proper handling of fsync and EIO

[MarkCallaghan]

PostgreSQL, MyRocks and InnoDB have bugs related to this, but RocksDB is OK.
Is the right thing done by WiredTiger and the replication log

https://www.postgresql.org/message-id/flat/CAMsr%2BYHh%2B5Oq4xziwwoEfhoTZgr07vdGG%2Bhu%3D1adXx59aTeaoQ%40mail.gmail.com#CAMsr+YHh+5Oq4xziwwoEfhoTZgr07vdGG+hu=1adXx5...@mail.gmail.com

LWN has a story, but it is subscriber only until next week - https://lwn.net/Articles/752063/

[Alex Gorrod]

Thanks for the heads up Mark - WiredTiger does retry fsync calls, so may have the same issue. I've opened https://jira.mongodb.org/browse/WT-4045 to track the diagnosis.