Currently our deployment of backend looks like this:

./bin/install
./bin/scripts/server

Under the hood it calls v2-build, cabal install, v2-test, haddock and finally v2-run.

• https://gitlab.iscpif.fr/gargantext/haskell-gargantext/blob/d362b468db3373972c5ef0ba24413c577355bdbd/bin/install
• https://gitlab.iscpif.fr/gargantext/haskell-gargantext/blob/d362b468db3373972c5ef0ba24413c577355bdbd/bin/scripts/server

Couple of issues:

1. Do we need haddock at all? I don't know if it's served anywhere. We can check if haddock compiles on the CI, not use the servers to do that?
2. Do we need cabal install? It seems to break template haskell with git (haskell-gargantext!422 (comment 14571)) and I don't see any point of recompiling again and installing just to run via cabal v2-run later.
3. Could we just ship the CI binaries to servers, without recompiling? I imagine something like this: git pull to get nix deps correctly. Then just nix-shell and run the CI compiled binary that is associated with current git rev (of course this should be scripted). I don't propose docker not to introduce new tech.
4. In fact I'm not even sure how many recompilations are there: we v2-build with -O2 options, but then run (possibly) without them...