Your solution will not work, as you can have a race condition where any number of threads match on Foo and calculate the same max value before each creates a new Foo with the next number. This is because you have not done anything to serialize the calculation of the next sequence number.
What I have done is create a node for each sequence generator I need. I then query for this node, update its sequence number, and return the new value. I do this in a transaction and I lock the node with a 'set' operations before I do anything. This seems to lock the node, making all threads serialize behind this lock if there is a race condition among multiple threads. I have done this in a custom procedure for my application because I need it for other custom procedures, but I have created an example with some Junit tests to show you the approach and the issue concurrency.
Here is a snippet of code with the concurrency issue. This is running in a test container. First I delete all data and then create a new 'UniqueId' node, which is used to track the sequence for invoice numbers. The sequence is seeded with a starting id of 100. In the test method, I created 1000 threads, each trying to get the next sequence number by matching on the node with the name 'invoice', calculating the next number in the sequence, and updating the node. The 'with' clause does not lock the node, allowing any number of threads to read the same value of id, thus calculating the same value of id+1 and persisting (overwriting) and returning the same value.
@BeforeEach
void setUpdata() {
try(final Session session = driver.session()) {
session.run("MATCH(n:UniqueId) DELETE n");
session.run("CREATE(:UniqueId {name:'invoice', id:100})");
}
}
@Test
public void testConcurrencyWithCallable() throws InterruptedException {
List<Future<Long>> futures = new ArrayList<>();
Callable<Long> callable = () -> {
try (final Session session = driver.session()) {
return session.executeWrite(tx -> {
Result result = tx.run("""
MATCH (id:UniqueId{name:'invoice'})
WITH id, id.id as x
SET id.id = x + 1
RETURN id.id as id"""
);
return result.single().get("id").asLong();
});
}
};
for (int i = 0; i < 1000; i++) {
futures.add(threadPool.submit(callable));
}
List<Long> ids = new ArrayList<>();
futures.forEach(x -> {
try {
Long id = x.get();
ids.add(id);
} catch (InterruptedException | ExecutionException e) {
throw new RuntimeException(e);
}
});
ids.sort(Comparator.comparingLong(x->x));
ids.forEach(System.out::println);
threadPool.awaitTermination(10, TimeUnit.SECONDS);
threadPool.shutdown();
}
Following is the beginning of the sequence of id's returned. As you can see, there is a lot of duplications because multiple threads concurrently read the same id value in the 'with' clause.
100
100
100
100
100
100
100
100
100
100
100
100
100
100
100
100
100
100
100
100
100
100
100
100
100
100
100
100
101
101
101
101
101
101
101
101
101
101
101
101
101
101
101
101
101
101
101
101
101
101
101
101
101
101
101
101
101
101
101
101
101
101
102
102
102
102
102
102
102
102
102
To fix this issue, we need to lock the UniqueId node with the name 'invoice', so we can can calculate the next sequence number sequentially. To do this, we immediately execute a 'set' operation so the node is locked, then calculate the next value by adding 1. Below is the new query to get the next sequence number and update the node in a serialized fashion.
MATCH (id:UniqueId{name:'invoice'})
SET id.id = id.id + 1
RETURN id.id as id
The following is the sequence of id's with the updated query:
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
You can do something more like this in a transaction, which will serialize the update of this node, giving you a sequence generator without duplicates.
MATCH (n:Foo{name:'sequence_name'})
SET n.next_number = n.next_number + 1
//Set other attributes
RETURN n.next_number
If you need to other stuff before calculating the next number, you can set a temporary property to lock the node, do other needed processing, then remove the temporary property at the end of the query.
MATCH (n:Foo{name:'sequence_name'})
SET n._LOCK = true
//do other stuff
SET n.next_number = n.next_number + 1
REMOVE n._LOCK
RETURN n.next_number
An interesting implementation would be to wrap this code in an APOC custom procedure so you can call it as a procedure when needed, eliminating duplicating the code in each query that the sequence number is needed.
Here is an example. Create the custom procedure in the 'system' database. Execute :use system to switch to the system database, and :use neo4j to switch back.
CALL apoc.custom.installProcedure(
'getNextSequenceNumber(sequence::STRING) :: (nextValue::INT)',
'
MATCH (n:Sequence{name:$sequence})
SET n.next_number = n.next_number + 1
RETURN n.next_number as nextValue
',
'neo4j',
'write'
)
After creating the Sequence node with name equal to 'sequence_name' and an initial next_number value of 100, you can get the next number in the sequence as follows:
call custom.getNextSequenceNumber('sequence_name') yield nextValue
return nextValue
Second call:
You can create any number of sequence nodes, just use a different name value that represent what the sequence is for, then call the custom procedure with the sequence's name.
