Gets broadcast threshold in propagating component assignment.

Gets checkpoint interval.

Gets storage level for intermediate datasets that require multiple passes.

Gets whether to use vertex labels as component identifiers.

Gets whether local checkpoints are being used instead of regular checkpoints.

The max number of iterations of algorithm to be performed.

Sets the algorithm to use for computing connected components. Supported values:

• ConnectedComponents.ALGO_GRAPHX: use the GraphX implementation
• ConnectedComponents.ALGO_GRAPHFRAMES: deprecated alias for ConnectedComponents.ALGO_TWO_PHASE
• ConnectedComponents.ALGO_TWO_PHASE: use the two-phase label propagation implementation
• ConnectedComponents.ALGO_RANDOMIZED_CONTRACTION: use the randomized contraction implementation

Sets a broadcast threshold in propagating component assignments (default: 1,000,000). If a node degree is greater than this threshold at some iteration, its component assignment will be collected and then broadcasted back to propagate the assignment to its neighbors. Otherwise, the assignment propagation is done by a normal Spark join. This parameter is only used when the algorithm is set to "graphframes". If the value is -1, then the skewness problem is left to the Apache Spark AQE optimizer.

**WARNING** using a broadcast threshold is non-free! Under the hood it is calling an action, and if a broadcast threshold is set, then AQE is disabled to avoid wrong results! If your graph does not contain gigantic components, it is strongly recommended to set this value to -1. On benchmarks setting it to -1 gains about x5 better results in performance.

**WARNING** the current default value is 1,000,000. It is left for backward compatibility only. In the future versions it may be set to -1 as more reasonable for the most real-world cases (e.g., the data deduplication problem).

Sets checkpoint interval in terms of number of iterations (default: 2). Checkpointing regularly helps recover from failures, clean shuffle files, shorten the lineage of the computation graph, and reduce the complexity of plan optimization. As of Spark 2.0, the complexity of plan optimization would grow exponentially without checkpointing. Hence, disabling or setting longer-than-default checkpoint intervals are not recommended. Checkpoint data is saved under org.apache.spark.SparkContext.getCheckpointDir with prefix of the algorithm name. If the checkpoint directory is not set, this throws a java.io.IOException. Set a nonpositive value to disable checkpointing. This parameter is only used when the algorithm is set to "graphframes". Its default value might change in the future.

Sets storage level for intermediate datasets that require multiple passes (default: MEMORY_AND_DISK).

!! WARNING: INTERNAL API — FOR VERY EXPERIENCED USERS ONLY !!

Sets whether the graph has already been prepared before being passed to the algorithm, skipping the internal graph preparation step. The default is false, meaning the algorithm will always prepare the graph itself, which is the safe and recommended behaviour.

Only set this to true if you have already performed all required preparation steps yourself and you fully understand what those steps are for the specific algorithm you are using. The preparation requirements differ significantly between algorithms:

• two_phase and randomized_contraction each require their own distinct preparation steps. These are NOT interchangeable. You MUST study the internal source code of the algorithm you intend to use and replicate its exact preparation logic before enabling this flag.

Incorrect use of this flag WILL produce silently wrong results with no error or warning at runtime. There is no validation that the graph has been correctly prepared. You are entirely responsible for ensuring correctness.

Sets whether to use vertex labels as component identifiers (default: false). When true, vertex labels will be used as component identifiers instead of computing connected components.

Sets whether to use local checkpoints instead of regular checkpoints (default: false). Local checkpoints are faster but less reliable as they don't survive node failures.