Re: Re: In-core regression tests for replication, cascading, archiving, PITR, etc.

Michael Paquier wrote:

> - Manage node information using package/class PostgresNode.pm and have
> RecoveryTest use it. I have actually made PostgresNode bare-bone and simple
> on purpose: one can initialize the node, append configuration parameters to
> it and manage it through start/stop/restart (we may want to add reload and
> promote actually if needed).

This looks great as a starting point. I think we should make TestLib
depend on PostgresNode instead of the other way around. I will have a
look at that (I realize this means messing with the existing tests).

> I have also arrived at the conclusion that it is not really worth
> adding a node status flag in PostgresNode because the port number
> saved there is sufficient when doing free port lookup, and the list of
> nodes used in a recovery test are saved in an array.

I don't disagree with this in principle, but I think the design that you
get a new PostgresNode object by calling get_free_port is strange. I
think the port lookup code should be part of either TestLib or
PostgresNode, not RecoveryTest.

> - Add new module RecursiveCopy to be used for base backups. This removes
> the dependency with Archive::Tar. PostgresNode makes use of that when
> initializing a node from a backup.

Great.

> - Tests have been updated to use the PostgresNode objects instead of the
> port number as identifier. That's more portable.

Makes sense.

--
Álvaro Herrera http://www.2ndQuadrant.com/
PostgreSQL Development, 24x7 Support, Remote DBA, Training & Services