{
  "conclusions": {
    "checkpoint_bounds_lineage_replay": true,
    "lost_wide_dependency_can_replay_multiple_parent_partitions": true,
    "mapreduce_materializes_each_round": true,
    "same_numeric_result": true,
    "task_side_effects_can_repeat": true
  },
  "mapreduce": {
    "durable_partition_writes": 16,
    "input_partition_reads": 32,
    "result": {
      "0": 1.554067,
      "1": 0.644089,
      "2": 1.651844,
      "3": 0.15
    },
    "trace": [
      "round 1: read graph+ranks, materialize all rank partitions",
      "round 2: read graph+ranks, materialize all rank partitions",
      "round 3: read graph+ranks, materialize all rank partitions",
      "round 4: read graph+ranks, materialize all rank partitions"
    ]
  },
  "model": "deterministic RDD lineage/materialization model",
  "narrow_dependency_loss": {
    "lost_partition": 2,
    "parent_partitions_touched": [
      2
    ],
    "restored": [
      40
    ]
  },
  "not_modeled": [
    "a Spark or Hadoop runtime",
    "network transfer, disk throughput, scheduler locality, speculative execution",
    "arbitrary DAGs or production failure recovery"
  ],
  "spark_checkpoint_after_iteration_2": {
    "duplicate_side_effect_keys": [
      "iteration-3-partition-0",
      "iteration-3-partition-1",
      "iteration-3-partition-2",
      "iteration-3-partition-3",
      "iteration-4-partition-0",
      "iteration-4-partition-1",
      "iteration-4-partition-2",
      "iteration-4-partition-3"
    ],
    "durable_partition_writes": 4,
    "graph_partition_reads": 4,
    "lineage_start": 2,
    "rank_task_runs": 24,
    "recomputed_task_attempts": [
      "iteration-3-partition-0",
      "iteration-3-partition-1",
      "iteration-3-partition-2",
      "iteration-3-partition-3",
      "iteration-4-partition-0",
      "iteration-4-partition-1",
      "iteration-4-partition-2",
      "iteration-4-partition-3"
    ],
    "result": {
      "0": 1.554067,
      "1": 0.644089,
      "2": 1.651844,
      "3": 0.15
    },
    "trace": [
      "cache invariant graph partitions",
      "iteration 1: shuffle contributions, compute four rank partitions",
      "iteration 2: shuffle contributions, compute four rank partitions",
      "checkpoint iteration 2: materialize four rank partitions",
      "iteration 3: shuffle contributions, compute four rank partitions",
      "iteration 4: shuffle contributions, compute four rank partitions",
      "lost rank partition 2: replay from durable boundary 2"
    ]
  },
  "spark_without_checkpoint": {
    "duplicate_side_effect_keys": [
      "iteration-1-partition-0",
      "iteration-1-partition-1",
      "iteration-1-partition-2",
      "iteration-1-partition-3",
      "iteration-2-partition-0",
      "iteration-2-partition-1",
      "iteration-2-partition-2",
      "iteration-2-partition-3",
      "iteration-3-partition-0",
      "iteration-3-partition-1",
      "iteration-3-partition-2",
      "iteration-3-partition-3",
      "iteration-4-partition-0",
      "iteration-4-partition-1",
      "iteration-4-partition-2",
      "iteration-4-partition-3"
    ],
    "durable_partition_writes": 0,
    "graph_partition_reads": 4,
    "lineage_start": 0,
    "rank_task_runs": 32,
    "recomputed_task_attempts": [
      "iteration-1-partition-0",
      "iteration-1-partition-1",
      "iteration-1-partition-2",
      "iteration-1-partition-3",
      "iteration-2-partition-0",
      "iteration-2-partition-1",
      "iteration-2-partition-2",
      "iteration-2-partition-3",
      "iteration-3-partition-0",
      "iteration-3-partition-1",
      "iteration-3-partition-2",
      "iteration-3-partition-3",
      "iteration-4-partition-0",
      "iteration-4-partition-1",
      "iteration-4-partition-2",
      "iteration-4-partition-3"
    ],
    "result": {
      "0": 1.554067,
      "1": 0.644089,
      "2": 1.651844,
      "3": 0.15
    },
    "trace": [
      "cache invariant graph partitions",
      "iteration 1: shuffle contributions, compute four rank partitions",
      "iteration 2: shuffle contributions, compute four rank partitions",
      "iteration 3: shuffle contributions, compute four rank partitions",
      "iteration 4: shuffle contributions, compute four rank partitions",
      "lost rank partition 2: replay from durable boundary 0"
    ]
  },
  "workload": {
    "algorithm": "PageRank-like update",
    "iterations": 4,
    "partitions": 4
  }
}