sparkplug.core — amperity/sparkplug-core 0.1.2

This namespace provides the main API for writing Spark tasks.

Most operations in this namespace place the RDD last in the argument list, just like Clojure collection functions. This lets you compose them using the thread-last macro (->>), making it simple to migrate existing Clojure code.

aggregate clj

(aggregate aggregator combiner zero rdd)

Aggregate the elements of each partition in rdd using aggregator, then merge the results for all partitions using combiner. Both functions will be seeded with the neutral zero value.

This is an action that causes computation.

broadcast clj

(broadcast spark-context value)

Broadcast a read-only variable to the cluster, returning a reference for reading it in distributed functions. The variable data will be sent to each cluster only once.

The returned broadcast value can be resolved with deref or the @ reader macro.

cartesian clj

(cartesian rdd1 rdd2)

Construct an RDD representing the cartesian product of two RDDs. Returns a new pair RDD containing all combinations of elements between the datasets.

cogroup clj

(cogroup rdd1 rdd2)

(cogroup rdd1 rdd2 rdd3)

(cogroup rdd1 rdd2 rdd3 rdd4)

Produe a new RDD containing an element for each key k in the given pair RDDs mapped to a tuple of the values from all RDDs as lists.

If the input RDDs have types (K, A), (K, B), and (K, C), the grouped RDD will have type (K, (list(A), list(B), list(C))).

cogroup-partitioned clj

(cogroup-partitioned rdd1 rdd2 partitions)

(cogroup-partitioned rdd1 rdd2 rdd3 partitions)

(cogroup-partitioned rdd1 rdd2 rdd3 rdd4 partitions)

Produe a new RDD containing an element for each key k in the given pair RDDs mapped to a tuple of the values from all RDDs as lists. The resulting RDD partitions may be controlled by setting partitions to an integer number or a Partitioner instance.

If the input RDDs have types (K, A), (K, B), and (K, C), the grouped RDD will have type (K, (List(A), List(B), List(C))).

collect clj

(collect rdd)

Collect the elements of rdd into a vector on the driver. Be careful not to realize large datasets with this, as the driver will likely run out of memory.

This is an action that causes computation.

combine-by-key clj

(combine-by-key seq-fn conj-fn merge-fn rdd)

(combine-by-key seq-fn conj-fn merge-fn num-partitions rdd)

Combine the elements for each key using a set of aggregation functions.

If rdd contains pairs of (K, V), the resulting RDD will contain pairs of type (K, C). Callers must provide three functions:

• seq-fn which turns a V into a C (for example, vector)
• conj-fn to add a V to a C (for example, conj)
• merge-fn to combine two C's into a single result

count clj

(count rdd)

Count the number of elements in rdd.

This is an action that causes computation.

count-by-key clj

(count-by-key rdd)

Count the distinct key values in rdd. Returns a map of keys to integer counts.

This is an action that causes computation.

count-by-value clj

(count-by-value rdd)

Count the distinct values in rdd. Returns a map of values to integer counts.

This is an action that causes computation.

distinct clj

(distinct rdd)

(distinct num-partitions rdd)

Construct an RDD containing only a single copy of each distinct element in rdd. Optionally accepts a number of partitions to size the resulting RDD with.

filter clj

(filter f rdd)

Filter the elements of rdd to the ones which satisfy the predicate f.

first clj

(first rdd)

Find the first element of rdd.

This is an action that causes computation.

fold clj

(fold f zero rdd)

Aggregate the elements of each partition in rdd, followed by the results for all the partitions, by using the given associative function f and a neutral zero value.

This is an action that causes computation.

foreach clj

(foreach f rdd)

Apply the function f to all elements of rdd. The function will run on the executors where the data resides.

Consider foreach-partition for efficiency if handling an element requires costly resource acquisition such as a database connection.

This is an action that causes computation.

foreach-partition clj

(foreach-partition f rdd)

Apply the function f to all elements of rdd by calling it with a sequence of each partition's elements. The function will run on the executors where the data resides.

This is an action that causes computation.

full-outer-join clj

(full-outer-join rdd1 rdd2)

(full-outer-join rdd1 rdd2 partitions)

Perform a full outer join of rdd1 and rdd2.

