bb71e7b1c7
CI / changes (pull_request) Successful in 2s
CI / unit (pull_request) Successful in 12s
CI / integration (pull_request) Successful in 26s
CI / ui (pull_request) Successful in 1m11s
CI / conformance (pull_request) Successful in 10s
CI / gate (pull_request) Successful in 0s
CI / deploy (pull_request) Successful in 1m56s
The bot runs on its own host and exports no telemetry (otelcol is unreachable
from there), so it was a monitoring blind spot. Observe its Bot API health
centrally by wrapping the HTTP client — one place, no per-call-site
instrumentation — and relay it up the existing bot-link as a periodic Health
message the gateway turns into its own metrics.
- proto: add Health (delta connect / api / 429 counters + a last-ok stamp) to
the FromBot oneof (additive, backward-compatible).
- bot: platform/telegram/internal/health wraps the Bot API HTTP client — it
stamps liveness on any 2xx (so the getUpdates long-poll keeps it fresh even
when idle), classifies transport/5xx failures (getUpdates vs other) and 429s,
and honours a 429's Retry-After (bounded) so the bot backs off; a 4xx other
than 429 is a normal per-request outcome and is not counted.
- bot-link client: flush the reporter as a Health message every 30s over a
single-sender loop (a gRPC stream forbids concurrent Send).
- gateway: fold each report into bot_tg_errors_total{kind} and the
bot_tg_last_ok_unix liveness gauge.
- grafana: alerts (bot disconnected, bot not reaching the Bot API, sustained
429s) routed to the operator email — which does not go through the bot — plus
a Telegram-bot dashboard.
- docs (ARCHITECTURE, compose comments); unit tests (observer classification,
Retry-After, snapshot/commit, hub last-ok monotonicity).
146 lines
5.1 KiB
Go
146 lines
5.1 KiB
Go
// Package health tracks the Telegram bot's Bot API health and reports it to the gateway over the
|
|
// bot-link. The bot exports no telemetry of its own — otelcol lives on the main host, unreachable
|
|
// from the bot host — so the gateway turns these reports into its own metrics (see
|
|
// gateway/internal/botlink). Health is observed CENTRALLY by wrapping the Bot API HTTP client
|
|
// ([Reporter.Wrap]); every request is classified there with no per-call-site instrumentation.
|
|
package health
|
|
|
|
import (
|
|
"bytes"
|
|
"context"
|
|
"encoding/json"
|
|
"io"
|
|
"net/http"
|
|
"strconv"
|
|
"strings"
|
|
"sync/atomic"
|
|
"time"
|
|
|
|
botlinkv1 "scrabble/pkg/proto/botlink/v1"
|
|
)
|
|
|
|
// maxBackoff caps how long a single 429 makes the bot wait, so a hostile or buggy Retry-After can
|
|
// never wedge the bot.
|
|
const maxBackoff = 30 * time.Second
|
|
|
|
// Reporter accumulates the bot's Bot API health between reports: three delta counters and the last
|
|
// successful-response stamp. It is safe for concurrent use. The bot-link client periodically
|
|
// [Reporter.Snapshot]s it, sends the snapshot as a botlink Health, and [Reporter.Commit]s it on a
|
|
// successful send.
|
|
type Reporter struct {
|
|
connectFailures atomic.Int64
|
|
apiErrors atomic.Int64
|
|
rateLimited atomic.Int64
|
|
lastOKUnix atomic.Int64
|
|
}
|
|
|
|
// NewReporter returns an empty Reporter.
|
|
func NewReporter() *Reporter { return &Reporter{} }
|
|
|
|
// Snapshot reads the current delta counters and the last-ok stamp into a Health message WITHOUT
|
|
// resetting them. The counters are cleared only by [Reporter.Commit] after a successful send, so a
|
|
// failed send loses nothing.
|
|
func (r *Reporter) Snapshot() *botlinkv1.Health {
|
|
return &botlinkv1.Health{
|
|
ConnectFailures: uint64(max(r.connectFailures.Load(), 0)),
|
|
ApiErrors: uint64(max(r.apiErrors.Load(), 0)),
|
|
RateLimited: uint64(max(r.rateLimited.Load(), 0)),
|
|
LastOkUnix: r.lastOKUnix.Load(),
|
|
}
|
|
}
|
|
|
|
// Commit subtracts a just-sent snapshot's delta counters, so counts accrued between the snapshot and
|
|
// the send survive into the next report. last_ok is a stamp, not a delta, so it is left as is.
