Background
Sentry's ingestion pipeline was built & optimized for event-like data, i.e. a high volume of small submissions. We have been raising limits on ingestion endpoints (currently 200 MB) but the fundamental limitation remains because the code assumes that we can keep in-flight requests in memory.
There are several use cases that require larger attachments than the current limits allow, roughly divided into two categories:
- Data that the user wants to see attached to their events. These require little to no processing ("plain attachments").
- Large crash reports such as minidumps. These are processed server-side to get meaningful stack traces for crashes.
The first category is easiest to implement because we can stream data into storage with relatively few steps. The second category is problematic because it requires the full ingestion & processing pipeline to be able to handle large files.
Milestones
P1 - EA by mid 2026
- Allow uploading plain attachments up to 1 GB.
- Allow uploading minidumps and apple crash reports up to 500 MB. Reports larger than 200 MB will not be PII-scrubbed so this feature will be opt-in.
P2 - EA by end of 2026
- Allow uploading plain attachments up to 10 GB.
P3 - No timeline
- Streaming Processing. Allow uploading minidumps and other attachments up to 10 GB while still supporting processing features like data scrubbing and symbolication. It is unclear whether we will ever support this since it requires a large change in our processing architecture.
References
Internal project: Linear
Background
Sentry's ingestion pipeline was built & optimized for event-like data, i.e. a high volume of small submissions. We have been raising limits on ingestion endpoints (currently 200 MB) but the fundamental limitation remains because the code assumes that we can keep in-flight requests in memory.
There are several use cases that require larger attachments than the current limits allow, roughly divided into two categories:
The first category is easiest to implement because we can stream data into storage with relatively few steps. The second category is problematic because it requires the full ingestion & processing pipeline to be able to handle large files.
Milestones
P1 - EA by mid 2026
P2 - EA by end of 2026
P3 - No timeline
References
Internal project: Linear