Skip to main content
By default the SDK logs through Python’s standard logging module, to the logger named westyx_nexus, which carries only a NullHandler. That is what the Logging HOWTO asks of a library: silent until the application configures logging, visible as soon as it does, with nothing passed to the SDK.
The name is exported as LOGGER_NAME.

What the SDK logs

The startup summary

After the first sync both clients log one info record naming what they received:
Names only, never a value. Each list is sorted and capped at 50 names, with a +N more entry after that, so a large service stays one readable line. It is how you confirm from a pod’s log which configs, secrets and flags a deployment actually pulled.

Resources the server did not deliver

A sync can report resources it left out - a secret that has no value yet (secret_not_set), a config whose {{secret:KEY}} reference does not resolve (config_unresolved_secret). Such a secret or config reads as absent. Each omission is logged as a warning the first time it appears; client.sync_errors lists the current ones. log_sync_errors=False keeps the list without the log lines.

Secret values never reach the logger

No log record carries a secret’s value, the API key or a session token. This is enforced: a test drives a sync, a write and the WIF path through a capturing logger and fails if any of them appears. Key names do appear - in the startup summary and in the omission warning. A name identifies what to look up; it is not the credential.

A custom logger

NexusConfig.logger takes any object with these methods - a Protocol, so there is no base class to inherit:
A logger written without warning still works: its warnings arrive at info. StdlibNexusLogger(name="myapp.nexus") nests the SDK under your own logger hierarchy.

structlog

Loguru

A logger that raises is contained

A logger whose method raises never fails an SDK operation. Every internal log call is wrapped, so a failing sink - a full disk, a down syslog target - is dropped and the read, refresh, or create() that logged carries on. The stream observer is contained the same way. The Workload Identity credential source is deliberately not - it is on the auth path, so its error surfaces rather than being swallowed.