Remote Clone Flow

In remote clone, client and donor are separate instances communicating through network.

The clone plugin defines and uses a RPC protocol with several COM_ commands. The storage engine interface does not deal with them directly, the map to the same handlerton API as the local clone which does not use the RPC protocol.

Starting the Clone

1. The client plugin establishes a regular MySQL protocol connection to the donor, then sends COM_CLONE MySQL protocol command to switch to the clone protocol.
2. The client plugin calls clone_apply_begin(HA_CLONE_MODE_VERSION). The storage engine creates a locator that does not point to any snapshot but contains version negotiation information instead.
3. The client plugin sends COM_INIT to the donor with the locator payload.
4. The donor plugin calls clone_begin(HA_CLONE_MODE_START). It checks the version information in the passed version locator. Then it creates a snapshot and prepares an actual clone locator. All subsequent flow uses this locator.
5. The donor plugin sends COM_RES_LOCS with the locators as the response to the previous COM_INIT.
6. The client plugin calls clone_apply_begin(HA_CLONE_MODE_START).

Data Copy

1. The client plugin sends COM_EXECUTE to the donor which asks to stream the data.
2. The donor plugin calls clone_copy (same as the local clone), whose output is then serialized into two responses:
   1. COM_RES_DATA_DESC for metadata
   2. COM_RES_DATA for the data itself
3. The client plugin passes the two response payload to clone_apply, which is then handled the same as for the local clone, with the additional requirement that any application of the memory data is acknowledge by sending COM_ACK to the donor.

Finishing the Clone

1. After donor completes sending all the data, the sends a COM_RES_COMPLETE packet.
2. The client plugin calls clone_apply_end.
3. The client plugin sends COM_EXIT to the donor.
4. The donor calls clone_end.