Skip to content

find_dropped_events doesn't say that buckets snap to whole intervals #2276

Open

Description

@sentry-junior

Pattern: find_dropped_events with start/end and a coarse interval returns full-interval volumes for the first and last buckets, not volumes for the requested window. Nothing in the tool output or the description says so. Callers that sum the buckets for a fixed window overcount.

Reports: 1 (Sentry MCP user feedback MCP-SERVER-GQD, python caller, server 0.42.0). The caller worked around it with interval=1h and told its user about the alignment limit.

What happens

  1. The caller passes start=<…T18:01Z>, end=<…T06:01Z>, interval=1d. The tool forwards them unchanged to GET /organizations/{org}/events-dropped/ (sampled span for this caller's call).
  2. Sentry builds the outcomes query through QueryDefinition → get_constrained_date_range (getsentry/sentry src/sentry/snuba/sessions_v2.py:451-487), which calls to_intervals (src/sentry/snuba/metrics/utils.py:414-449). That floors start to the interval and rounds end up to the next boundary. With 1d, the range becomes <day>T00:00Z to <last day + 1>T00:00Z. So the first and last buckets hold whole-day totals.
  3. The endpoint's meta.start / meta.end report the requested range, not the aligned one (src/sentry/api/endpoints/organization_events_dropped.py:233-234). The MCP tool doesn't return meta.start/meta.end at all (packages/mcp-core/src/tools/catalog/find-dropped-events.ts:212-217). Only each bucket's own start/end gives this away.
  4. The description (find-dropped-events.ts:83-84, :128-137) tells callers to "use '1d' and sum the buckets" for longer ranges, without saying that the first and last buckets can cover time outside the window.

Suggested fix (MCP side)

  • When start/end don't fall on interval boundaries, add a note that gives the aligned range that was counted and says the edge buckets are whole intervals. Suggest a finer interval (for example 1h) for exact windows.
  • Add one line to the interval hint saying buckets align to whole intervals in UTC.

Possible Sentry API follow-up (not filed): meta.start/meta.end could report the aligned range that was actually queried. Owner: whoever owns organization_events_dropped.py. The MCP note above doesn't depend on that change.

Not this issue: the same report's search_errors time-range problem is #2146, and its explanation mismatch is #2147.

via automation evt_905b9c478746fd9deeb853f85ad7441b · Run ID dispatch_ebd17b42670133526c625ae4ebab5254.

--

View Session [Sentry]

Activity

  1. linear-code commented on Nov 15, 2026

    @linear-code
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    MCPSentry MCP tools, transports, and documentationmcp-feedback-pattern

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions