UI: Topology Visualization

ui/app/components/flex-masonry.js

@@ -0,0 +1,66 @@
+import Component from '@glimmer/component';
+import { tracked } from '@glimmer/tracking';
+import { run } from '@ember/runloop';
+import { action } from '@ember/object';
+import { minIndex, max } from 'd3-array';
+
+export default class FlexMasonry extends Component {
+  @tracked element = null;
+
+  @action
+  captureElement(element) {
+    this.element = element;
+  }
+
+  @action
+  reflow() {
+    run.next(() => {
+      // There's nothing to do if this is a single column layout
+      if (!this.element || this.args.columns === 1 || !this.args.columns) return;
+
+      const columns = new Array(this.args.columns).fill(null).map(() => ({
+        height: 0,
+        elements: [],
+      }));
+
+      const items = this.element.querySelectorAll('.flex-masonry-item');
+
+      // First pass: assign each element to a column based on the running heights of each column
+      for (let item of items) {
+        const styles = window.getComputedStyle(item);
+        const marginTop = parseFloat(styles.marginTop);
+        const marginBottom = parseFloat(styles.marginBottom);
+        const height = item.clientHeight;
+
+        // Pick the shortest column accounting for margins
+        const column = columns[minIndex(columns, c => c.height)];
+
+        // Add the new element's height to the column height
+        column.height += marginTop + height + marginBottom;
+        column.elements.push(item);
+      }
+
+      // Second pass: assign an order to each element based on their column and position in the column
+      columns
+        .mapBy('elements')
+        .flat()
+        .forEach((dc, index) => {
+          dc.style.order = index;
+        });
+
+      // Guarantee column wrapping as predicted (if the first item of a column is shorter than the difference
+      // beteen the height of the column and the previous column, then flexbox will naturally place the first
+      // item at the end of the previous column).
+      columns.forEach((column, index) => {
+        const nextHeight = index < columns.length - 1 ? columns[index + 1].height : 0;
+        const item = column.elements.lastObject;
+        if (item) {
+          item.style.flexBasis = item.clientHeight + Math.max(0, nextHeight - column.height) + 'px';
+        }
+      });
+
+      // Set the max height of the container to the height of the tallest column
+      this.element.style.maxHeight = max(columns.mapBy('height')) + 1 + 'px';
+    });
+  }
+}

