Fair策略

image.png

FairSharePolicy.FairShareComparator

@Override
    public int compare(Schedulable s1, Schedulable s2) {
      double minShareRatio1, minShareRatio2;
      double useToWeightRatio1, useToWeightRatio2;
      double weight1, weight2;
      Resource minShare1 = Resources.min(RESOURCE_CALCULATOR, null,
          s1.getMinShare(), s1.getDemand());
      Resource minShare2 = Resources.min(RESOURCE_CALCULATOR, null,
          s2.getMinShare(), s2.getDemand());
      boolean s1Needy = Resources.lessThan(RESOURCE_CALCULATOR, null,
          s1.getResourceUsage(), minShare1);
      boolean s2Needy = Resources.lessThan(RESOURCE_CALCULATOR, null,
          s2.getResourceUsage(), minShare2);

      minShareRatio1 = (double) s1.getResourceUsage().getMemory()
          / Resources.max(RESOURCE_CALCULATOR, null, minShare1, ONE).getMemory();
      minShareRatio2 = (double) s2.getResourceUsage().getMemory()
          / Resources.max(RESOURCE_CALCULATOR, null, minShare2, ONE).getMemory();

      weight1 = s1.getWeights().getWeight(ResourceType.MEMORY);
      weight2 = s2.getWeights().getWeight(ResourceType.MEMORY);
      if (weight1 > 0.0 && weight2 > 0.0) {
        useToWeightRatio1 = s1.getResourceUsage().getMemory() / weight1;
        useToWeightRatio2 = s2.getResourceUsage().getMemory() / weight2;
      } else { // Either weight1 or weight2 equals to 0
        if (weight1 == weight2) {
          // If they have same weight, just compare usage
          useToWeightRatio1 = s1.getResourceUsage().getMemory();
          useToWeightRatio2 = s2.getResourceUsage().getMemory();
        } else {
          // By setting useToWeightRatios to negative weights, we give the
          // zero-weight one less priority, so the non-zero weight one will
          // be given slots.
          useToWeightRatio1 = -weight1;
          useToWeightRatio2 = -weight2;
        }
      }

      int res = 0;
      if (s1Needy && !s2Needy)
        res = -1;
      else if (s2Needy && !s1Needy)
        res = 1;
      else if (s1Needy && s2Needy)
        res = (int) Math.signum(minShareRatio1 - minShareRatio2);
      else
        // Neither schedulable is needy
        res = (int) Math.signum(useToWeightRatio1 - useToWeightRatio2);
      if (res == 0) {
        // Apps are tied in fairness ratio. Break the tie by submit time and job
        // name to get a deterministic ordering, which is useful for unit tests.
        res = (int) Math.signum(s1.getStartTime() - s2.getStartTime());
        if (res == 0) {
          res = s1.getName().compareTo(s2.getName());
        }
      }
      return res;
    }

➢ 实际最小资源份额：minShare = Min（资源需求量Demand，配置的最小资源MinShare）
➢ 是否饥饿：isNeedy = 资源使用量 < minShare（实际最小资源份额）
➢ 资源分配比：minShareRatio = 资源使用量 / Max（mindshare, 1）
➢ 资源使用权重比：useToWeightRatio = 资源使用量 / 权重

image.png

• FSQueue 中MinShare和MaxShare定义
  MinShare = MinResources
MaxShare = max(maxResource，minShare)
  如未定义maxResource, 默认为queueMaxResourcesDefault (Integer.MAX_VALUE)

• FSAppAttempt中MinShare和MaxShare定义
  MinShare = 0
  MaxShare = Integer.MAX_VALUE