- Fix not breaking out of the VIEWS-infinite-loop when the "SHOULD_EXIT_WHOLE_SCRIPT_UPON_ERROR" is set to "false".
- Exit the script when no HDFS-active-node was found, independently of the "SHOULD_EXIT_WHOLE_SCRIPT_UPON_ERROR".
- Fix view_name-recognition in a log-message, by using the more advanced "Perl-Compatible Regular Expressions" in "grep".
- Add error-handling for "compute stats" errors.
- Assign the WRITE and EXECUTE permissions to the DBs' HDFS-directories, in order to be able to create tables on top of them, in the Impala Cluster.
- Make sure the "copydb" function returns early, when it encounters a fatal error, while respecting the "SHOULD_EXIT_WHOLE_SCRIPT_UPON_ERROR" config.
- Add a configuration-"switch" to control whether the script exits upon an error or not.
- Allow the script to exit when a table could not be created.
- Show the elapsed time for processing each database.
Update the code of acquiring the entities from Ocean cluster, through hive, in order to optimize the process and account for additional reserved keywords in Impala.
Co-authored-by: Antonis Lempesis <antleb@di.uoa.gr>
- Update the algorithm for creating views that depend on other views; overcome some bash-instabilities.
- Upon any error, fail the whole process, not just the current DB-creation, as those errors usually indicate a bug in the initial DB-creation, that should be fixed immediately.
- Enhance parallel-copy of large files by "hadoop distcp" command.
- Reduce the "invalidate metadata" commands to just the current DB's tables, in order to eliminate the general overhead on Impala.
- Show the number of tables and views in the logs.
- Fix some log-messages.
- Set Steps 2-7 and 9 to limit the amount of files generated by Spark, from 8000, down to 100, to improve file-transfer and querying performance.
- Allow the workflow to run up to Step10. The Step11 seems to have some issues even when using hive-action.
- Update the algorithm for creating views that depend on other views.
- Add check for successful execution of the "hadoop distcp" command.
- Add a check for successful copy operation of all entities.
- Upon facing an error in a DB, exit the method, instead of the whole script.
- Improve logging.
- Code polishing.
- Use only hive commands in the Ocean Cluster, as the "impala-shell" will be removed from there to free-up resources.
- Hugely improve the performance in every aspect of the copying process: a) speedup file-transferring and DB-deletion, b) eliminate permissions-assignment, "load" operations and "use $db" queries, c) retry only the "create view" statements and only as long as they depend on other non-created views, instead of trying to recreate all tables and views 5 consecutive times.
- Add error-checks for the creation of tables and views.