For each element (k, v) in rdd1, the resulting RDD will either contain all pairs (k, (Some(v), Some(w))) for (k, w) in rdd2, or the pair (k, (Some(v), None)) if no elements in other have key k. Similarly, for each element (k, w) in rdd2, the resulting RDD will either contain all pairs (k, (Some(v), Some(w))) for v in rdd1, or the pair (k, (None, Some(w))) if no elements in rdd1 have key k.

Hash-partitions the resulting RDD using the existing partitioner/parallelism level unless partitions is be provided as an integer number or a partitioner instance.

group-by clj

(group-by f rdd)

(group-by f num-partitions rdd)

Group the elements of rdd using a key function f. Returns a pair RDD with each generated key and all matching elements as a value sequence.

group-by-key clj

(group-by-key rdd)

(group-by-key num-partitions rdd)

Group the entries in the pair rdd by key. Returns a new pair RDD with one entry per key, containing all of the matching values as a sequence.

intersection clj

(intersection rdd1 rdd2)

Construct an RDD representing the intersection of elements which are in both RDDs.

into clj

(into coll rdd)

(into coll xf rdd)

Collect the elements of rdd into a collection on the driver. Behaves like clojure.core/into, including accepting an optional transducer. Automatically coerces Scala tuples into Clojure vectors.

Be careful not to realize large datasets with this, as the driver will likely run out of memory.

This is an action that causes computation.

join clj

(join rdd1 rdd2)

(join rdd1 rdd2 partitions)

Construct an RDD containing all pairs of elements with matching keys in rdd1 and rdd2. Each pair of elements will be returned as a tuple of (k, (v, w)), where (k, v) is in rdd1 and (k, w) is in rdd2.

Performs a hash join across the cluster. Optionally, partitions may be provided as an integer number or a partitioner instance to control the partitioning of the resulting RDD.

key-by clj

(key-by f rdd)

Creates pairs from the elements in rdd by using f to compute a key for each value.

keys clj

(keys rdd)

Transform rdd by replacing each pair with its key. Returns a new RDD representing the keys.

left-outer-join clj

(left-outer-join rdd1 rdd2)

(left-outer-join rdd1 rdd2 partitions)

Perform a left outer join of rdd1 and rdd2.

For each element (k, v) in rdd1, the resulting RDD will either contain all pairs (k, (v, Some(w))) for (k, w) in rdd2, or the pair (k, (v, None)) if no elements in rdd2 have key k.

Hash-partitions the resulting RDD using the existing partitioner/parallelism level unless partitions is be provided as an integer number or a partitioner instance.

lookup clj

(lookup rdd k)

Find all values in the rdd pairs whose keys is k. The key must be serializable with the Java serializer (not Kryo) for this to work.

This is an action that causes computation.

map clj

(map f rdd)

Map the function f over each element of rdd. Returns a new RDD representing the transformed elements.

map->pairs clj

(map->pairs f rdd)

Map the function f over each element of rdd. Returns a new pair RDD representing the transformed elements.

map-partitions clj

(map-partitions f rdd)

(map-partitions f preserve-partitioning? rdd)

Map the function f over each partition in rdd, producing a sequence of results. Returns an RDD representing the concatenation of all the partition results. The function will be called with an iterator of the elements of each partition.

map-partitions->pairs clj

(map-partitions->pairs f rdd)

(map-partitions->pairs f preserve-partitioning? rdd)

Map the function f over each partition in rdd, producing a sequence of key-value pairs. The function will be called with an iterator of the elements of the partition.

map-partitions-indexed clj

(map-partitions-indexed f rdd)

Map the function f over each partition in rdd, producing a sequence of results. Returns an RDD representing the concatenation of all the partition results. The function will be called with the partition index and an iterator of the elements of each partition.

map-vals clj

(map-vals f rdd)

Map the function f over each value of the pairs in rdd. Returns a new pair RDD representing the transformed pairs.

mapcat clj

(mapcat f rdd)

Map the function f over each element in rdd to produce a sequence of results. Returns an RDD representing the concatenation of all element results.

mapcat->pairs clj

(mapcat->pairs f rdd)

Map the function f over each element in rdd to produce a sequence of key-value pairs. Returns a new pair RDD representing the concatenation of all result pairs.

mapcat-vals clj

(mapcat-vals f rdd)

Map the function f over each value of the pairs in rdd to produce a collection of values. Returns a new pair RDD representing the concatenated keys and values.

max clj

(max rdd)

(max compare-fn rdd)

Find the maximum element in rdd in the ordering defined by compare-fn.

