Monitoring Error Tracking

See what a project received in the last 30 days and what happened to it, what its SDKs dropped, and what your alerts sent.

Last updated 19 November 2026View as Markdown

Replace project-id with the project's id.

What a project received

Prerequisites: errors:ListProjects on the project (without it, the project is not found).

  1. Open the project, then the Stats tab.

Error events per day charts the last 30 days (UTC) by outcome. Last 30 days has a row per kind of data (error events, attachments, transactions, spans, sessions, profiles and the kinds not supported yet) with a column per outcome, and the reasons, with their counts, in its Note.

What happened to each event, and why

Outcome What it means Reasons
Accepted Stored; Size is what it took
Refused Not stored; the request is answered 429 or another 4xx, unless the rest of its envelope was taken Over the organization's, the project's or the DSN key's per-minute limit; over the daily cap, the monthly cap or the stored-data cap; our database was unavailable, our storage was full or our ingest was busy (the SDK sends again); too large; unreadable; no DSN key, an unknown DSN key, a revoked DSN key or an inactive project or organization, a DSN key of another project; outside the accepted time window
Not kept Arrived, and is not stored by design Not supported yet; an attachment over the size limit or without its event; a server's process-wide session
Lost Acknowledged, then not stored: our fault, and it should always be zero Our database rejected it; unreadable; our ingest was busy (a transaction sent with an error event, a profile or sessions that found no room)
Written late An error event stored after waiting 10 seconds or more for our database, or written after an ingest restart; also counted under Accepted Our database was unavailable; our database rejected it at first; our ingest restarted
Dropped by the SDK What the SDKs report they left out The SDK's own, in its words, such as sample rate, before send, ratelimit backoff, queue overflow or network error

An error event that was acknowledged is kept through a restart of the ingest: what the database could not take before the stop is kept on disk and written when the ingest is back, and counted as Written late. Profiled milliseconds are not a row: the Profiles and Profile chunks rows count what arrived.

A rise in refused events means a limit is biting (see Quotas); a rise in dropped by the SDK means its sampling or beforeSend is. To hear about refused or lost events, create an alert rule Events are refused or lost (see Create an alert rule).

The Projects page shows each project's Unresolved issues and its accepted error events of the last 24 hours (Events, 24 h); dcloud errors projects list gives the same.

What your alerts sent

The Settings tab's Recently fired lists the last 50 firings of the project's alert rules, tests included, with what was sent ("sent: emailed 2; slack webhook 200", or why it failed). It needs errors:ListAlertRules on the project; dcloud errors alert-rules list gives the same.

Usage and cost

Error events, spans, profile hours and stored bytes are metered and appear under Billing. Error events and stored bytes are charged; spans and profile hours are metered but not charged yet (see What is Error Tracking?).

Did this page help you?