graph-sequencer

Sort items in a graph using a topological sort while resolving cycles with priority groups.

Say you have some sort of graph of dependencies: (using an adjacency list)

let graph = new Map([
  ["task-a", ["task-d"]], // task-a depends on task-d
  ["task-b", ["task-d", "task-a"]],
  ["task-c", ["task-d"]],
  ["task-d", ["task-a"]],
]);

You could run a topological sort on these items, but you'd still end up with cycles:

task-a -> task-d -> task-a

To resolve this you pass "priority groups" to graph-sequencer:

let groups = [
  ["task-d"], // higher priority
  ["task-a", "task-b", "task-c"], // lower priority
];

The result will be a chunked list of items sorted topologically and by the priority groups:

let chunks = [
  ["task-d"],
  ["task-a", "task-c"],
  ["task-b"]
];

You can then run all these items in order with maximum concurrency:

for (let chunk of chunks) {
  await Promise.all(chunk.map(task => exec(task)));
}

However, even with priority groups you can still accidentally create cycles of dependencies in your graph.

graph-sequencer will return a list of the unresolved cycles:

let cycles = [
  ["task-a", "task-b"] // task-a depends on task-b which depends on task-a
];

However, graph-sequencer will still return an "unsafe" set of chunks. When it comes across a cycle, it will add another chunk with the item with the fewest dependencies remaining which will often break cycles.

All together that looks like this:

const graphSequencer = require('graph-sequencer');

graphSequencer({
  graph: new Map([
    ["task-a", ["task-d"]], // task-a depends on task-d
    ["task-b", ["task-d", "task-a"]],
    ["task-c", ["task-d"]],
    ["task-d", ["task-a"]],
  ]);
  groups: [
    ["task-d"], // higher priority
    ["task-a", "task-b", "task-c"], // lower priority
  ],
})
// {
//   safe: true,
//   chunks: [["task-d"], ["task-a", "task-c"], ["task-b"]],
//   cycles: [],
// }