Upgrade plan from 1.6 to 1.10

[Mons]

Epic: Tarantool installation and upgrades

SME: @ sergepetrenko

We have a cluster of 1.6 with the following configuration:

• 7 nodes
• full mesh
• one rw, others ro
• rich application with background fibers and network operations (net.box & http)
• rw have background fibers modify data and work with network
• balancer, that routes request traffic to ro or rw depending on instance status and role (in fact, balancers are 1.10)

We know, that application works on 1.10 without any modifications due to compatibility layer, based on kit, config and spacer/v1 (from github.com/moonlibs).

We need a guide to perform live migration without downtime from 1.6 to 1.10, i.e. we can't stop writes or reads "for some time"

[Totktonada]

@Mons There is the instruction: https://www.tarantool.io/en/doc/1.10/book/admin/upgrades/

Are there something that it does not cover? Are you expect other actions from us aside revisiting of the instruction?

[ligurio]

Found issues: tarantool/tarantool#4948, tarantool/tarantool#4946

[ligurio]

Status: waiting for testing on stage

[ligurio]

Upgrade scenario step-by-step:

Before upgrade:

1. check cluster health, it should be ok before upgrade
2. execute box.info() and check that replicas in operation mode
3. check replication lag, it shouldn't be too high
4. check tarantool logs for errors on each replica

Upgrade

1. enable RPM/DEB repository for tarantool 1.10 and install tarantool package using package manager
2. replace box.replication_source to box.replication everywhere in used Lua applications
3. restart one of old instance v1.6 to run updated binary
4. make sure that restarted instance is still in a cluster and it is still in operation
5. repeat steps 3-4 for each replica in a cluster
6. restart master instance and execute box.schema.upgrade() on it
7. make sure cluster is healthy, there are no errors in logs and output of box.info() doesn't report any issues

After upgrade

1. stop one-by-one instances with old tarantool versions and each time and check that cluster continue works
2. success criteria: all instances has new tarantool version v1.10, replication lag ~= 0, no errors in logs

References:

• https://www.tarantool.io/en/doc/1.6/book/replication/
• https://www.tarantool.io/en/doc/1.10/book/admin/upgrades/
• https://www.tarantool.io/en/doc/1.10/whats_new/#version-1-10

[patiencedaur]

@ sergepetrenko has a lot of useful notes about it, which would make an excellent Habr article.

[Totktonada]

NB: https://github.com/tarantool/tarantool/wiki/Caveats-when-upgrading-from-tarantool-1.6

[Totktonada]

BTW, https://github.com/tarantool/tarantool/wiki/Upgrade-instructions collects all such instructions. Maybe it worth to mention it somewhere in the documentation.

[patiencedaur]

Related: #836