`dista.metrics`

Prometheus metrics: counters and a histogram, and the /metrics page.

With DISTA_METRICS_PORT set, the head and every worker answer HTTP on that port with Prometheus's text format, which Kubernetes monitoring scrapes as it is. It's off by default, since in local mode the head and its workers share a host.

The server is the smallest one that does: the event loop polls its listener, and each connection is answered on the spot. The request is read for up to READ_TIMEOUT_MS and only its first line matters; every path gets the metrics. A scrape is a few kilobytes, so answering it blocks the loop for well under a millisecond, but a client that connects and sends nothing holds it for the timeout.

Structs

`Histogram`

struct Histogram

A Prometheus histogram of durations.

Implements Copyable, Movable

Fields: counts: List[Int]: Per bucket, how many observations were at most its bound.; count: Int; sum: Float64

`init`

def __init__(out self)

`observe`

def observe(mut self, seconds: Float64)

`write`

def write(self, mut out: String, name: String, help: String)

Functions

`bucket_bound`

def bucket_bound(i: Int) -> Float64

Bucket i's upper bound, in seconds.

`write_metric`

def write_metric(mut out: String, kind: String, name: String, help: String, value: Int)

One counter or gauge, without labels.

`write_outcomes`

def write_outcomes(mut out: String, name: String, help: String, ok: Int, err: Int)

A counter split by outcome, ok and error.

`metrics_listener`

def metrics_listener() -> Optional[Listener]

A listener on DISTA_METRICS_PORT, on every interface, if it's set.

`answer_scrape`

def answer_scrape(l: Listener, body: String)

Accepts one connection on l and answers it with body. A client that misbehaves only loses its answer.

Aliases

comptime READ_TIMEOUT_MS = 200

comptime BUCKETS = 6

Task duration buckets: 1 ms, 10 ms, 100 ms, 1 s, 10 s and 60 s.