Fixes the Cypher files execution not happening sequentially

[ncordon]

What

Fixes runFiles so that the execution happens sequentially, rather than in parallel.

It wasn't working because of two reasons:

• We were creating a list with streams produced by runManyStatements, which starts a thread. Which means all threads were started effectively at the same time for the different files it receives as input.
• Stream.reduce doesn't have good semantics to achieve the sequential execution either. On the one hand it's eager (or terminal as they call it in the docs), which means the threads are fired when we call the reduce phase and not when the stream is consumed. On the other hand it asks for associativity, which means we could group operations like (read the + as in execute the operation, not just concatenating them, since we have said it's eager):

a + b + c + d
as
(a + b) + (c + d)

flatMap on the other hand is an intermediate operation (it creates a lazy stream) and the threads will not be fired until the Stream is consumed.

Why

apoc.cypher.runFiles and apoc.cypher.runSchemaFile were not being executed sequentially.

That was a problem creating flakiness in the modified testRunFilesMultiple test.

It was reproducible doing apoc.cypher.runFiles with the following (added in the PR to the test) files:

UNWIND RANGE(0,2) as id
CREATE (n:Node {id:id})
RETURN n.id as id;

CALL apoc.util.sleep(2000)

MATCH (n) DELETE n; // The files definitely run in parallel cause I see 4 deleted here

and

UNWIND RANGE(0,2) as id
CREATE (n:Node {id:id})
RETURN n.id as id;

CALL apoc.util.sleep(1000)

MATCH (n) DELETE n;

[AzuObs]

I think that you're right and that reduce was the problem. The java docs for Stream.reduce says that this function is not constrained to run sequentially.