danielcweeks commented on PR #12868: URL: https://github.com/apache/iceberg/pull/12868#issuecomment-3164904270
I don't think this is necessarily the right way to tackle this problem. As @nastra pointed out, this is somewhat specific to the S3FileIO implementation (and other implementations) that share underlying pools, but closing one client closes them all. In this case I think it's related to how we configure the S3 Client in the S3FileIO. There is a rather nuanced issue with how the AWS SDK v2 treats the http client. If you use the `awsClientBuilder.httpClientBuilder(apacheHttpClientBuilder)` as we do, then the S3Client will treat the client as an owned client an close the underlying resources when `S3Client::close()` is called. However, there is different behavior if you call `awsClientBuilder.httpClient(<client>)`, which treats the client as a shared resource and does not close it. This surprisingly nuanced behavior, but I think what we need to do is associate the http client with something higher-level (table?) so that it closes with the appropriate resources. -- 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] --------------------------------------------------------------------- To unsubscribe, e-mail: [email protected] For additional commands, e-mail: [email protected]