This is an action that causes computation.

min clj

(min rdd)

(min compare-fn rdd)

Find the minimum element in rdd in the ordering defined by compare-fn.

This is an action that causes computation.

reduce clj

(reduce f rdd)

Aggregate the elements of rdd using the function f. The reducing function must accept two arguments and should be commutative and associative so that it can be computed correctly in parallel.

This is an action that causes computation.

reduce-by-key clj

(reduce-by-key f rdd)

Aggregate the pairs of rdd which share a key by combining all of the values with the reducing function f. Returns a new pair RDD with one entry per unique key, holding the aggregated values.

right-outer-join clj

(right-outer-join rdd1 rdd2)

(right-outer-join rdd1 rdd2 partitions)

Perform a right outer join of rdd1 and rdd2.

For each element (k, w) in rdd2, the resulting RDD will either contain all pairs (k, (Some(v), w)) for (k, v) in rdd1, or the pair (k, (None, w)) if no elements in rdd1 have key k.

Hash-partitions the resulting RDD using the existing partitioner/parallelism level unless partitions is be provided as an integer number or a partitioner instance.

sample clj

(sample fraction rdd)

(sample fraction replacement? rdd)

(sample fraction replacement? seed rdd)

Generate a randomly sampled subset of rdd with roughly fraction of the original elements. Callers can optionally select whether the sample happens with replacement, and a random seed to control the sample.

sort-by-key clj

(sort-by-key rdd)

(sort-by-key ascending? rdd)

(sort-by-key compare-fn ascending? rdd)

(sort-by-key compare-fn ascending? num-partitions rdd)

Reorder the elements of rdd so that they are sorted according to their natural order or the given comparator f if provided. The result may be ordered ascending or descending, depending on ascending?.

subtract clj

(subtract rdd1 rdd2)

Remove all elements from rdd1 that are present in rdd2.

subtract-by-key clj

(subtract-by-key rdd1 rdd2)

Construct an RDD representing all pairs in rdd1 for which there is no pair with a matching key in rdd2.

take clj

(take n rdd)

Take the first n elements of the RDD.

This currently scans the partitions one by one on the driver, so it will be slow if a lot of elements are required. In that case, use collect to get the whole RDD instead.

This is an action that causes computation.

take-ordered clj

(take-ordered n rdd)

(take-ordered n compare-fn rdd)

Take the first n (smallest) elements from this RDD as defined by the elements' natural order or specified comparator.

This currently scans the partitions one by one on the driver, so it will be slow if a lot of elements are required. In that case, use collect to get the whole RDD instead.

This is an action that causes computation.

union clj

(union rdd1 rdd2)

(union rdd1 rdd2 & rdds)

Construct a union of the elements in the provided RDDs. Any identical elements will appear multiple times.

vals clj

(vals rdd)

Transform rdd by replacing each pair with its value. Returns a new RDD representing the values.

zip-indexed clj

(zip-indexed rdd)

Zip the elements in rdd with their indices. Returns a new pair RDD with the element/index tuples.

The ordering is first based on the partition index and then the ordering of items within each partition. So the first item in the first partition gets index 0, and the last item in the last partition receives the largest index.

This method needs to trigger a spark job when rdd contains more than one partition.

zip-unique-ids clj

(zip-unique-ids rdd)

Zip the elements in rdd with unique long identifiers. Returns a new pair RDD with the element/id tuples.

Items in the kth partition will get ids k, n+k, 2*n+k, ..., where n is the number of partitions. So the ids won't be sequential and there may be gaps, but this method won't trigger a spark job, unlike zip-indexed.