Files
silo-server/internal/telemetry/telemetry.go
203a18ae83 feat(observability): OpenTelemetry logs+traces with secret redaction and slog standardization (#290)
* feat(observability): OpenTelemetry logs+traces with secret redaction

Part of #265. Adds opt-in OpenTelemetry (logs + traces) alongside the existing
stderr + opslog pipeline, plus secret redaction on all sinks. Default-off: with
no OTEL_* / SILO_OTEL_ENABLED config, behavior is unchanged.

Bootstrap (internal/telemetry):
- Setup() builds one shared resource, a TracerProvider (parent-based trace-id
  ratio sampler), a LoggerProvider, and the W3C TraceContext+Baggage propagator
  from env. It installs NO MeterProvider — metrics stay on Prometheus, and the
  built-in no-op global MeterProvider keeps the trace instrumentation libs from
  double-emitting. Shutdown is deferred with a flush timeout.
- Logs are bridged via otelslog fan-out (slog.MultiHandler), level-gated by the
  shared LevelVar and best-effort so a failing collector can't break the console
  or DB branches. stderr + opslog stay untouched.

Secret redaction (internal/logredact):
- A slog.Handler masks secret-keyed attributes (password, token, api_key,
  authorization, cookie, ...) — including .With-bound attrs, nested groups,
  secret-keyed group subtrees, and values behind a LogValuer — on the console
  and OTLP sinks, with a no-op fast path when a record has no secret keys.
  opslog.shouldRedact delegates to logredact.SecretKey so all sinks share one
  marker list.

Rotation is infra-managed (no custom file sink): container runtime for stderr,
collector/backend for OTLP, opslog partition-pruning for the DB. Documented in
docs/architecture/observability.md.

Verification: go build ./..., go vet, gofmt -l — clean; go test
./internal/telemetry/ ./internal/logredact/ -race pass.

AI-use disclosure: implemented with AI assistance (Claude Code), including
adversarial reviews that hardened the bootstrap and fixed two redaction leak
paths; reviewed by the author.

* refactor(observability): slog context+component sweep, sloglint gate (phase 3)

Part of #265. Builds on the OTel bootstrap + redaction commit.

Standardizes every log call site onto the context-carrying slog variants so
records correlate with the active OpenTelemetry trace, and locks the standard
in with a machine gate so future code (human- or AI-authored) can't drift back.

- Call-site sweep: converted the remaining slog.<Level>(...) calls to the
  slog.<Level>Context(ctx, ...) form wherever a context.Context is in scope
  (background/init calls with no ctx are left as-is), across 183 files. Applied
  via a type-aware AST codemod. Log levels and message strings are preserved
  verbatim; a component attr (canonical per-package name) is added to direct
  package-level slog calls. Bound-logger calls keep their existing .With
  bindings. The main.go and telemetry package conversions rode with their file
  in the previous commit to keep each file within a single commit.
- Enforcement (.golangci.yml): enable sloglint with context=scope, static-msg,
  key-naming-case=snake, no-mixed-args. After the sweep all four report zero
  violations repo-wide (tests included), so make lint / CI now blocks any
  regression to the non-context form. The gate ships with the sweep because it
  cannot be green until the legacy sites are converted.

Metrics remain on Prometheus; no behavior change to /metrics or Grafana.

Verification: go build ./..., go vet ./..., gofmt -l — clean; sloglint (all 4
rules) 0 violations repo-wide; log levels verified unchanged.

AI-use disclosure: implemented with AI assistance (Claude Code), including the
codemod; reviewed by the author.

* fix(observability): honor per-signal OTLP protocol and secret WithGroup names

Two Codex review findings on PR #290:

- telemetry: OTEL_EXPORTER_OTLP_{TRACES,LOGS}_PROTOCOL now override the
  generic OTEL_EXPORTER_OTLP_PROTOCOL per signal, so mixed collector
  setups (e.g. HTTP logs + gRPC traces) build the right exporter.
- logredact: entering a group whose name is secret-bearing (e.g.
  WithGroup("authorization")) now masks every leaf in that subtree,
  matching how slog.Group("authorization", ...) is masked as a whole.

