Elasticsearch Sense contents to demonstrate how transactions with respect to search visibility may be implemented with parent-child relationships. Just copy paste into Sense and run the statements top to bottom.

transactions-sense-example

# Transactions with parent-child relationships.

# Delete the index, just in case it still exists.
DELETE /myindex

# Create the index with a mapping for the actual documents and a mapping for the parent transaction type.
PUT /myindex
{
  "mappings": {
    "transaction": {
      "properties": {
        "id": {
          "type": "string",
          "index": "not_analyzed"
        }
      }
    },
    "document": {
      "_parent": {
        "type": "transaction"
      },
      "properties": {
        "transaction_id": {
          "type": "string",
          "index": "not_analyzed"
        },
        "data": {
          "type": "string",
          "index": "not_analyzed"
        }
      }
    }
  }
}

# Index two documents with transaction 1 as parent.
PUT /myindex/document/1?parent=1
{
  "transaction_id": "1",
  "data": "part1"
}
PUT /myindex/document/2?parent=1
{
  "transaction_id": "1",
  "data": "part2"
}

# Try to search for one of the documents, including a check for the existence of the parent document. This will fail because transaction 1 does not exist yet.
GET /myindex/document/_search
{
  "query": {
    "filtered": {
      "query": {
        "match": {
          "data": "part1"
        }
      },
      "filter": {
        "has_parent": {
          "type": "transaction",
          "query": {"match_all": {}}
        }
      }
    }
  }
}

# Index another document of the same transaction.
PUT /myindex/document/3?parent=1
{
  "transaction_id": "1",
  "data": "part3"
}

# Assuming that the third document completes the document group, create the transaction to make all three documents visible for search.
PUT /myindex/transaction/1
{
  "id": "1"
}

# Again try to search for one of the documents. This will succeed now.
GET /myindex/document/_search
{
  "query": {
    "filtered": {
      "query": {
        "match": {
          "data": "part1"
        }
      },
      "filter": {
        "has_parent": {
          "type": "transaction",
          "query": {"match_all": {}}
        }
      }
    }
  }
}

# We can also retrieve all documents belonging to some transaction.
GET /myindex/document/_search
{
  "query": {
    "filtered": {
      "filter": {
        "has_parent": {
          "type": "transaction",
          "filter": {
            "term": {
              "id": "1"
            }
          }
        }
      }
    }
  }
}

# Index documents of another transaction.
PUT /myindex/document/10?parent=2
{
  "transaction_id": "2",
  "data": "part1"
}
PUT /myindex/document/11?parent=2
{
  "transaction_id": "2",
  "data": "part2"
}
PUT /myindex/document/12?parent=2
{
  "transaction_id": "2",
  "data": "part3"
}

# The second transaction is not visible, so searching for all "part1" documents will only show the one for transaction 1.
GET /myindex/document/_search
{
  "query": {
    "filtered": {
      "query": {
        "match": {
          "data": "part1"
        }
      },
      "filter": {
        "has_parent": {
          "type": "transaction",
          "query": {"match_all": {}}
        }
      }
    }
  }
}

# Create the second transaction.
PUT /myindex/transaction/2
{
  "id": "2"
}

# Sarch will show us the "part1" documents of all completed transactions.
GET /myindex/document/_search
{
  "query": {
    "filtered": {
      "query": {
        "match": {
          "data": "part1"
        }
      },
      "filter": {
        "has_parent": {
          "type": "transaction",
          "query": {"match_all": {}}
        }
      }
    }
  }
}