Global `EnvFilter` prevents topic spans from being constructed at the default log level
评估
- 难度
- 4/5
- 预计耗时
- 3-5 天
- 新手友好度
- 68/100
- Issue 类型
- 缺陷
- 描述清晰度
- 描述清楚
- 活跃度
- 活跃
- 技术栈
- rust
调研方向
Start in crates/tracing/src/init.rs, especially the subscriber construction around lines 90-93 and default_env_filter, then inspect crates/tracing/src/layers/metrics.rs and the existing tracing tests. Run the real pluto_tracing::init stack at the default info level and verify that topic labels are retained while console and Loki verbosity still follow --log-level. Add a regression test covering warn/error metrics and their topic labels.
由索引模型根据 Issue 内容生成。
描述
Summary
EnvFilter is attached as a global layer on the subscriber stack, so it gates Subscriber::enabled for every layer — including MetricsLayer: init.rs#L90-L93.
Every topic span in the workspace is DEBUG-level, and the default --log-level is info (common.rs#L103, default_env_filter). At info the span is never constructed, MetricsLayer::on_new_span never fires, no SpanTopic extension is stored, and event_topic walks an empty scope.
Result: 100% of warn/error events are counted under topic="" at the default configuration. The topic mechanism only works under RUST_LOG=debug.
Evidence
Probed against the real MetricsLayer, both directions:
| subscriber stack | console filter | debug_span! constructed |
app_log_warn_total label |
|---|---|---|---|
EnvFilter global (current) |
info |
no (is_disabled=true) |
topic="" |
EnvFilter global (current) |
debug |
yes | topic="<name>" |
EnvFilter on fmt only + MetricsLayer.with_filter(LevelFilter::DEBUG) |
info |
yes | topic="<name>" |
The third row is the proposed fix: the console still emits at info (the WARN printed normally), while the metric keeps its label.
Scope
This predates the topic work and is not caused by it. It already affects the two topics on main — health and stacksnipe — and will silently neutralise all 17 topics once #588 / #669 lands.
Why it matters
app_log_{warn,error}_total{topic}is the only per-component signal for warn/error rate. With one label value it answers no question that the unlabelled counter does not already answer — which is exactly what #588 set out to fix.- Generated alert rules are silently degraded.
define.rs#L375buildsincrease(app_log_warn_total{topic!~"vmock|tracker"}[30s]) > 2. Against a pluto node every series carriestopic="", which matches!~"vmock|tracker"— so the exclusion never excludes anything and the rule is strictly more trigger-happy than intended. This directly undercuts the re-validation planned in #583. - Grafana panels collapse.
dash_charon_overview.json#L1701-L1714aggregatesby (topic); pluto contributes a single empty series next to charon's ~20. - It blocks #47 (
app/tracer). The same global filter drops debug-level spans before they can reach an OTLP exporter, so an OTel integration inherits the problem rather than working around it.
Charon contrast
The counter is level-independent by construction: incWarnCounter(ctx) is the first statement of log.Warn, before the logger is consulted — log.go#L107-L108, same for log.Error. Charon's topic rides on context.Context, so console verbosity and metric labelling are independent concerns. Pluto couples them through the span, which is what makes the filter placement load-bearing.
Proposed fix
Move EnvFilter off the subscriber and onto the layers that should honour the operator's verbosity — fmt and Loki — and give MetricsLayer its own LevelFilter::DEBUG filter so topic spans are always materialised:
let registry = Registry::default()
.with(fmt_layer.with_filter(env_filter))
.with(MetricsLayer.with_filter(LevelFilter::DEBUG));
Loki needs the same treatment (it currently sits behind the global filter too), and EnvFilter is not Clone, so the filter has to be built per layer from the same directive string.
The alternative — promoting every topic span to info — is a smaller diff but makes span names visible in default console output and leaves the counter level-dependent, which is the coupling charon does not have.
Acceptance
-
app_log_{warn,error}_totalcarries itstopiclabel with the subscriber built bypluto_tracing::initat the defaultinfolevel. - A regression test in
crates/tracingpins this against the real layer stack, not a hand-built one. -
--log-levelstill controls console and Loki verbosity exactly as before.
Context
Diagnosed while reviewing #669; see this comment for the original analysis. Related: #588 (topic labels), #631 (span propagation), #583 (alert-rule re-validation), #47 (app/tracer).
- 主要语言
- Rust
- 星标
- 8
- 派生
- 5
- 平均合并
- 4 天 18 小时
- 30 天内合并 PR
- 4
环境准备
- 提供 Dockerfile 或 Docker Compose 文件
- 有 Pull Request 模板
- 阅读贡献指南
从这里开始
- 先读完整个 Issue,再读项目的贡献指南。
- 在 Issue 下留言说明你要接手 —— 这能避免两个人做同样的事。
- Fork 仓库,在一个分支上完成修改。
- 提交 Pull Request,并在描述里引用这个 Issue 编号。
NethermindEth/pluto 的其他 Issue
-
难度 4/5 3-5 天 新手友好度 52/100
NethermindEth/pluto#719 ·
-
enhancement rust
难度 4/5 3-5 天 新手友好度 45/100
NethermindEth/pluto#640 ·
-
rust
难度 5/5 一周以上 新手友好度 45/100
NethermindEth/pluto#638 · 1 条评论 ·
-
enhancement rust
难度 5/5 一周以上 新手友好度 35/100
NethermindEth/pluto#635 ·
-
Improve `alpha test peers`: match Charon's probes; reduce timeouts可能已有人在做 @varex83agent 于 23 天前认领。 未关闭bug track:orchestration-cli
难度 5/5 一周以上 新手友好度 28/100
NethermindEth/pluto#632 ·
查看 NethermindEth/pluto 的全部 Issue
相似的 Issue
-
难度 2/5 1-3 小时 新手友好度 85/100
farion1231/cc-switch#8829 ·
维护者通常 1 天内回复
-
bug ci good first issue
难度 2/5 1-3 小时 新手友好度 88/100
维护者通常 1 天内回复
-
难度 1/5 1 小时以内 新手友好度 92/100
longbridge/gpui-kit#3677 ·
维护者通常 1 天内回复
-
E-pr-welcome
难度 2/5 1-3 小时 新手友好度 77/100
seanmonstar/reqwest#3121 · 2 条评论 ·
维护者通常 1 天内回复
-
难度 1/5 1 小时以内 新手友好度 68/100
ilya-zlobintsev/LACT#1235 ·
维护者通常 1 天内回复