Skip to content

A quick and efficient data structure for indexing MQTT subscriptions in a highly concurrent environment.

Notifications You must be signed in to change notification settings

simon622/mqtt-tree

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

47 Commits
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Logo

MQTT Tree

MQTT Tree is a dependency free java implementation, designed to be used for storing, matching and retrieving MQTT subscriptions in a highly concurrent environment.

Table of Contents

  1. About
  2. Specification Conformance
  3. Memory & Time Performance
  4. Prefix Search
  5. Benchmarking
  6. Known Issues
  7. Example Usage

About

Broker and gateway implementations of MQTT will all need the ability to quickly and efficiently look up subscriptions. Subscriptions should be indexed (these contain wildcard characters) and then PUBLISH paths (which do not) ask the tree for all those subscriptions held matching the PUBLISH topic.

I have found myself writing this same code a number of times for different broker/GW implementations, so it made sense to make it available for others who may need to same thing.

Typical use would be to index your client session subscriptions using this tree for exceptionally fast lookup of subscriptions in high load environments. Modifications and Reads are entirely synchronized and can be performed without external synchronization.

Specification Conformance

I have attempted to make the tree compliant with both the normative and non-normative conformance statements in the MQTT Version 5 specification document. You will find specification compliance unit tests located in MqttSpecificationComplianceTests.java. Whilst these test pass, there are certain aspects of path segmentation matching that are subjective. I will be seeking clarification, but there are tests in the non-formal test class which demonstrate this.

Memory & Time Performance

The tree uses a Tries algorithm internally (prefix tree), where each level of the tree is a normalized path segment. Thus a million subscriptions to the same topic path would only result in that topic path being stored once. Branch nodes along the path will diverge where new path segments are added. Therefore search performance and modification performance are a function of the number of levels of the tree traversed.

Memberships to a given leaf node (subscriptions) are purposefully arbitrary types, allowing the calling code to specify their own member types. The stipulation is that members are maintained in a HashSet, therefore the equals and hashCode methods of any chosen member type should be consistent.

Prefix Search

Being able to visualise the large datasets that are generated by topic trees is a constant struggle. I have included an optional wrapper class which you can choose to wrap your MqttTree in, which will index all of your paths and make them searchable against a single character prefix. I have done this using a radix tree, which indexes the topics during creation for look-ahead searching. This is a decent enough index for thousands of nodes, when there are millions, the overhead of the extra index will be burdensome, so use wisely. I have knocked together a quick Java swing application to show how the prefix searching works.

//Wrap your tree in a searchable tree, and expose the new method
ISearchableMqttTree<String> searchableTree = new SearchableMqttTree(
                new MqttTree<String>(MqttTree.DEFAULT_SPLIT, true));

searchableTree.prefixSearch("some/pre", 100);

Benchmarking

I have yet to undertake a formal benchmark of the tree, but I would be happy to hear the results should someone else wish to compare against other implementations. I would advise that for a real comparison, a benchmark should incorporate concurrent read/updates to the tree.

Known Issues

As outlined in the compliance section, certain aspects of path matching are open to interpretation. Where consecutive path segments are defined the tree will presently match separators in these locations. I will look to change this when time allows. Further, I will be looking to optimize the matching algorithm further.

Example Usage

public class Example {
    public static void main(String[] args) throws MqttTreeException, MqttTreeLimitExceededException {
        MqttTree<String> tree = new MqttTree<String>(MqttTree.DEFAULT_SPLIT, true);
        tree.withMaxPathSegments(1024);
        tree.withMaxMembersAtLevel(1024);

        tree.subscribe("/this/is/a/topic", "ClientId1", "ClientId2");
        tree.subscribe("/this/+/a/topic", "ClientId3");
        tree.subscribe("/this/#", "ClientId4");
        Set<String> m = tree.search("/this/is/a/topic");
        System.out.println(String.format("matching search had [%s] members", m.size()));
        m = tree.search("/is/a/different/topic");
        System.out.println(String.format("non-matching search had [%s] members", m.size()));
        tree.unsubscribe("/this/is/a/topic", "ClientId2");
        m = tree.search("/this/is/a/topic");
        System.out.println(String.format("matching search had [%s] members", m.size()));
    }
}

About

A quick and efficient data structure for indexing MQTT subscriptions in a highly concurrent environment.

Topics

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages