gbif / pipelines

Pipelines for data processing (GBIF and LivingAtlases)
Apache License 2.0
40 stars 28 forks source link

Optimising Data Cleaning for files used for GBIF downloads #1077

Open fmendezh opened 5 months ago

fmendezh commented 5 months ago

Multiple data cleaning routines are currently applied during the production of occurrence downloads. These routines can be moved to an earlier stage when the HDFS View files are created. This change could reduce the total time needed to produce downloads and also facilitate incremental table builds. The routines that can be applied are encapsulated as Spark UDFs in the occurrence-table-build-spark project. The fields to which these routines are applied are controlled by the HiveQueries utility class. Additionally, all verbatim fields are initialized by cleaning delimiters in the OccurrenceHDFSTableDefinition.verbatimField(Term term) method, and String values are also initialized in the OccurrenceHDFSTableDefinition.interpretedField(Term term) method.