WIP: Add ability to define custom entries in pg_ident.conf

[dineshba]

Trying to fix #633

Tasks:

• Able to give pg_ident content in clusterSpec
• Update pg_ident.conf file whenever there is a change in pg_ident in clusterSpec

How to use in clusterSpec:

stolonctl update -p '{"pgIdent": {"omicron": [{"systemUsername": "bryanh", "databaseUsername": "bryanh"}, {"systemUsername": "robert", "databaseUsername": "bob"}]}}'

Note: Raising WIP PR to get faster feedback on the approach.

[sgotti]

@dineshba Thanks for the PR!

• Is it ready for review?
• Do you think we need some integration tests for this?
• We should probably document the new parameter in the cluster spec.

[dineshba]

Hi @sgotti

It is better to add some integration tests. So I will add integration test and update the document about the new parameter. Thanks for the suggestion

So, it is not ready for review. I will mark the PR as WIP.

[porty]

Entries will likely be written out in an undefined order (https://go.dev/blog/maps#iteration-order) - sorting the entries before writing them out might be beneficial

[porty]

f is an os.File which doesn't guarantee full writes

[porty]

I found that attempting to replace an entry for one of the ident maps fails due to apimachinery unable to merge map entries (or something along those lines). Attempting to do so gives this error:

$ bin/stolonctl update \
    --cluster-name stolon-cluster \
    --store-backend=etcdv3 \
    -p '{ "pgIdent": { "map1": [{ "systemUsername": "sysuser1", "databaseUsername": "dbuser1" }, { "systemUsername": "sysuser2", "databaseUsername": "dbuser2" }] } }'
failed to patch cluster spec: failed to merge patch cluster spec: merging an object in json but data type is not struct, instead is: map

(where map1 already contained some entries).

Overall I think structuring the pg_ident.conf entries in to a nice JSON representation, only to put them back as lines in a file is overkill, and confusing - I've found that using []string instead (similar to pgHBA) is simpler and less error prone.