Skip to content

Variables nothing reads

What the boot says of a variable under the prefix that no config reads — misspelled, retired, shadowed by a pin — by name, never by value.

A variable under the prefix that no config reads used to cost the deployment its value in silence: a misspelled key, or a name a release renamed, left the default in place and nothing said so. The boot now does, once per variable, at warn on nest_rs::config, by name and never by value:

MessageWhenFields
config variable read by no configa config of this binary read the variable’s namespace, and none read its keyvariable, namespace, and suggestion — the nearest key that was read, when one is close enough, or the one key it names once renamed (HSTS for SECURITY_HEADERS_HSTS)
variable read by nothing since 7.0the key is one 7.0 retired, every row of the upgrading guide’s table, which the crate that retired it declares and the namespace that reads it now was read (<PREFIX>_HTTP__CORS_EXPOSED, <PREFIX>_OAUTH_RESOURCE__RESOURCE, <PREFIX>_REDIS__WORKER__SHUTDOWN_TIMEOUT_SECS) — the words OpenTelemetry’s retired namespace and the console’s <PREFIX>_LOG are said in toovariable, namespace — the one reading it now — and read_as, the variable it became, when it became one
config variable from the .env cascade shadowed by a config pinned in code; export it into the real environment, or set it in the pina .env file sets a key a config reads, but the app pinned that config in code (Module::for_root(cfg)), which cuts its whole namespace off the cascadevariable, namespace, and pinned_by — the import that pinned it, `HttpModule::for_root(..)` at `imports[4]` of `ApiModule`
config variable under a misspelled namespacethe variable’s namespace is a near miss of one this binary read — other separators (<PREFIX>_REDIS_QUEUE__LEASE_SECS for redis__queue, <PREFIX>_SEAORM_URL for seaorm), another case, the prefix’s included, since the loader folds none (<PREFIX>_Seaorm__URL, nestrs_seaorm__url), or one misspelled segment of six letters or more, a family member’s included (<PREFIX>_SEAROM__URL, <PREFIX>_OAUTH__RESOURSE__RESOURCE) — and its key is one that namespace reads, or one edit from onevariable, namespace (the linked spelling), and suggestion — the name that is read

Everything else stays silent, and that is the design. One .env serves several binaries (what is one app’s alone sits in its own apps/<app>/.env), so a namespace this binary does not read, and that is no near miss of one it does, is another binary’s — and a near miss is held to both halves of the variable for that reason. A namespace under six letters has no room for a typo that is not also a word (auth and authz beside authn, es beside ws), so only its separators and its case count; and a misspelled namespace still carries the key the deployment meant, which the near namespace reads, while another binary’s variable carries its own — <PREFIX>_OPENAI__API_KEY beside a linked openapi names no key openapi reads, and stays silent. A key holding __ under a namespace it does read is a sub-namespace another binary links — redis__queue under redis — reported only when it is a near miss of a key read here. The framework’s process-level names, <PREFIX>_LOG_FORMAT, <PREFIX>_LOG_SOURCE_LOCATION, <PREFIX>_ENV and NESTRS_ENV_PREFIX, are never reported; the retired <PREFIX>_LOG is said by the console itself.

A retired key whose loss would leave a policy weaker than the one set is no line: the boot refuses it, <VAR> is read by nothing since 7.0 — set <NEW>: <why> — a security header’s flat key (<PREFIX>_HTTP__HSTS), which would go unsent, and <PREFIX>_GRAPHQL__DISABLE_INTROSPECTION, whose successor is its opposite. Under a config pinned in code, the pin holds the policy and cuts its namespace off the .env cascade, so such a key in a .env file is said retired like any other, while one the deployment sets is still refused.

The check runs where each config is read, against the keys its overlay actually asked for — the namespace’s misspellings included, at that same read, ended by an error or not, so a renamed required variable is named before the boot error its absence causes. A key read only behind a branch — HttpCors reads its other keys only when CORS_ORIGINS is set — is therefore reported exactly when the deployment did not take the branch, which is when it does nothing. A read made before any subscriber listens waits for the first read that has one, and a deployment that filters nest_rs::config out pays nothing for the scan. Upgrading from 6.x shows what it prints for a boot that kept a renamed variable.