|
|
func (r *Reporter) Commit(sent *botlinkv1.Health) {
|
|
r.connectFailures.Add(-int64(sent.GetConnectFailures()))
|
|
r.apiErrors.Add(-int64(sent.GetApiErrors()))
|
|
r.rateLimited.Add(-int64(sent.GetRateLimited()))
|
|
}
|
|
|
|
// Wrap returns an http.Client-shaped observer over base for tgbot.WithHTTPClient: it stamps liveness
|
|
// on every successful response, counts transport and 5xx failures (split getUpdates vs other) and
|
|
// 429s, and honours a 429's Retry-After (bounded by maxBackoff) so the bot backs off instead of
|
|
// hammering. A 4xx other than 429 (a user blocked the bot, a malformed request) is a normal
|
|
// per-request outcome, not a health signal, and is not counted.
|
|
func (r *Reporter) Wrap(base Doer) Doer { return &observer{base: base, r: r} }
|
|
|
|
// Doer is the minimal HTTP surface the Bot API client needs; both *http.Client and [observer]
|
|
// satisfy it, and it matches the go-telegram/bot HttpClient interface.
|
|
type Doer interface {
|
|
Do(*http.Request) (*http.Response, error)
|
|
}
|
|
|
|
// observer is the Bot API HTTP client wrapper (see [Reporter.Wrap]).
|
|
type observer struct {
|
|
base Doer
|
|
r *Reporter
|
|
}
|
|
|
|
// Do performs the request and records its health outcome.
|
|
func (o *observer) Do(req *http.Request) (*http.Response, error) {
|
|
resp, err := o.base.Do(req)
|
|
poll := strings.HasSuffix(req.URL.Path, "/getUpdates")
|
|
if err != nil {
|
|
// A cancelled context is a shutdown, not a Bot API failure.
|
|
if req.Context().Err() == nil {
|
|
o.fail(poll)
|
|
}
|
|
return resp, err
|
|
}
|
|
switch {
|
|
case resp.StatusCode == http.StatusTooManyRequests:
|
|
o.r.rateLimited.Add(1)
|
|
o.backoff(req.Context(), resp)
|
|
case resp.StatusCode >= 200 && resp.StatusCode < 300:
|
|
o.r.lastOKUnix.Store(time.Now().Unix())
|
|
case resp.StatusCode >= 500:
|
|
o.fail(poll)
|
|
}
|
|
return resp, err
|
|
}
|
|
|
|
// fail records a connect failure on the getUpdates long-poll or an api error otherwise.
|
|
func (o *observer) fail(poll bool) {
|
|
if poll {
|
|
o.r.connectFailures.Add(1)
|
|
} else {
|
|
o.r.apiErrors.Add(1)
|
|
}
|
|
}
|
|
|
|
// backoff waits out a 429's Retry-After (bounded, cancellable). It buffers the response body first so
|
|
// the wait holds no connection open and the Bot API library can still parse the 429 afterwards.
|
|
func (o *observer) backoff(ctx context.Context, resp *http.Response) {
|
|
body, _ := io.ReadAll(resp.Body)
|
|
resp.Body.Close()
|
|
resp.Body = io.NopCloser(bytes.NewReader(body))
|
|
d := retryAfter(resp.Header.Get("Retry-After"), body)
|
|
if d <= 0 {
|
|
return
|
|
}
|
|
if d > maxBackoff {
|
|
d = maxBackoff
|
|
}
|
|
select {
|
|
case <-ctx.Done():
|
|
case <-time.After(d):
|
|
}
|
|
}
|
|
|
|
// retryAfter resolves the 429 back-off from the Retry-After header, falling back to the Bot API
|
|
// body's parameters.retry_after (both are seconds). It returns 0 when neither gives a positive value.
|
|
func retryAfter(header string, body []byte) time.Duration {
|
|
if secs, err := strconv.Atoi(strings.TrimSpace(header)); err == nil && secs > 0 {
|
|
return time.Duration(secs) * time.Second
|
|
}
|
|
var payload struct {
|
|
Parameters struct {
|
|
RetryAfter int `json:"retry_after"`
|
|
} `json:"parameters"`
|
|
}
|
|
if err := json.Unmarshal(body, &payload); err == nil && payload.Parameters.RetryAfter > 0 {
|
|
return time.Duration(payload.Parameters.RetryAfter) * time.Second
|
|
}
|
|
return 0
|
|
}
|