I'm going to revert the default configuration change - this is a minumal, least risk approach to this issue.
Its possible that we might want to pick the rocksdb fixes as well but that diverges Ubuntu from upstream in a core part of ceph which I'm not 100% comfortable with.
Feedback on the upstream tracker bug indicates there is a lot of evidence that increasing concurrency has not resulted in data loss.
I'm going to revert the default configuration change - this is a minumal, least risk approach to this issue.
Its possible that we might want to pick the rocksdb fixes as well but that diverges Ubuntu from upstream in a core part of ceph which I'm not 100% comfortable with.
Feedback on the upstream tracker bug indicates there is a lot of evidence that increasing concurrency has not resulted in data loss.