Upgrades may not replace the rabbit.schema file

[lukebakken]

rabbitmq-users message: link

Reproduction steps:

• Install and start RabbitMQ
• Stop RabbitMQ, upgrade to newer version with different rabbit.schema file
• Start RabbitMQ

The old cuttlefish schema may be in use. See the 3.7.9 to 3.7.10 upgrade for an example.

[mohag]

A workaround seems to be to delete /var/lib/rabbitmq/schema/rabbit.schema and restart RabbitMQ. The /usr/lib/rabbitmq/bin/rabbitmq-server script will copy the newer version in. (I'm not sure if there might be other dependencies though)

The package scripts might be one place to correct it, if there is a reason to copy it instead of using it from the installation location.

(If changes are ever made that will break old configs, it might be necessary to keep the old schema and do something like providing a rabbitmq-update-conf-format (or more likely a rabbitmq-diagnostics subcommand) to let the user manually apply it?)

[lukebakken]

@dumbbell @michaelklishin - another fix would be to always copy the rabbit.schema file when RabbitMQ starts. The reason that may be a better fix that something in packaging is that we don't know if RABBITMQ_SCHEMA_DIR is set during install / upgrade, correct?

[michaelklishin]

Always copying the schema sounds reasonable. I don't see any major risks with that.

[mohag]

I'm not sure why ${RABBITMQ_SCHEMA_DIR} doesn't just default to ${RABBITMQ_HOME}/priv/schema/? Is this to allow to user to load additional schemas?

[lukebakken]

I don't know the history behind making the schema dir configurable 🤷‍♂️

[michaelklishin]

@mohag schema files also come from plugins, so your understanding is basically correct.

[michaelklishin]

@hairyhum has most context here

[michaelklishin]

I think copying the file and overriding the one that's already there is what most people would expect: we want to have the latest schema file. The overhead of this is negligible.