feat(disk_balance): add do_disk_migrate_replica to support migrate origin data

[foreverneverer]

1. support disk migration rpc request(RPC_REPLICA_DISK_MIGRATE) and check whether the request is valid feat(disk_balance): support and validate disk migration rpc #660.
2. add do_disk_migrate_replica to support migrate origin data
3. close origin replica and update replica dir feat(disk_balance): close origin replica and update replica dir #668

replica migration status

   client(shell)----------->replicaServer--------->metaServer
         |                      |                     |
         |------- start ------->|----> IDLE           |
         |                      |      | (validate)   |
         |                      |     MOVING          |
         |                      |      | (copy data)  |
         |                      |     MOVED           |     
         |                      |      | (rename dir) |
         |                      |     CLOSED          |
         |                      |      |              |
         |          ------------|<- Learning <--------|
         |         |            |                     |
         |  LearningSucess      |                     |
         |         |            |                     |
         |          ----------->|                     |

#660 complete the migration request args check, this pr add do_disk_migrate_replica and support migrate origin data. The migration status from MOVING to MOVED

[neverchanje]

This entire execution spends too long, it will probably block other tasks on the REPLICATION_LONG thread pool.

[foreverneverer]

right, but the process need spend the long time, if not allow block REPLICATION_LONG, put DEFAULT?

[neverchanje]

So how does the user know the result that failure happened?

[foreverneverer]

copy data is designed async and no response the status, the result now need check the log

[foreverneverer]

or I can add query_status rpc in later pr?

[hycdong]

I agree with @neverchanje, there should be a way to let user notice failure happened, query_status rpc may be helpful.

[foreverneverer]

I can add it in later pr

[neverchanje]

So what is your plan? You certainly can not actively notify the user. One possible way is to save the error, and every time user queries the status, if failure happened, you respond with that error.

[foreverneverer]

Right, now has one way is run disk-migrate -g gpid -n node(not add other argument), if has one task is running, it will show:

 Internal server error [DiskMigrate to session 127.0.0.1:45801(replica)] failed: ErrorCode({Errno:ERR_BUSY}):Existed migrate task(replication::disk_migration_status::MOVED) is running]

We can get the task process but not need query_status, I think we can use this first, if need query, I can add it in later pr

[neverchanje]

Are _target_replica_dir = fmt::format("{}{}", replica_dir, kReplicaDirTempSuffix); and _target_data_dir = utils::filesystem::path_combine(_target_replica_dir, kDataDirFolder); the same?

[foreverneverer]

filesystem::path_combine("/abc", "file") => /abc/file, but format("/abc", "file")=>/abcfile

[hycdong]

When will this condition happen? could you please take an example for me?

[foreverneverer]

Now hasn't this condition, add it only make sure the process must under MOVING.

[hycdong]

OK, I suggest that you can consider this condition in your further pull request.

[hycdong]

I agree with @neverchanje, there should be a way to let user notice failure happened, query_status rpc may be helpful.