MVP Error Tracking: 4 Decisions for Grouping, Search, and Alerting
The page says a Node.js SaaS catalog import has produced zero updates for 45 minutes. The on-call opens the alert and needs three things immediately: whether the Sentry alternative's error-tracking API received a failure, which shop and scheduled run stopped, and whether the run failed or never started. TL;DR: a simple error-tracking API is a reasonable Sentry alternative for one small SaaS app…
The Node.js SaaS catalog import has not produced any updates for 45 minutes, prompting the on-call to open an alert. The immediate requirements include verifying if the Sentry error-tracking API received a failure, identifying the affected shop and scheduled run that stopped, and determining whether the run failed or never started.
Essentially, a straightforward error-tracking API could serve as a viable alternative to Sentry for this small SaaS application when the primary focus is capturing exceptions and triaging them. This involves capturing an error, examining its grouped issue and event payload, searching through the error set, and resolving the group post-fix deployment.
However, it is not a replacement for Sentry's comprehensive notification routing, frontend debugging capabilities, exploration of distributed traces, source-map processing, crash symbolication, or Session Replay features. Crucially, an error tracker cannot report a run that generated no errors if it never ran at all. To address this, a heartbeat monitor like Healthchecks or the implementation of an explicit result signal, overseen by the poller, is essential.
This separation is a key decision. The real question is what should have triggered the zero-results page? Analyzing backward, a customer-facing query or a downstream freshness check might notice a lack of catalog changes, but this detection takes into account the schedule interval, job runtime, and any delay before an individual inspects the result.
Therefore, an earlier signal should encompass the outcome of a single run, encompassing stable run ID, tenant ID, scheduled time, completion time, result count, and status. If the worker encounters an error, error tracking records the exception and retains the event payload for subsequent triage. Conversely, if the worker fails to start, only a missing heartbeat alert will indicate the silence.
There are two distinct failure modes, and merging them into an alert for error_count = 0 creates a blind spot. A loud failure occurs when the import begins and raises an exception, which can be grouped and accompanied by event details that include the run and tenant context. Conversely, a silent failure arises when the scheduler, queue handoff, or worker fails to produce any outcome.
A deadline-based heartbeat must alert on the absence of a completion. The Service Level Objective (SLO) should reflect the user-visible outcome, not the tool itself, such as "scheduled imports produce a terminal outcome within the agreed completion window." The specific window should be outlined in the service contract and capacity model; it cannot be deduced from an error-tracking product.
Waking up at 45 minutes is reasonable only if the typical and high-percentile runtimes provide sufficient response time before the SLO is compromised. This is where a well-structured error dashboard may mislead an infrastructure lead. A zero new exception count could signify either zero failures or zero execution. It is crucial to instrument the outcome, not solely the exception.
In the scenario of an e-commerce importer, the same correlation fields (tenant_id, import_id, and run_id) should be attached to the schedule dispatch, worker completion, and captured exception. These fields facilitate cost and ownership attribution and enable trace_id and span_id correlation where available. However, merely having these fields does not constitute a distributed trace query experience or a span tree.
The result signal should be emitted once at the terminal boundary, after the importer has verified the number of accepted records. Avoid generating an exception for a legitimate zero-result run. Record the count and let the business rule determine whether zero is expected for that tenant and feed. Otherwise, a low-volume merchant may trigger recurring false pages.
Capacity planning is vital. A poller that individually searches every tenant increases query traffic based on the tenant count, while a batched control-plane check reduces calls but amplifies the impact of a broken poll cycle. The choice of either approach should be documented, along with exposing the poller's successful evaluation time.
An alerting process lacking a self-signal merely shifts the silent-failure problem. Sensitive payloads should also be handled with care. Error events are diagnostic records, not an additional customer database. To comply with deletion obligations, exclude or mask credentials, payment data, and raw personal information. OWASP's logging guidance serves as a useful baseline for deciding what must be excluded or masked.
The decision to employ Sentry or a simple error-tracking API for a Node.js SaaS largely depends on the specific requirements of the import workflow. The primary considerations include who owns notification delivery, the level of frontend context needed, and whether a missing run is treated as a first-class event. The optimal choice will vary depending on the team's needs, the required cost-attribution consequences, and the potential impact on the billing model.
Written by urgent.news from Dev.to's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.