* fix(observability): address review feedback on telemetry bootstrap

- Telemetry setup failure no longer kills boot: Setup returns usable
  no-op providers alongside the error and main logs and continues with
  telemetry disabled, honoring the best-effort contract.
- Honor OTEL_TRACES_SAMPLER (always_on/off, traceidratio, parentbased_*
  variants); unsupported values fall back to parentbased_traceidratio.
- Attach node identity as semconv service.instance.id instead of the
  non-semconv node.name.
- Rename opslog retention-scope log attrs to target_component/target_level
  so they no longer collide with the canonical component routing key, and
  tag those lines with component=opslog.
- Fix stale levelGated comment casing; use WarnContext in the telemetry
  shutdown defer; document the LogValuer double-resolve on the redaction
  slow path.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

---------

Co-authored-by: Quick <31828688+Quick104@users.noreply.github.com>
Co-authored-by: Claude Fable 5 <noreply@anthropic.com>
2026-07-09 08:53:52 -04:00

196 lines
7.0 KiB
Go

package telemetry
import (
"context"
"errors"
"log/slog"
"sync"
"go.opentelemetry.io/otel"
"go.opentelemetry.io/otel/attribute"
otlplog "go.opentelemetry.io/otel/exporters/otlp/otlplog/otlploggrpc"
otlploghttp "go.opentelemetry.io/otel/exporters/otlp/otlplog/otlploghttp"
"go.opentelemetry.io/otel/exporters/otlp/otlptrace/otlptracegrpc"
"go.opentelemetry.io/otel/exporters/otlp/otlptrace/otlptracehttp"
logapi "go.opentelemetry.io/otel/log"
"go.opentelemetry.io/otel/log/global"
lognoop "go.opentelemetry.io/otel/log/noop"
"go.opentelemetry.io/otel/propagation"
sdklog "go.opentelemetry.io/otel/sdk/log"
"go.opentelemetry.io/otel/sdk/resource"
sdktrace "go.opentelemetry.io/otel/sdk/trace"
semconv "go.opentelemetry.io/otel/semconv/v1.26.0"
)
// Providers holds the constructed OpenTelemetry providers. When telemetry is
// disabled, LoggerProvider is a no-op so callers can build an slog handler
// unconditionally without branching.
type Providers struct {
// LoggerProvider is the OTel logs provider. Never nil (no-op when disabled).
LoggerProvider logapi.LoggerProvider
}
// noopShutdown is an inert shutdown used when telemetry is disabled.
func noopShutdown(context.Context) error { return nil }
// noopProviders returns Providers backed by no-op implementations, used both
// when telemetry is disabled and on setup failure so callers always get a
// usable value.
func noopProviders() *Providers {
return &Providers{LoggerProvider: lognoop.NewLoggerProvider()}
}
// Setup initializes the OpenTelemetry SDK from cfg. When cfg.Enabled is false it
// installs nothing (leaving all otel globals as their built-in no-ops) and
// returns a Providers holding no-op providers plus a no-op shutdown.
//
// When enabled it builds one shared resource, a composite W3C propagator, a
// TracerProvider, and a LoggerProvider, registering each as the corresponding
// otel global. It deliberately does NOT build or register a MeterProvider:
// metrics stay on Prometheus, and the built-in no-op global MeterProvider keeps
// the trace instrumentation libraries from double-emitting metrics.
//
// The returned shutdown flushes and closes all providers; it is idempotent and
// joins any errors from the underlying shutdown funcs.
//
// On failure Setup installs no globals and still returns usable no-op providers
// and a no-op shutdown alongside the error, so callers can log the error and
// keep running with telemetry disabled rather than treating it as fatal.
func Setup(ctx context.Context, cfg Config) (*Providers, func(context.Context) error, error) {
if !cfg.Enabled {
return noopProviders(), noopShutdown, nil
}
// resource.New can return a usable, fully-merged resource together with a
// non-fatal error (e.g. ErrSchemaURLConflict when a detector's schema differs
// from ours). Per OTel guidance, use the returned resource and treat the error
// as a warning; only abort if no resource came back. This preserves the
