How events are received

What the ingest accepts from the Sentry SDKs, the size limits, what it keeps and drops, and how it answers when a limit is reached.

Last updated 19 November 2026View as Markdown

The SDKs send to https://errors.<domain>/api/<project number>/envelope/ (and the older store/ endpoint), and browsers post CSP reports to …/security/. Events are stored on Deutsche Cloud's own machines, in the project's region.

Authentication

The SDK names its DSN key in the X-Sentry-Auth or Authorization: Sentry … header, as sentry_key in the query string, or in the envelope's own header (how a tunnel forwards it).

Status Message When
401 missing sentry_key No key in any of those places
401 unknown or revoked DSN key A key that does not exist or was revoked
403 the DSN key does not belong to this project The key is of another project than the URL names

What is received

Item What happens
event (errors, messages) Grouped into issues, stored for the project's retention
attachment with its event in the same envelope Stored with the event; one without its event, or over the size limit, is not kept
client_report Counted on the Stats tab as dropped by the SDK, with its reasons
transaction with its spans Stored for the project's retention: the trace view and the Transactions tab
session, sessions Release health, kept 90 days (a Node.js server's process-wide session is counted, not kept)
profile, profile_chunk Profiles, kept for the project's retention and at most 30 days
A CSP report Each violation becomes an error event; other report types are not kept
span (span streaming), replays, check_in, logs, feedback, metrics Not supported yet: answered with a one-hour rate limit for that category, so the SDK stops sending it

Bodies may be gzip, deflate, brotli or zstd. An envelope holds at most 1000 items. With the platform's default settings, an event or a transaction may be up to 1 MiB, an attachment up to 1 MiB and a request up to 20 MiB decompressed. A larger transaction or attachment is dropped without an error and counted on the Stats tab; the other refusals are answered:

Status Message When
400 invalid envelope: more than 1000 items An envelope of more than 1000 items
413 request body too large, or envelope too large A request over 20 MiB, sent or decompressed
413 event larger than 1 MiB An error event over 1 MiB
415 unsupported content encoding Another Content-Encoding

What is changed on the way in

  • With Keep client IP addresses off (the default), the sender's address is removed from the event and from its request headers; with it on, it is kept when the SDK asks for it ({{auto}}).
  • A transaction's request cookies, credential headers (Authorization, Cookie and similar) and query strings, in URLs and span data, are removed.
  • Stack frames are source-mapped with the uploaded maps (see Upload source maps).

An error event's other data is stored as sent: scrub what must not be stored in the SDK's beforeSend, and keep sendDefaultPii: false.

Limits and how they are answered

A refused request is answered 429 "rate limited" with Retry-After and X-Sentry-Rate-Limits naming the refused categories, which the SDKs honour by pausing those categories. A refusal of transactions in an envelope that also carries an error event, or of profiles in an envelope that carries anything else, is not a 429: the rest is taken, and the refusal rides on the 200 answer as an X-Sentry-Rate-Limits entry.

Limit Refuses For
The project's rate limit errors the rest of the minute
A key's rate limit errors, and separately transactions the rest of the minute
errors.events-per-minute errors the rest of the minute
errors.events-per-day errors a minute at a time, until the UTC day ends
errors.events-per-month errors a minute at a time, until the month ends or the quota is raised
errors.transactions-per-minute, …-per-month transactions only as above
errors.profiles-per-minute, …-per-month profiles only as above
errors.stored-bytes every category a minute at a time, until retention frees space
The event database unavailable every category ten seconds at a time, until it is back

Per-minute limits are counted on each ingest machine separately, so with more than one machine the effective ceiling is higher than the setting. When the ingest is busy, an item waits up to two seconds for room. An error event that still finds none is answered 429 with a one-second Retry-After, and so is a transaction sent without an error event. A transaction sent with an error event, and profiles and sessions, that find no room were already acknowledged: they are counted as lost on the Stats tab.

Every outcome is counted on the project's Stats tab with its reason: accepted, refused, not kept, lost after acceptance, written late, or dropped by the SDK (see What happened to each event, and why). An alert rule Events are refused or lost tells you when it happens (see Create an alert rule).

Did this page help you?