otel-supabase-edge-style

Compare original and translation side by side

🇺🇸

Original

English

🇨🇳

Translation

Chinese

OTel Supabase Edge Style

OTel Supabase Edge 风格

Hosted Supabase Edge Functions do not support the normal native OTel SDK lifecycle today. Use a tiny provider-neutral shim that looks like OTel and can be deleted later if native support lands.

Good reusable API names:

tracer.startActiveSpan("edge.chat", async (span) => { ... });
span.setAttributes({ "tenant.id": tenantId });
span.setStatus({ code: SpanStatusCode.ERROR });
meter.createCounter("llm.tokens.input").add(tokens, attrs);
histogram.record(durationMs, attrs);

Bad reusable API names:

sendSuperlogSpan(...);
recordCounter(...);
trackSuperlogEvent(...);

Keep the endpoint and ingest key inline in one setup area near the top of the function. Do not read them from env — the ingest key is project-scoped + write-only and Edge runtimes are flaky about env propagation anyway.

const SUPERLOG_ENDPOINT = "https://intake.superlog.sh";
const SUPERLOG_KEY = "superlog_live_…"; // set by superlog-onboard skill on pairing

目前托管版Supabase Edge Functions不支持常规的原生OTel SDK生命周期。可以使用一个轻量级、与供应商无关的类OTel适配层，待原生支持落地后即可移除。

推荐的可复用API命名：

tracer.startActiveSpan("edge.chat", async (span) => { ... });
span.setAttributes({ "tenant.id": tenantId });
span.setStatus({ code: SpanStatusCode.ERROR });
meter.createCounter("llm.tokens.input").add(tokens, attrs);
histogram.record(durationMs, attrs);

不推荐的可复用API命名：

sendSuperlogSpan(...);
recordCounter(...);
trackSuperlogEvent(...);

将端点和接入密钥集中配置在函数顶部的初始化区域。不要从环境变量中读取——接入密钥是项目级别的且仅可写入，而且Edge运行时在环境变量传播方面存在不稳定问题。

const SUPERLOG_ENDPOINT = "https://intake.superlog.sh";
const SUPERLOG_KEY = "superlog_live_…"; // set by superlog-onboard skill on pairing

Signals

信号类型

Traces: span around the function operation and each critical provider call.
Logs: concise structured OTLP log records for critical outcomes.
Metrics: tenant-tagged counters/histograms for critical operations.
LLM calls: capture
```
llm.tokens.input
```
/
```
llm.tokens.output
```
only when provider instrumentation is unavailable; tag explicit token counters with tenant, provider, model, use case, call site, and outcome. Do not calculate LLM cost in edge function code; Superlog estimates it centrally from provider/model/token data.

Use

EdgeRuntime.waitUntil(...)

or the runtime's equivalent when available so OTLP fetches can finish after the response is returned.

追踪（Traces）：为函数操作及每个关键供应商调用创建span。
日志（Logs）：为关键结果生成简洁的结构化OTLP日志记录。
指标（Metrics）：为关键操作添加租户标签的计数器/直方图。
LLM调用：仅当供应商工具不可用时，才捕获
```
llm.tokens.input
```
/
```
llm.tokens.output
```
；为明确的令牌计数器添加租户、供应商、模型、使用场景、调用位置及结果标签。不要在边缘函数代码中计算LLM成本；Superlog会基于供应商/模型/令牌数据集中估算成本。

如果可用，请使用

EdgeRuntime.waitUntil(...)

或运行时的等效方法，确保OTLP请求在返回响应后仍能完成。