// "telemetry is best-effort, must never brick the server" contract.
res, err := buildResource(ctx, cfg)
if err != nil {
if res == nil {
return noopProviders(), noopShutdown, err
}
slog.WarnContext(ctx, "telemetry: resource built with warnings", "component", "telemetry", "error", err)
}
var shutdownFuncs []func(context.Context) error
// TracerProvider. Build the exporter/provider before mutating any process
// globals so a failure leaves nothing installed.
traceExp, err := newTraceExporter(ctx, cfg)
if err != nil {
return noopProviders(), noopShutdown, err
}
tp := sdktrace.NewTracerProvider(
sdktrace.WithResource(res),
sdktrace.WithBatcher(traceExp),
sdktrace.WithSampler(newSampler(cfg)),
)
shutdownFuncs = append(shutdownFuncs, tp.Shutdown)
// LoggerProvider.
logExp, err := newLogExporter(ctx, cfg)
if err != nil {
// Best-effort cleanup of what we already built before failing. Nothing
// is installed as a global yet, so this only drains the trace batcher.
_ = tp.Shutdown(ctx)
return noopProviders(), noopShutdown, err
}
lp := sdklog.NewLoggerProvider(
sdklog.WithResource(res),
sdklog.WithProcessor(sdklog.NewBatchProcessor(logExp)),
)
shutdownFuncs = append(shutdownFuncs, lp.Shutdown)
// Both exporters constructed successfully: now install the globals. Setting
// them last preserves the "install nothing on failure" invariant.
otel.SetTracerProvider(tp)
global.SetLoggerProvider(lp)
otel.SetTextMapPropagator(propagation.NewCompositeTextMapPropagator(
propagation.TraceContext{},
propagation.Baggage{},
))
shutdown := newIdempotentShutdown(shutdownFuncs)
return &Providers{LoggerProvider: lp}, shutdown, nil
}
// buildResource constructs the single shared resource describing this process.
func buildResource(ctx context.Context, cfg Config) (*resource.Resource, error) {
attrs := []attribute.KeyValue{semconv.ServiceName(cfg.ServiceName)}
if cfg.ServiceVersion != "" {
attrs = append(attrs, semconv.ServiceVersion(cfg.ServiceVersion))
}
if cfg.NodeID != "" {
// service.instance.id is the semconv slot for per-instance identity;
// backends key on it to distinguish nodes sharing one service.name.
attrs = append(attrs, semconv.ServiceInstanceID(cfg.NodeID))
}
return resource.New(ctx,
resource.WithFromEnv(),
resource.WithProcess(),
resource.WithHost(),
resource.WithAttributes(attrs...),
)
}
// newSampler maps the configured OTEL_TRACES_SAMPLER strategy to an SDK
// sampler. Unrecognized values were already normalized to the parent-based
// trace-id-ratio default by parseSampler.
func newSampler(cfg Config) sdktrace.Sampler {
switch cfg.Sampler {
case SamplerAlwaysOn:
return sdktrace.AlwaysSample()
case SamplerAlwaysOff:
return sdktrace.NeverSample()
case SamplerTraceIDRatio:
return sdktrace.TraceIDRatioBased(cfg.SamplerRatio)
case SamplerParentBasedAlwaysOn:
return sdktrace.ParentBased(sdktrace.AlwaysSample())
case SamplerParentBasedAlwaysOff:
return sdktrace.ParentBased(sdktrace.NeverSample())
default:
return sdktrace.ParentBased(sdktrace.TraceIDRatioBased(cfg.SamplerRatio))
}
}
// newTraceExporter builds the OTLP trace exporter for the configured protocol.
func newTraceExporter(ctx context.Context, cfg Config) (sdktrace.SpanExporter, error) {
if cfg.TracesProtocol == ProtocolHTTP {
return otlptracehttp.New(ctx)
}
return otlptracegrpc.New(ctx)
}
// newLogExporter builds the OTLP log exporter for the configured protocol.
func newLogExporter(ctx context.Context, cfg Config) (sdklog.Exporter, error) {
if cfg.LogsProtocol == ProtocolHTTP {
return otlploghttp.New(ctx)
}
return otlplog.New(ctx)
}
// newIdempotentShutdown returns a shutdown func that runs the accumulated
// shutdown funcs exactly once, joining their errors.
func newIdempotentShutdown(funcs []func(context.Context) error) func(context.Context) error {
var once sync.Once
var err error
return func(ctx context.Context) error {
once.Do(func() {
var errs []error
for _, fn := range funcs {
if fn == nil {
continue
}
if e := fn(ctx); e != nil {
errs = append(errs, e)
}
}
err = errors.Join(errs...)
})
return err
}
}