ranflarion opened a new issue, #10493:
URL: https://github.com/apache/arrow-rs/issues/10493
**Describe the bug**
`AvroObjectReader`'s `spawn` / `spawn_stream` helpers convert every error
from the underlying `ObjectStore` with `AvroError::General(e.to_string())`
(arrow-avro/src/reader/async_reader/store.rs), and the header fetch loop wraps
its error the same way (`"Error fetching Avro header from file reader: {err}"`
in arrow-avro/src/reader/async_reader/builder.rs). The typed
`object_store::Error` is destroyed at that point: `err.source()` returns
nothing and there is no way to downcast back to `object_store::Error`.
This matters for callers implementing retry or error-classification logic on
top of the reader: a permanent `NotFound`/`PermissionDenied` is
indistinguishable from a transient transport fault without string matching. We
hit this while adding retry-on-transient-store-fault logic around an Avro scan
and could not avoid retrying 404s.
`ParquetObjectReader` already handles this correctly: its identical `spawn`
helper takes `E: Into<ParquetError>` and parquet provides `impl
From<object_store::Error> for ParquetError {
ParquetError::External(Box::new(e)) }` (parquet/src/errors.rs), so the full
source chain survives.
**To Reproduce**
```rust
let store = Arc::new(InMemory::new());
let mut reader = AvroObjectReader::new(store, Path::from("missing.avro"));
let err = reader.get_bytes(0..10).await.unwrap_err();
// err is AvroError::General("Object at location missing.avro not found:
...")
// err.source() is None; no downcast to object_store::Error is possible
```
**Expected behavior**
Align with `ParquetObjectReader`: add a feature-gated `impl
From<object_store::Error> for AvroError` mapping to
`AvroError::External(Box::new(e))`, change the private `spawn` bound to `E:
Into<AvroError>`, and let the header fetch propagate the error instead of
reformatting it. `AvroError::External` already reports its inner error via
`source()`, and `From<AvroError> for ArrowError` maps it through
`from_external_error`, so the chain then survives all the way to `ArrowError`.
**Additional context**
Error `Display` text changes for store failures (`External: ...` instead of
`Avro error: ...`); variant selection within the `#[non_exhaustive]`
`AvroError` is not a stability contract. Happy to submit a PR.
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
