Upgrading the crud module

Compatibility limitation during upgrade

Starting from crud 1.7.0, the module automatically switches the cluster to safe mode when vshard rebalancing starts. Safe mode prevents writing data to an incorrect replica set during rebalancing.

The following safe mode specifics should be taken into account:

• safe mode is enabled independently on each storage instance participating in rebalancing;
• safe mode is not disabled automatically after rebalancing completes and must be disabled explicitly.

To check whether safe mode is enabled on a storage instance:

tarantool> _crud.rebalance_safe_mode_status()

or via the tnt_crud_storage_safe_mode_enabled metric.

To return to normal mode after rebalancing:

1. Wait until bucket migration (vshard rebalancing) is complete.

2. On each router, clear the CRUD routing cache:

   crud.rebalance.router_cache_clear()
   

3. On each storage instance, disable safe mode via the _crud Storage API. The command must be executed on all storage instances, including replicas:

   _crud.rebalance_safe_mode_disable()
   

Requests to spaces that use the vinyl engine are always executed in safe mode regardless of the current safe mode status.

When upgrading from crud versions below 1.7.0 to 1.7.0 and later, the following compatibility limitation applies:

In CRUD 1.7.0–1.7.4, storages require bucket_id. Therefore, get, update, and delete operations sent by a router running CRUD below 1.7.0 fail with an error. Other operations (insert, replace, upsert, select, count) work normally.

Starting from CRUD 1.7.5, storages support compatibility mode with routers below 1.7.0. If bucket_id is missing, the storage executes the operation without bucket ref, writes a warning to the log, and, if the metrics module is installed, also increments tnt_crud_storage_nil_bucket_id_compat_total.

On storages running CRUD 1.7.5+, while not all routers have been upgraded to CRUD 1.7.0 or later, get, update, and delete operations are executed in compatibility mode with reduced rebalancing safety. Storages write a warning to the log and, if the metrics module is installed, also increment tnt_crud_storage_nil_bucket_id_compat_total.

Upgrade order

When upgrading crud, follow this order:

When upgrading crud to versions below 1.7.0 or to versions starting from 1.7.5, use the standard upgrade order — first storages, then routers:

1. Upgrade storage replica nodes
2. Upgrade storage master nodes
3. Upgrade routers

When upgrading through crud 1.7.0–1.7.4, upgrade routers first and then storages. Do not run rebalancing until the upgrade is complete:

1. Upgrade routers
2. Upgrade storage replica nodes
3. Upgrade storage master nodes

For details on the standard cluster upgrade order, see the corresponding section.