ui/app/components/topo-viz.js

@@ -0,0 +1,271 @@
+import Component from '@glimmer/component';
+import { tracked } from '@glimmer/tracking';
+import { action, set } from '@ember/object';
+import { run } from '@ember/runloop';
+import { scaleLinear } from 'd3-scale';
+import { extent, deviation, mean } from 'd3-array';
+import { line, curveBasis } from 'd3-shape';
+
+export default class TopoViz extends Component {
+  @tracked element = null;
+  @tracked topology = { datacenters: [] };
+
+  @tracked activeNode = null;
+  @tracked activeAllocation = null;
+  @tracked activeEdges = [];
+  @tracked edgeOffset = { x: 0, y: 0 };
+
+  get isSingleColumn() {
+    if (this.topology.datacenters.length <= 1) return true;
+
+    // Compute the coefficient of variance to determine if it would be
+    // better to stack datacenters or place them in columns
+    const nodeCounts = this.topology.datacenters.map(datacenter => datacenter.nodes.length);
+    const variationCoefficient = deviation(nodeCounts) / mean(nodeCounts);
+
+    // The point at which the varation is too extreme for a two column layout
+    const threshold = 0.5;
+    if (variationCoefficient > threshold) return true;
+    return false;
+  }
+
+  get datacenterIsSingleColumn() {
+    // If there are enough nodes, use two columns of nodes within
+    // a single column layout of datacenters to increase density.
+    return !this.isSingleColumn || (this.isSingleColumn && this.args.nodes.length <= 20);
+  }
+
+  // Once a cluster is large enough, the exact details of a node are
+  // typically irrelevant and a waste of space.
+  get isDense() {
+    return this.args.nodes.length > 50;
+  }
+
+  dataForNode(node) {
+    return {
+      node,
+      datacenter: node.datacenter,
+      memory: node.resources.memory,
+      cpu: node.resources.cpu,
+      allocations: [],
+      isSelected: false,
+    };
+  }
+
+  dataForAllocation(allocation, node) {
+    const jobId = allocation.belongsTo('job').id();
+    return {
+      allocation,
+      node,
+      jobId,
+      groupKey: JSON.stringify([jobId, allocation.taskGroupName]),
+      memory: allocation.allocatedResources.memory,
+      cpu: allocation.allocatedResources.cpu,
+      memoryPercent: allocation.allocatedResources.memory / node.memory,
+      cpuPercent: allocation.allocatedResources.cpu / node.cpu,
+      isSelected: false,
+    };
+  }
+
+  @action
+  buildTopology() {
+    const nodes = this.args.nodes;
+    const allocations = this.args.allocations;
+
+    // Wrap nodes in a topo viz specific data structure and build an index to speed up allocation assignment
+    const nodeContainers = [];
+    const nodeIndex = {};
+    nodes.forEach(node => {
+      const container = this.dataForNode(node);
+      nodeContainers.push(container);
+      nodeIndex[node.id] = container;
+    });
+
+    // Wrap allocations in a topo viz specific data structure, assign allocations to nodes, and build an allocation
+    // index keyed off of job and task group
+    const allocationIndex = {};
+    allocations.forEach(allocation => {
+      const nodeId = allocation.belongsTo('node').id();
+      const nodeContainer = nodeIndex[nodeId];
+      if (!nodeContainer)
+        throw new Error(`Node ${nodeId} for alloc ${allocation.id} not in index.`);
+
+      const allocationContainer = this.dataForAllocation(allocation, nodeContainer);
+      nodeContainer.allocations.push(allocationContainer);
+
+      const key = allocationContainer.groupKey;
+      if (!allocationIndex[key]) allocationIndex[key] = [];
+      allocationIndex[key].push(allocationContainer);
+    });
+
+    // Group nodes into datacenters
+    const datacentersMap = nodeContainers.reduce((datacenters, nodeContainer) => {
+      if (!datacenters[nodeContainer.datacenter]) datacenters[nodeContainer.datacenter] = [];
+      datacenters[nodeContainer.datacenter].push(nodeContainer);
+      return datacenters;
+    }, {});
+
+    // Turn hash of datacenters into a sorted array
+    const datacenters = Object.keys(datacentersMap)
+      .map(key => ({ name: key, nodes: datacentersMap[key] }))
+      .sortBy('name');
+
+    const topology = {
+      datacenters,
+      allocationIndex,
+      selectedKey: null,
+      heightScale: scaleLinear()
+        .range([15, 40])
+        .domain(extent(nodeContainers.mapBy('memory'))),
+    };
+    this.topology = topology;
+  }
+
+  @action
+  captureElement(element) {
+    this.element = element;
+  }
+
+  @action
+  showNodeDetails(node) {
+    if (this.activeNode) {
+      set(this.activeNode, 'isSelected', false);
+    }
+
+    this.activeNode = this.activeNode === node ? null : node;
+
+    if (this.activeNode) {
+      set(this.activeNode, 'isSelected', true);
+    }
+
+    if (this.args.onNodeSelect) this.args.onNodeSelect(this.activeNode);
+  }
+
+  @action
+  associateAllocations(allocation) {
+    if (this.activeAllocation === allocation) {
+      this.activeAllocation = null;
+      this.activeEdges = [];
+
+      if (this.topology.selectedKey) {
+        const selectedAllocations = this.topology.allocationIndex[this.topology.selectedKey];
+        if (selectedAllocations) {
