_request_duration_seconds_bucket{le="+Inf"} 400
my_app_request_duration_seconds_sum 150.0
my_app_request_duration_seconds_count 400Grafana for Visualization
While Prometheus stores and queries metrics, Grafana is the go-to tool for visualizing them. Grafana connects to various data sources (like Prometheus) and allows you to build interactive dashboards.
With Grafana, you can:
- Create graphs, charts, and tables.
- Combine data from multiple metrics.
- Set up alerts based on visual thresholds.
- Share dashboards with your team.
Monitoring Integration Check
You've learned about collecting and exposing metrics, and how Prometheus and Grafana work together. Let's test your understanding!
Recap & Next Steps
Great job! You've covered the essentials of metrics and monitoring integration:
- The importance of observability and the role of metrics.
- How to collect custom metrics in Erlang, including using
gen_serverprocesses. - The concept of exporters and the Prometheus text format.
- How Prometheus acts as a pull-based time-series database.
- How Grafana provides powerful visualization for your metrics.
With these tools, you can gain deep insights into your Erlang applications!
常见问题解答
「指标与监控集成」课时是免费的吗？
是的 — 「指标与监控集成」的完整文本可在网页上免费阅读。要进行交互式练习（内置代码编辑器和全天候 AI 导师）并解锁 Erlang OTP: Distributed & Fault-Tolerant Systems Programming 课程的其余内容，请升级到 CoddyKit PRO。 Erlang OTP: Distributed & Fault-Tolerant Systems Programming 课程共包含 4 节课。
「指标与监控集成」这节课中我会学到什么？
实现指标收集，并与 Prometheus 和 Grafana 等外部监控系统集成，以提升系统可观测性。 你通过在浏览器中直接运行的动手代码来练习 Erlang OTP: Distributed & Fault-Tolerant Systems Programming，全天候 AI 导师会在你学习这节课的过程中回答你的问题。
学习 Erlang OTP: Distributed & Fault-Tolerant Systems Programming 需要有经验吗？
无需任何先前经验。CoddyKit 上的 Erlang OTP: Distributed & Fault-Tolerant Systems Programming 课程适合初学者到高级学习者，你可以从这里开始或从头开始，按照自己的节奏学习。 这是第 3 节课，共 4 节。
「指标与监控集成」课时需要多长时间？
大多数 CoddyKit 课程大约需要 5–10 分钟。每节课都很精短且互动，所以你能稳步进步，并在网页和应用中从离开的地方继续。
我能在这节 Erlang OTP: Distributed & Fault-Tolerant Systems Programming 课中编写并运行代码吗？
能。每节 Erlang OTP: Distributed & Fault-Tolerant Systems Programming 课都包含内置代码编辑器，你可以在浏览器中直接编写并运行真实代码，并获得即时 AI 反馈 — 无需本地设置。