When Alibaba Cloud schedules a migration event for your ApsaraDB for Redis instance, the system moves the instance to new server nodes to upgrade software, hardware, or network facilities. This page answers common questions about how migration events work and what actions you can take.
What happens during a migration event
A migration event has four stages:
| Stage | What happens | Impact |
|---|---|---|
| Task creation | Alibaba Cloud notifies you by text message, phone call, email, or internal message. | None. |
| Start time | The system provisions new resources and begins synchronizing data. | You can use the database, but cannot perform instance-level operations (such as changing configurations or migrating across zones). The scheduled switchover time cannot be changed at this stage. |
| Scheduled switchover time | The system cuts over your workloads to the new instance. | The instance experiences a transient connection interruption and enters a read-only state for up to 30 seconds. If the instance uses the cluster or read/write splitting architecture, each shard experiences a transient connection and stays in the read-only state for up to 30 seconds individually. Data Management (DMS) and Data Transmission Service (DTS) are temporarily affected and return to normal after the switchover completes. |
| Completion | The task is removed from the scheduled events list. | None. The zone, account, network, and endpoint are unaffected. |
Prepare your application for the switchover
During the switchover, the instance may drop connections for up to 30 seconds. Configure your application to automatically reconnect after a connection drop. To reduce business impact, reschedule the switchover to off-peak hours before the migration starts. See View and manage scheduled events.
FAQ about timing
How much notice do I get before a migration?
At least 40 hours in advance. After the scheduled switchover time is set, the system performs the switchover within the upcoming maintenance window.
How do I change the scheduled switchover time?
Change it in the console or by calling ModifyActiveOperationTasks. For step-by-step instructions, see View and manage scheduled events.
Why can't I change the switchover time?
If the migration has already started (the task has passed its start time), the scheduled switchover time is locked and cannot be changed.
Can I estimate when a migration will finish?
No. Only the start time is shown. The duration depends on factors such as network latency, task queue length, and data size.
FAQ about migration behavior
Why does Alibaba Cloud migrate instances?
To improve service performance and stability. The system migrates instances to new server nodes to upgrade software, hardware, and network facilities.
Are instances migrated to a different zone?
No. Migration stays within the current zone.
Can I cancel a migration task?
No. Migration tasks run at high urgency and cannot be cancelled. You can reschedule the switchover to a more convenient time, such as off-peak hours.
How are shards and proxy nodes switched over in a cluster or read/write splitting instance?
Data shards are switched over in parallel.
Proxy nodes are switched over in parallel, with a maximum of half the proxy nodes switched at the same time.
What happens if the switchover fails?
Within 1 minute of the switchover attempt, the system checks whether it succeeded. If it failed, the system waits briefly and then re-initiates the switchover automatically.
How do I know when a migration is complete?
The task disappears from the scheduled events list. To receive automatic notifications when migration tasks complete, set up alert rules in Cloud Monitor. For details, see Alert settings.
Why does my instance go through two switchovers?
Most migration tasks require only one switchover at the scheduled time. However, if your instance is deployed across two zones, a second switchover occurs within 120 minutes of the first. The second switchover moves the master node back to its original primary zone, keeping the instance co-located with other services it interacts with. No action is required. The interval between the two switchovers depends on how quickly the new replica node is rebuilt after the first switchover.