diff options
Diffstat (limited to 'scalding/src/main')
-rw-r--r-- | scalding/src/main/scala/sandcrawler/MissingColumnDumpJob.scala | 67 |
1 files changed, 67 insertions, 0 deletions
diff --git a/scalding/src/main/scala/sandcrawler/MissingColumnDumpJob.scala b/scalding/src/main/scala/sandcrawler/MissingColumnDumpJob.scala new file mode 100644 index 0000000..cc3bf23 --- /dev/null +++ b/scalding/src/main/scala/sandcrawler/MissingColumnDumpJob.scala @@ -0,0 +1,67 @@ +package sandcrawler + +import java.util.Properties + +import cascading.property.AppProps +import cascading.tuple.Fields +import com.twitter.scalding._ +import com.twitter.scalding.typed.TDsl._ +import parallelai.spyglass.base.JobBase +import parallelai.spyglass.hbase.HBaseConstants.SourceMode +import parallelai.spyglass.hbase.HBasePipeConversions +import parallelai.spyglass.hbase.HBaseSource + +// This nasty, no-good, horrible Job outputs a list of keys ("sha1:A234...") +// for which the given "column" does not have a value set. +// It does this using a self-join because SpyGlass's HBase SCAN support seems +// to be extremely limited. +class MissingColumnDumpJob(args: Args) extends JobBase(args) with HBasePipeConversions { + + val output = args("output") + + val allKeys : TypedPipe[String] = MissingColumnDumpJob.getHBaseKeySource( + args("hbase-table"), + args("zookeeper-hosts")) + .read + .fromBytesWritable('key) + .toTypedPipe[String]('key) + + val existingKeys : TypedPipe[(String,Boolean)] = MissingColumnDumpJob.getHBaseColSource( + args("hbase-table"), + args("zookeeper-hosts"), + args("column")) + .read + .fromBytesWritable('key) + .toTypedPipe[String]('key) + .map{ key => (key, true) } + + val missingKeys : TypedPipe[String] = allKeys + .groupBy( identity ) + .leftJoin(existingKeys.groupBy(_._1)) + .toTypedPipe + .collect { case (key, (_, None)) => key } + + missingKeys + .write(TypedTsv[String](output)) + +} + +object MissingColumnDumpJob { + + // eg, "wbgrp-journal-extract-0-qa",7 "mtrcs-zk1.us.archive.org:2181" + def getHBaseColSource(hbaseTable: String, zookeeperHosts: String, col: String) : HBaseSource = { + HBaseBuilder.build( + hbaseTable, + zookeeperHosts, + List(col), + SourceMode.SCAN_ALL) + } + + def getHBaseKeySource(hbaseTable: String, zookeeperHosts: String) : HBaseSource = { + HBaseBuilder.build( + hbaseTable, + zookeeperHosts, + List("f:c"), + SourceMode.SCAN_ALL) + } +} |