encoredev Encore.ts structured logging
Use encore.dev/log for application events that need structured fields or trace integration. Preserve an application's established field names and event conventions when they are consistent with this guidance.
Decide what to record
Add logs for information Encore cannot infer from infrastructure operations:
- Business state transitions
- Decisions and the inputs that determined them
- Retries, fallbacks, and degraded operation
- Failures at the boundary that decides whether to retry, recover, or abort
- External-system interactions that require domain context
- Security-relevant or administrative actions
Encore already traces API requests, database queries, service calls, cache operations, and Pub/Sub activity. Do not narrate that execution with logs such as request started, querying database, or request completed.
Use metrics for aggregate counts and levels. Use traces for call flow and timing. Use logs for the context needed to investigate an individual event.
Emit structured logs
Keep the message stable and put variable data in fields:
import log from "encore.dev/log";
log.info("order rejected", {
event: "order.rejected",
orderId,
productId,
reason: "insufficient_inventory",
requestedQuantity,
availableQuantity,
});
Avoid interpolating identifiers into the message. Stable messages can be grouped, while fields remain searchable.
For events used by dashboards, alerts, or automation, include a stable event field. Prefer <domain>.<past-tense-event>, such as payment.authorized or subscription.cancelled. Diagnostic logs do not require an event name.
Keep each field name and type consistent across events. Prefer numeric values with explicit units and enum-like values over formatted prose:
log.info("payment authorized", {
event: "payment.authorized",
amountMinorUnits: 14900,
currency: "SEK",
});
Choose a level
error: an unexpected failure prevented an operation from completingwarn: the application recovered or continued in a degraded stateinfo: a meaningful, relatively low-volume business eventdebug: diagnostic detail not required for normal operationtrace: high-volume internal state
Expected outcomes such as validation failures, missing resources, rejected logins, or declined payments are not error events. Use their business meaning to choose a level.
The default minimum level is trace. Configure log_level in encore.app before adding debug or trace calls to frequently executed code.
Record errors once
error and warn accept the original error as their first argument:
try {
await chargeCustomer(customerId, amount);
} catch (err) {
log.error(err, "payment authorization failed", {
customerId,
amountMinorUnits: amount,
currency,
});
throw err;
}
Passing the original error preserves its type, message, stack trace, and cause. Converting it to a string discards that information.
Log a propagated failure at the layer that owns the recovery decision and has the domain context to describe its impact. A lower layer should log only when it handles or suppresses the error, retries, detects an invariant violation, or holds context that propagation will lose.
Add shared context
Use log.with() when several events from one domain operation share fields:
const logger = log.with({ importId, tenantId });
logger.info("user import completed", { succeededCount, failedCount });
logger.warn("user import row rejected", { rowNumber, reason: "invalid_email" });
Encore already attaches the service, endpoint, trace ID, and authenticated user ID to request logs. Do not add duplicate copies.
Protect sensitive data
Encore does not redact log fields. Never log credentials, tokens, session identifiers, cookies, authorization headers, secret configuration, payment details, or complete request and response objects.
Prefer internal identifiers over personal data. Use a stable non-reversible fingerprint when correlation requires a value that should not be stored.
The sensitive API option redacts request and response payloads from traces; it does not redact values passed to log.
Control volume
Summarize batches and long-running operations with outcomes, counts, and durations. Avoid one log per item unless each failure requires investigation.
Log selected metadata from large values rather than complete provider responses, records, configuration objects, arrays, or document bodies. Bound previews and review them for sensitive data before logging.
When reviewing an existing application, check loops, Pub/Sub subscribers, middleware, and frequently called helpers first. A small number of these call sites often produces most of the log volume.
See the Encore.ts logging guide and encore.dev/log API reference.