Blobstore can throw FileAlreadyExistsException

[rymsha]

With a chance of nodes to be created with exactly the same blob (and it does not already exist) - there is a race condition

if (file.exists) -> return false for both threads/instances
----------------> race is here
file.write -> both threads/instances want to write.

one of the nodes won't be created.

There is another race: when one thread creates blob - and writes data to it, while another sees the blob, and tries to read contents of it, there is a chance that contents are not yet fully written - because file writes are not atomic and we don't use write-then-copy approach in FileBlobStore. This leads to broken blobs cached.

[rymsha]

closed by mistake