switch to using conda postgresql
greglandrum opened this issue · comments
Now that conda has postgresql 9.5 builds available for all platforms, I think we should consider switching the rdkit-postgresql recipe over to using those. This would simplify things, lessen the amount of work that we need to do, and possibly make things easier for users (since it's easy to imagine others starting to take advantage of the conda postgresql builds.
When we do that we can remove the postgresql and postgresql95 recipes along with the rdkit-postgresql95 recipe. Unless, of course, you see a reason to maintain support for older postgresql versions in the conda recipes.
I wasn't aware about the updated anaconda postgresql packages (at the time of the 9.5 release they were still providing postgres 9.1 for linux and osx alone).
I am a bit afraid that updating rdkit and postgres independently from each other and/or supporting multiple versions of postgres may become more difficult. The only reason for supporting older versions of postgresql is that users often can't just conda update from one version to another without migrating the deployed database.
On the other hand the potential simplification and the direct availability of windows packages are definitely interesting. Do you think it may be ok to assume that users deploying rdkit and postgres in a production environment could be skilled enough to eventually use their own recipes and conda channel?
I was also totally unaware of it (it's new) until it came up on the mailing list.
The version compatibility thing is a real concern. If we were to assume (and I have no idea how good of an assumption this is) that no one is currently running a production instance using the rdkit postgresql recipes then we can choose the current conda-supported postgresql version as a baseline and manage future version-incompatibilities as we come to them.
Seem reasonable?
in general, yes, it would sound quite reasonable to me. but one other thing I'm noticing is that the postgresql build from anaconda doesn't seem to install any other extensions apart pl/pgsql.
I am not sure about how much that would be limiting, but if it's not blocking, maybe we could switch to building the rdkit packages using the anaconda postgresql package, but leave the alternate recipes available for a while for users that may have stronger requirements? The best long term solution would probably consist in coordinating with the upstream distribution to solve the possible limitations without duplicating the effort.