sqlite: implement backup api #17
Conversation
Signed-off-by: David Crawshaw <crawshaw@tailscale.com>
raggi
force-pushed
the
crawshaw/backup
branch
from
208ec67
to
7a5c4d2
Compare
|
This is not safe as-is, sqlite itself does not lock results, which the backup API also uses. We will need to add a lock shared between the backup context and the connection context that serializes calls to step and query/result functions. I tested THREADSAFE=1 to see if that would do the work for us, it does not. |
It is likely useful to report progress information and offer abort for large databases, so expose this in the high level API in a somewhat safe way. A lock was not introduced into the driver directly at this time because doing so is much more complex, particularly as Stmt.DBHandle would need to recover that lock - essnetially requiring global state with further synchronization.
raggi
force-pushed
the
crawshaw/backup
branch
from
7ce3b4e
to
54958c6
Compare
|
TODO: test if we can perform online backups of two schemas concurrently At time of writing control has two databases attached in the writable connection, main and noise. Ideally to backup both efficiently using the online backup API we'd be able to start a backup of both of these schemas and Step() them concurrently, and Finish() them at the same time, so the backups are in sync. It's not immediately clear if this is allowed in the backup API, so we need to test it. |
Signed-off-by: David Crawshaw crawshaw@tailscale.com