[bolt v3] Bolt.Sips.transaction crash with invalid query

[dominique-vassard]

Considering the awaited return, which happened in bolt version 1 and 2 (Neo4j < 3.4)

iex(5)> Bolt.Sips.transaction(Bolt.Sips.conn(), fn conn ->
...(5)> Bolt.Sips.query(conn, "RETRN 1 AS num")           
...(5)> end)
{:ok,
 {:error,
  [
    code: "Neo.ClientError.Statement.SyntaxError",
    message: "Invalid input 'R': expected 'u/U' (line 1, column 4 (offset: 3))\n\"RETRN 1 AS num\"\n    ^"
  ]}}

The same code in a bolt V3 Neo4j server:

iex(7)> Bolt.Sips.transaction(Bolt.Sips.conn(), fn conn ->  
...(7)> Bolt.Sips.query(conn, "RETRN 1 AS num")             
...(7)> end)
** (MatchError) no match of right hand side value: {:error, %Bolt.Sips.Internals.Error{code: "Neo.ClientError.Request.Invalid", connection_id: 10928, function: :commit, message: "Message 'COMMIT' cannot be handled by a session in the READY state.", type: :cypher_error}}
    (bolt_sips) lib/bolt_sips/protocol.ex:117: Bolt.Sips.Protocol.handle_commit/2
    (db_connection) lib/db_connection/holder.ex:270: DBConnection.Holder.holder_apply/4
    (db_connection) lib/db_connection.ex:1468: DBConnection.run_commit/3
    (db_connection) lib/db_connection.ex:1006: DBConnection.checkin/4
    (db_connection) lib/db_connection.ex:1450: DBConnection.commit/3
    (db_connection) lib/db_connection.ex:1377: DBConnection.run_transaction/4

This is due to how DBConnection handles the transaction:

• execute the query
• rollback if there is an error

But in Bolt v3, a rollback is automatically performed at server-level if query is invalid. All is required is a RESET message...

[dominique-vassard]

Investigation leads to discover some weird stuff. When server sends back a FAILURE message:

• If it's due to a invalid query, a RESET and then everything's fine
• If it's due to an invalid message (for ex. COMMIT, ROLLBACK), then socket is closed and can't be used anymore

I opened a topic on Neo4j's forum in order to know if it's the expected behaviour or not.

[florinpatrascu]

a rollback initiated by the server sounds awkward, from the server' design perspective. I'm following your thread on on their forums and await for replies.

[sashman]

Are there any updates/breadcrumbs on this? I am facing a similar issue where I have a large number of queries failing in a transaction with:

** (Protocol.UndefinedError) protocol String.Chars not implemented for %MatchError{term: {:error, %Bolt.Sips.Internals.Error{code: "Neo.ClientError.Request.Invalid", connection_id: 352, function: :commit, message: "Message 'COMMIT' cannot be handled by a session in the READY state.", type: :cypher_error}}} of type MatchError (a struct). This protocol is implemented for the following type(s): Phoenix.LiveComponent.CID, Integer, Date, DateTime, Version, Version.Requirement, Float, NaiveDateTime, Atom, List, URI, BitString, Time

[Anders2303]

Irregardless of whether the rollback should be initiated by the server, could we atleast make it so the error message contains the reason the query crashes (ala the error message we got in v.2)? As it stands, I need to manually print out the error reason to see what went wrong:

Bolt.Sips.transaction(Bolt.Sips.conn. fn conn -> 
 case Bolt.Sips.query(conn, cypher, attrs) do
   {:error, reason} ->
     IO.inspect(reason)
     throw(reason)
   res ->
     res
   end
end)

[sukidhar]

Considering the awaited return, which happened in bolt version 1 and 2 (Neo4j < 3.4)

iex(5)> Bolt.Sips.transaction(Bolt.Sips.conn(), fn conn ->
...(5)> Bolt.Sips.query(conn, "RETRN 1 AS num")           
...(5)> end)
{:ok,
 {:error,
  [
    code: "Neo.ClientError.Statement.SyntaxError",
    message: "Invalid input 'R': expected 'u/U' (line 1, column 4 (offset: 3))\n\"RETRN 1 AS num\"\n    ^"
  ]}}

The same code in a bolt V3 Neo4j server:

iex(7)> Bolt.Sips.transaction(Bolt.Sips.conn(), fn conn ->  
...(7)> Bolt.Sips.query(conn, "RETRN 1 AS num")             
...(7)> end)
** (MatchError) no match of right hand side value: {:error, %Bolt.Sips.Internals.Error{code: "Neo.ClientError.Request.Invalid", connection_id: 10928, function: :commit, message: "Message 'COMMIT' cannot be handled by a session in the READY state.", type: :cypher_error}}
    (bolt_sips) lib/bolt_sips/protocol.ex:117: Bolt.Sips.Protocol.handle_commit/2
    (db_connection) lib/db_connection/holder.ex:270: DBConnection.Holder.holder_apply/4
    (db_connection) lib/db_connection.ex:1468: DBConnection.run_commit/3
    (db_connection) lib/db_connection.ex:1006: DBConnection.checkin/4
    (db_connection) lib/db_connection.ex:1450: DBConnection.commit/3
    (db_connection) lib/db_connection.ex:1377: DBConnection.run_transaction/4

This is due to how DBConnection handles the transaction:

• execute the query
• rollback if there is an error

But in Bolt v3, a rollback is automatically performed at server-level if query is invalid. All is required is a RESET message...

I am experiencing similar crash, when query fails. anything to resolve this ?

[sukidhar]

Are there any updates/breadcrumbs on this? I am facing a similar issue where I have a large number of queries failing in a transaction with:

** (Protocol.UndefinedError) protocol String.Chars not implemented for %MatchError{term: {:error, %Bolt.Sips.Internals.Error{code: "Neo.ClientError.Request.Invalid", connection_id: 352, function: :commit, message: "Message 'COMMIT' cannot be handled by a session in the READY state.", type: :cypher_error}}} of type MatchError (a struct). This protocol is implemented for the following type(s): Phoenix.LiveComponent.CID, Integer, Date, DateTime, Version, Version.Requirement, Float, NaiveDateTime, Atom, List, URI, BitString, Time

Did you find any solution ?

[florinpatrascu]

No. PRs are very welcome.