You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
Feedback summary cache misses every hour for relative date ranges #131185
Found by reading the source at 7d94848 and by running the endpoint's get() with the database, the cache and the Seer call stubbed; not seen in a live sentry.io session.
Open User Feedback with a relative date filter, such as the default "Last 14 days". useFeedbackSummary sends statsPeriod=14d to /organizations/{org}/feedback-summary/.
The key changes every hour, so the first request in each new hour is a cache miss. Each miss sends up to 1,000 feedbacks or 1,000,000 characters to Seer's summarize endpoint. The one-day TTL is not used after the first hour.
One request per hour for 24 hours, same organization and projects (OrganizationFeedbackSummaryEndpoint.get() at 7d94848; Group query, get_projects, cache and get_summary_from_seer stubbed; clock moved with time_machine):
statsPeriod
Seer summary calls, master
With the categories key logic
14d (page default)
24
1
7d (endpoint default)
24
1
24h
24
24 (hour key kept)
The hour key came from the review of #93461, so that a short range such as the last hour gets its own summary. A fix that copies the categories logic (hour key below 2 days, day key from 2 days up) keeps that and is ready as a PR. The trade-off is the one the categories endpoint already makes: for ranges of 2 days or more, a summary can be up to a day old.
Activity
getsantry commentedon Nov 15, 2026
Assigning to @getsentry/support for routing ⏲️
linear-code commentedon Nov 15, 2026
ENG-9046