feat(server): accept minidumps in a gzip container #4029
Conversation
| @@ -97,7 +120,11 @@ async fn extract_multipart( | |||
| minidump_item.set_payload(ContentType::Minidump, embedded); | |||
| } | |||
|
|
|||
| validate_minidump(&minidump_item.payload())?; | |||
| if validate_minidump(&minidump_item.payload()).is_err() { | |||
There was a problem hiding this comment.
Small suggestion: Similar to how we check for embedded minidumps, could we invert the logic slightly to fully separate concerns? Right now, validate_gzip_minidump fails with an error "invalid minidump" where actually it tests for a gzip header.
I'm thinking about control flow like this:
- Check for a compression header
- If there is compression, then try to decompress and propagate all errors
- Replace the payload bytes with the decompressed bytes
- Now check the minidump header once.
There was a problem hiding this comment.
Yes, of course, this makes sense. I was only prepping a PoC to check for basic warning signs and whether you want automatic detection/decoding in the endpoint. I am also unhappy with the gzip being the fallback to the typical case 😅.
Even better.
Yes, definitely. I guess |
|
Just so you know, I am working on an escalating Android issue and will revisit this when I have more time. |
During the work on native assigned documentation issues, the topic of compressing minidumps when uploading them via
curldirectly to the minidump endpoint came up.This is my attempt at documenting the current process: https://github.com/getsentry/sentry-docs/pull/11304/files.
Uploading a compressed minidump without meta-data is relatively straightforward. You use a gzipped minidump as the sole
POSTcontent, provide the minidump content type, and define it as gzip-encoded. Axum transparently decodes the minidump before they even reach the relay endpoint, and you can process them if you are oblivious to the encoding.Unfortunately, due to the way encoding can't be supplied per multipart field (at least not RFC conforming) if a user wanted to supply meta-data with the minidump, they would have to construct the entire multipart content manually, which is error-prone, compress that whole body, and send it again as a single data request.
Curl doesn't provide any convenience for gzip encoding arbitrary POST content.
This simple change in Relay would allow users to make no change between their curl requests (they could use simple curl field parameters as explained in docs) besides uploading a gzipped minidump in place of a plain one.
Let me know if this would be an acceptable change compared to adding the documentation, which describes a severely manual and, thus, error-prone process.