Clash between ON CLUSTER in table DDL queries and Replicated engine

[Tristan971]

The Replicated database engine essentially makes (database-scoped, eg no user mgmt) DDL queries run with ON CLUSTER ... implicitly.

However, on top of this it also disallows database-queries having explicit ON CLUSTER ... in queries...

Which makes migrations fail with eg

Loading plausible..
Starting dependencies..
Starting repos..
Running migrations for Elixir.Plausible.Repo
Running migrations for Elixir.Plausible.IngestRepo
warning: code block contains unused literal "Migration adds a ALIAS columns needed to keep DB schema and api\nproperty naming in sync to reduce overhead in code.\n" (remove the literal or assign it to _ to avoid warnings)
└─ lib/plausible-0.0.1/priv/ingest_repo/migrations/20240502115822_alias_api_prop_names.exs: Plausible.IngestRepo.Migrations.AliasApiPropNames (module)

** (FunctionClauseError) no function clause matching in Ch.RowBinary."-inlined-decode_string_decode_rows/5-"/5    
    
    The following arguments were given to Ch.RowBinary."-inlined-decode_string_decode_rows/5-"/5:
    
        # 1
        "version 24.9.3.128 (official build)). (INCORRECT_QUERY) (version 24.9.2.42 (official build))\n"
    
        # 2
        [:u16, :i64, :string, :u64, :u64]
    
        # 3
        []
    
        # 4
        [["ode: 80. DB::Exception: There was an error on [chi-dev-app-main-0-1", 14650, 8017287196693246000, "e: 80. DB::Exception: It's not initial query. ON CLUSTER is not allowed for Replicated database. (IN", 6869189296659779395, 2891356424380962129], ["chi-dev-app-main-0-0", 9000, 80, "Code: 80. DB::Exception: It's not initial query. ON CLUSTER is not allowed for Replicated database. (INCORRECT_QUERY) (version 24.9.2.42 (official build))", 0, 0], ["chi-dev-app-main-0-1", 9000, 80, "Code: 80. DB::Exception: It's not initial query. ON CLUSTER is not allowed for Replicated database. (INCORRECT_QUERY) (version 24.9.3.128 (official build))", 1, 1]]
    
        # 5
        [:string, :u16, :i64, :string, :u64, :u64]
    
    (ch 0.2.5) lib/ch/row_binary.ex:608: Ch.RowBinary."-inlined-decode_string_decode_rows/5-"/5
    (ch 0.2.5) lib/ch/query.ex:183: DBConnection.Query.Ch.Query.decode/3
    (db_connection 2.6.0) lib/db_connection.ex:1433: DBConnection.decode/4
    (db_connection 2.6.0) lib/db_connection.ex:801: DBConnection.execute/4
    (ch 0.2.5) lib/ch.ex:87: Ch.query/4
    (ecto_sql 3.11.1) lib/ecto/adapters/sql.ex:519: Ecto.Adapters.SQL.query!/4
    (elixir 1.16.0) lib/enum.ex:1700: Enum."-map/2-lists^map/1-1-"/2

I'm mildly reluctant to make this a feature request, since it's probably niche especially for selfhosted, so... yeah, not too sure how to proceed besides doing the migrations by hand.

[Tristan971]

fwiw this was the "approach" of the CH team when clickhouse-copier ran into the same issue ClickHouse/ClickHouse#35570 (comment)

[Tristan971]

(which doesn't make sense to me, I mean... looks like a hack that works only in a small number of cases)

[Tristan971]

meh, it's really annoying and even they got it wrong multiple times... I think this makes more sense as an upstream bug of CH that ON CLUSTER {cluster} (ie w/ the macro) isn't entirely ignored for Replicated-engine-distributed queries... (as opposed to their odd check which is based on the name of the database which still makes no sense to me)

[Tristan971]

Well I raised ClickHouse/ClickHouse#79565 for now, we'll see what comes of it, because after a bit of playing around with the code, it'd be fairly ugly to add yet another parameter to all these on_cluster_statement lookups