Write to UFS when failed to write to Alluxio Worker

[JiamingMai]

1. Use addSuppressed instead of creating a new exception to throw the original exception.
2. Go ahead to write data to UFS when it encountered exception during the time writing data to Alluxio Worker.

[huanghua78]

Please add comments here:

    //writing to worker over netty failed. But Alluxio client continues writing to UFS.

[huanghua78]

Should we abort trying to write to worker for the following requests? And delete already written pages from worker cache.

The reasons are:

1. The subsequent write request might fail, too.
2. The worker will receive some data but there is already a hole (from the previous failed request). I am not sure if the worker/cache manager will handle this. I am afraid not. The cache manager on worker expects contiguous write. The write request does not have a POSITION for current data.
3. suppose for a 1MB page, the alluxio client has written 4K data at [0, 4096). And then failure for next [4096, 8192) write. And then a successful write of 4K at [8192, 12288). So, the page file on worker will have a hole, not filled with actually data. Subsequent reading from paging store will get all ZEROES for such hole. This is not correct.

[JiamingMai]

I submitted an issue #18157 to make this be a future work.

[JiamingMai]

alluxio-bot, merge this please