+          selectedAllocations.forEach(allocation => {
+            set(allocation, 'isSelected', false);
+          });
+        }
+        set(this.topology, 'selectedKey', null);
+      }
+    } else {
+      if (this.activeNode) {
+        set(this.activeNode, 'isSelected', false);
+      }
+      this.activeNode = null;
+      this.activeAllocation = allocation;
+      const selectedAllocations = this.topology.allocationIndex[this.topology.selectedKey];
+      if (selectedAllocations) {
+        selectedAllocations.forEach(allocation => {
+          set(allocation, 'isSelected', false);
+        });
+      }
+
+      set(this.topology, 'selectedKey', allocation.groupKey);
+      const newAllocations = this.topology.allocationIndex[this.topology.selectedKey];
+      if (newAllocations) {
+        newAllocations.forEach(allocation => {
+          set(allocation, 'isSelected', true);
+        });
+      }
+
+      this.computedActiveEdges();
+    }
+    if (this.args.onAllocationSelect)
+      this.args.onAllocationSelect(this.activeAllocation && this.activeAllocation.allocation);
+    if (this.args.onNodeSelect) this.args.onNodeSelect(this.activeNode);
+  }
+
+  @action
+  computedActiveEdges() {
+    // Wait a render cycle
+    run.next(() => {
+      const path = line().curve(curveBasis);
+      // 1. Get the active element
+      const allocation = this.activeAllocation.allocation;
+      const activeEl = this.element.querySelector(`[data-allocation-id="${allocation.id}"]`);
+      const activePoint = centerOfBBox(activeEl.getBoundingClientRect());
+
+      // 2. Collect the mem and cpu pairs for all selected allocs
+      const selectedMem = Array.from(this.element.querySelectorAll('.memory .bar.is-selected'));
+      const selectedPairs = selectedMem.map(mem => {
+        const id = mem.closest('[data-allocation-id]').dataset.allocationId;
+        const cpu = mem
+          .closest('.topo-viz-node')
+          .querySelector(`.cpu .bar[data-allocation-id="${id}"]`);
+        return [mem, cpu];
+      });
+      const selectedPoints = selectedPairs.map(pair => {
+        return pair.map(el => centerOfBBox(el.getBoundingClientRect()));
+      });
+
+      // 3. For each pair, compute the midpoint of the truncated triangle of points [Mem, Cpu, Active]
+      selectedPoints.forEach(points => {
+        const d1 = pointBetween(points[0], activePoint, 100, 0.5);
+        const d2 = pointBetween(points[1], activePoint, 100, 0.5);
+        points.push(midpoint(d1, d2));
+      });
+
+      // 4. Generate curves for each active->mem and active->cpu pair going through the bisector
+      const curves = [];
+      // Steps are used to restrict the range of curves. The closer control points are placed, the less
+      // curvature the curve generator will generate.
+      const stepsMain = [0, 0.8, 1.0];
+      // The second prong the fork does not need to retrace the entire path from the activePoint
+      const stepsSecondary = [0.8, 1.0];
+      selectedPoints.forEach(points => {
+        curves.push(
+          curveFromPoints(...pointsAlongPath(activePoint, points[2], stepsMain), points[0]),
+          curveFromPoints(...pointsAlongPath(activePoint, points[2], stepsSecondary), points[1])
+        );
+      });
+
+      this.activeEdges = curves.map(curve => path(curve));
+      this.edgeOffset = { x: window.visualViewport.pageLeft, y: window.visualViewport.pageTop };
+    });
+  }
+}
+
+function centerOfBBox(bbox) {
+  return {
+    x: bbox.x + bbox.width / 2,
+    y: bbox.y + bbox.height / 2,
+  };
+}
+
+function dist(p1, p2) {
+  return Math.sqrt(Math.pow(p2.x - p1.x, 2) + Math.pow(p2.y - p1.y, 2));
+}
+
+// Return the point between p1 and p2 at len (or pct if len > dist(p1, p2))
+function pointBetween(p1, p2, len, pct) {
+  const d = dist(p1, p2);
+  const ratio = d < len ? pct : len / d;
+  return pointBetweenPct(p1, p2, ratio);
+}
+
+function pointBetweenPct(p1, p2, pct) {
+  const dx = p2.x - p1.x;
+  const dy = p2.y - p1.y;
+  return { x: p1.x + dx * pct, y: p1.y + dy * pct };
+}
+
+function pointsAlongPath(p1, p2, pcts) {
+  return pcts.map(pct => pointBetweenPct(p1, p2, pct));
+}
+
+function midpoint(p1, p2) {
+  return pointBetweenPct(p1, p2, 0.5);
+}
+
+function curveFromPoints(...points) {
+  return points.map(p => [p.x, p.y]);
+}

ui/app/components/topo-viz/datacenter.js

@@ -0,0 +1,32 @@
+import Component from '@glimmer/component';
+
+export default class TopoVizDatacenter extends Component {
+  get scheduledAllocations() {
+    return this.args.datacenter.nodes.reduce(
+      (all, node) => all.concat(node.allocations.filterBy('allocation.isScheduled')),
+      []
+    );
+  }
+
+  get aggregatedAllocationResources() {
+    return this.scheduledAllocations.reduce(
+      (totals, allocation) => {
+        totals.cpu += allocation.cpu;
+        totals.memory += allocation.memory;
+        return totals;
+      },
+      { cpu: 0, memory: 0 }
+    );
+  }
+
+  get aggregatedNodeResources() {
+    return this.args.datacenter.nodes.reduce(
+      (totals, node) => {
+        totals.cpu += node.cpu;
+        totals.memory += node.memory;
+        return totals;
+      },
+      { cpu: 0, memory: 0 }
+    );
+  }
+}