We’re overhauling Dgraph’s docs to make them clearer and more approachable. If you notice any issues during this transition or have suggestions, please let us know.
To understand how query execution works, look at an example. 
{
    me(func: uid(0x1)) {
      rel_A
      rel_B {
        rel_B1
        rel_B2
      }
      rel_C {
        rel_C1
        rel_C2 {
          rel_C2_1
      }
      }
  }
}
Let’s assume we have 3 Alpha instances, and instance id=2 receives this query. These are the steps:
  • This query specifies the exact UID list (one UID) to start with, so there is no root query clause.
  • Retreive posting lists using keys = 0x1::rel_A, 0x1::rel_B, and 0x1::rel_C.
    • At worst, these predicates could belong to 3 different groups if the DB is sharded, so this would incur at most 3 network calls.
  • The above posting lists would include three lists of UIDs or values.
    • The UID results (id1, id2, …, idn) for rel_B are converted into queries for id1::rel_B1 id2::rel_B1, etc., and for id1::rel_B2 id2::rel_B2, etc.
    • Similarly, results for rel_C will be used to get the next set of UIDs from posting list keys like id::rel_C1 and id::rel_C2.
  • This process continues recursively for rel_C2_1 as well, and as deep as any query requires.
More complex queries may do filtering operations, or intersections and unions of UIDs, but this recursive walk to execute a number of (often parallel) Tasks to retrieve UIDs characterizes Dgraph querying. If the query was run via HTTP interface /query, the resulting subgraph then gets converted into JSON for replying back to the client. If the query was run via gRPC interface using the language clients, the subgraph gets converted to protocol buffer format and similarly returned to the client.