[throwawayie6]

I made something similar once, but simply wrote the output to a temporary file, and then just renamed it to the final name when it was done

From what I remember, this was an atomic operation, and if there was a download in progress it would continue using the old file, because the data was still on disk and the filename was simply a pointer to the node.

This may behave differently on other file systems. This was done on an old Linux server with ext3

Seems like a simpler solution than using a db