Troubleshooting

Start with the checks below. Set SW_AGENT_LOGGING_LEVEL=debug for a short test when you need more agent detail. Change it back after the test because debug logging adds work and can produce large logs.

No service or traces in SkyWalking

1. Confirm that Node.js is version 20 or later.
2. Confirm that agent.start() runs before Express, HTTP clients, database clients, and other instrumented modules are loaded.
3. Check SW_DISABLE. The exact value true keeps the agent stopped.
4. Check SW_AGENT_NAME and select that service name in the SkyWalking UI.
5. Send a request that uses a supported library.
6. Check that the path, suffix, and HTTP method are not filtered by trace settings.
7. Check the agent log for plugin version or OAP connection errors.

If the process is short-lived, call await agent.flush() before exit. flush() is a bounded best-effort attempt, not a delivery guarantee. See Data is missing when the process stops.

OAP connection errors

The collector setting must use host:port form:

export SW_AGENT_COLLECTOR_BACKEND_SERVICES=oap.example.com:11800

Check DNS, network access, firewall rules, and the OAP gRPC port.

One host:port uses the grpc-js dns: resolver (all A/AAAA records become endpoints and are re-resolved periodically). A comma-separated list uses a static resolver with pick_first: each name is a literal endpoint only — no DNS expansion or re-resolution per name, so discovery is weaker than a single DNS name (for example a headless Kubernetes service).

An option passed to agent.start() replaces the environment value. This includes an empty string:

// Do not do this. It removes the default collector address.
agent.start({ collectorAddress: '' });

For a TLS OAP endpoint, set SW_AGENT_SECURE=true. The agent uses the system trust store. It has no configuration for a custom CA or client certificate. Set SW_AGENT_AUTHENTICATION if OAP requires an agent token.

Under TLS with multiple hostnames, certificate verification follows the channel authority (the first list entry in the configured target). Endpoint pick order may be shuffled by grpc-js, but the target string — and therefore authority / SNI — stays in config order. Every backend must present a certificate that shares the needed SANs, or failover handshakes fail. Prefer one DNS name with multiple A/AAAA records for TLS high availability.

Channel disconnect lines are logged at error and recover lines at warn (throttled separately so a recover line is not swallowed by the disconnect window). For per-address grpc-js detail, set GRPC_TRACE=pick_first,subchannel.

Traces are missing during an OAP outage

The agent keeps finished segments in a memory buffer. When the buffer reaches SW_AGENT_MAX_BUFFER_SIZE, it removes the oldest finished segment and may log that the trace buffer reached maximum size. Restore the OAP connection; increasing the buffer only delays data loss and uses more process memory.

If a report attempt fails, the agent may log that it discarded N trace segment(s) after report failure. Those segments are not re-sent. Reporting is best-effort: failures discard data, and flush() only waits briefly then tries once more.

A library has no spans

• Check that the agent started before the library loaded.
• Check the library and plugin in Supported libraries.
• Most plugin version rules are broad, but CI tests exact versions from package-lock.json. Test other versions in your application.
• Webpack uses a smaller static plugin set. See Webpack.
• AWS SDK for JavaScript v3 is not covered by the AWS SDK v2 plugins.

Runtime metrics do not appear

1. Check that SW_AGENT_NODEJS_RUNTIME_METRICS_REPORTER_ACTIVE is not false.
2. Wait at least one report period. The default is 20 seconds.
3. Check that your OAP version includes the Node.js runtime meter rules and dashboard.
4. Check the OAP connection and agent log.

See Node.js runtime metrics for meter names and OAP setup.

Agent logs are hard to find

The default agent log level is warn.

Level	What you typically see
error	Auth rejection, channel disconnect (throttled), fatal boot failures
warn (default)	Channel recovered (throttled), trace buffer full, discarded segment batches, meter report failures (throttled)
info / debug	Lifecycle noise; span debug lines

Levels below the configured threshold are silent by design (warn / info / debug become no-ops when the threshold is higher).

• When NODE_ENV is not production, logs go to the console.
• When NODE_ENV=production, logs go to skywalking.log in the process working directory.
• Set SW_LOGGING_TARGET=console to use the console in production.

Agent to OAP over an HTTP proxy

Agent to OAP over an HTTP proxy is not supported. Every gRPC channel sets grpc.enable_http_proxy=0, so host http_proxy / https_proxy never affect OAP uplink. This applies to single-address and multi-address targets. After upgrade, traffic that previously relied on an HTTP CONNECT proxy to reach OAP will no longer use that proxy.

Data is missing when the process stops

agent.destroy() stops reporters but does not flush them. Use this order:

await agent.flush();
agent.destroy();

flush() waits a short time for in-flight work and may start one more report attempt. It does not guarantee delivery if OAP is slow or unreachable.

Do not use stop and restart as a normal agent update method inside one process. Module patches stay installed after destroy().

SQS receives only one message

This is current AWS SDK v2 SQS plugin behavior. The plugin removes MaxNumberOfMessages so it can link one message to one entry span. Disable AWS2SQS if the application requires batch receives.

Ask for help

If the checks do not solve the problem, open an issue in the Apache SkyWalking issue tracker. Include the Node.js Agent version, Node.js version, OAP version, target library versions, startup code, and relevant agent errors. Remove tokens and private data first.