Analytics systems · observability POC

Finding the new error in a sea of repeats.

Our SDK diagnostics service already captured uncaught errors. Datadog now selects the genuine SDK events, turns them into native Error Tracking issues, measures each group, and alerts Slack when a new issue appears.

Datadog-native
No service deployment
Native grouping
New-issue alerts

The problem

Tracking every error is not the same as understanding them.

The existing diagnostics service gave us the raw signal, but finding a new SDK failure meant manually reading logs, comparing messages and stack traces, and remembering what we had already seen. The current Datadog-only flow turns that repeated analysis into a strict selection, normalization, native grouping, and notification pipeline.

Current system

The live dashboard is the source of truth.

The canonical population requires service:sdk-diagnostics, the exact "SDK Diagnostic Event" marker, and @event_name:sdk.error.uncaught. Together they reject request wrappers, backend Java exceptions, and unrelated diagnostics logs.

Open the dashboard ↗Native issues, occurrence counts, and percentages Open the new-issue monitor ↗Slack alert with error context and a direct issue link

The mechanism

One event. One stable identity.

A canonical log is normalized before it becomes an issue. Datadog Error Tracking owns the fingerprint and native grouping; the dashboard reports each group’s occurrence and share.

01
Select
Three required signals isolate genuine Browser SDK uncaught-error diagnostic events.
02
Normalize
The log pipeline maps nested values to error.kind, error.message, error.stack, and status:error.
03
Group
Datadog Error Tracking creates the native fingerprint and groups equivalent occurrences into one issue.
04
Measure
The dashboard ranks groups by occurrence and percentage; generic Failed to fetch network noise is excluded.
05
Notify
A genuinely new issue group triggers a contextual Slack message with a direct Datadog issue link.

Interactive explainer

Follow the fingerprint.

Open full screen ↗Watch the demo ↗

Two-minute walkthrough

See the real path.

The recording is a historical walkthrough. For current queries, grouping, counts, and alert configuration, use the live dashboard and monitor above.

The takeaway

Grouping is the product. Notification is the final step.

Without reliable native grouping, Slack only makes noise arrive faster. With strict selection and Error Tracking issues, each new alert is configured to represent a new SDK problem worth investigating.