S3 backend: unexpected errors during high volume of reading

[dantengsky]

while reading data from s3, expected error encountered:

ERROR 1105 (HY000): Code: 4000, displayText = other error (object error: (op: read, path: 1/29/_b/b7b39054dd59427babc4f0eae3af59b3_v0.parquet, source: sending request:  Error { kind: ProtocolViolation, context: None, source: Some(Error { description: "Server returned nothing (no headers, no data)", code: 52, extra: None }), source_type: Some("curl::error::Error"), local_addr: Some([172.31.7.127:53456](http://172.31.7.127:53456/)), remote_addr: Some([52.219.88.122:443](http://52.219.88.122:443/)) })) (while in processor thread 0).

during the above execution, there are thousands of HTTP connections, which seems to be one of the causes.

[Xuanwo]

ProtocolViolation means the server made an unrecoverable HTTP protocol violation. This indicates a bug in the server. Ideally, we should not retry this error because it's likely to produce the same error.

But services like s3 could recover from this service very soon. Retrying this error is still meaningful.

Let's do this.