DB Connection Error - connection not available and request was dropped

Can you post the queries?

This is one of the query tests

  test "query comments with sortBy inserted_at and sortOrder DESC", context do
       current_user = tennat_with_current_user(context.tenant)
 
       request_and_parse_gql(
         """
         query {
             comments(sort: { sortBy: INSERTED_AT, sortOrder: DESC}) {
                 entries{
                     id
                 }
             }
         }
         """

,
         %{current_user: current_user}
       )
       |> graphql_response(
         expected_response(
           [
             context.comment4,
             context.comment3,
             context.comment2,
             context.comment1
           ],
           [:id]
         ),
         200
       )
 
       parse_graphql(
         """
           query {
             comments(sort: { sortBy: INSERTED_AT, sortOrder: DESC}, limit: 2) {
               paginationMetadata {
                 after
                 before
                 limit
                 totalCount
                 totalCountCapExceeded
               }
             }
           }
         """,
         %{current_user: current_user}
       )
       |> graphql_response(
         %{
           "comments" => %{
             "paginationMetadata" => %{
               "after" =>
                 Paginator.cursor_for_record(
                   context.comment3,
                   id: :asc,
                   inserted_at: :desc
                 ),
               "before" => nil,
               "limit" => 2,
               "totalCount" => 4,
               "totalCountCapExceeded" => false
             }
           }
         },
         200
       )
     end

Not GraphQL queries. Post the SQL queries.

can I know where I can find this

One cause of this in tests: parallel tests that all get blocked waiting for each other’s DB transaction to commit/rollback.

For instance, let’s say you have a Company schema with a name field covered by a unique index.

If multiple tests try to insert a record with the same name, all but the first one will block until the first one completes - and since ex_unit starts System.schedulers_online * 2 cases in parallel, this can cause connection starvation like you’re seeing if the pool is too small.

Boosting the pool size can help, but the better fix is to make sure your code either:

• inserts the shared records before switching the sandbox on
• inserts unique values into the unique-indexed column (with an ExMachina sequence, for instance)

Hi! Is it possible to get an explanation how one can: