aioprometheus中的quantile

### aioprometheus Quantile Usage and Implementation

In the context of `aioprometheus`, a library designed to work with Prometheus metrics within asynchronous Python applications, handling quantiles is essential for monitoring performance characteristics such as response times or request sizes. Unlike NumPy's direct computation method using `np.quantile`[^1], `aioprometheus` integrates this functionality into its metric types like histograms.

For implementing quantiles in `aioprometheus`, one typically uses histogram buckets rather than directly specifying quantile values at collection time. Histograms allow tracking the distribution of events over specified ranges (buckets). After collecting data points across these predefined intervals, external systems like Prometheus can calculate desired quantiles from aggregated bucket counts.

Here’s an example demonstrating how to set up a histogram in `aioprometheus`:

from aioprometheus import Counter, Gauge, Summary, Histogram, Registry, render

# Create registry instance.
registry = Registry()

# Define histogram with specific buckets suitable for latency measurements.
latency_histogram = Histogram('http_request_duration_seconds', 'HTTP Request duration in seconds',
                              buckets=[0.1, 0.5, 1.0, 2.0])

# Register the histogram with the registry.
registry.register(latency_histogram)

async def record_latency(duration):
    """Record HTTP request latencies."""
    latency_histogram.observe(duration)

Prometheus clients periodically scrape collected metrics including those defined by histograms. Once scraped, users leverage PromQL queries on the server side to compute various percentiles/quantiles based on observed samples stored within each bucket range.

--related questions--

1. How does one define appropriate bucket boundaries when setting up histograms?
2. What are some best practices for choosing which metrics should be tracked via histograms versus counters or gauges?
3. Can you provide examples of PromQL expressions used to extract meaningful insights from histogram-based metrics?
4. In what scenarios might it be more beneficial to use summaries instead of histograms for capturing percentile information?