[ https://issues.apache.org/jira/browse/SPARK-27228?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16800621#comment-16800621 ]
Lukas Waldmann commented on SPARK-27228: ---------------------------------------- That's a good question :) What code does it runs up to several hundreds of sql queries with different parameters and union the results before writing the result to Hive table. Input are Hive tables with up to several hundreds million lines code looks something like this: {code:java} void process(String dbName, String environment) { //For all items in call the sql snippet and union the results List<Metadata> mds = ...; Map<String, Dataset> res = new LinkedHashMap<>(); mds.stream() .forEach(md -> { try (InputStream is = getClass().getResourceAsStream("/" + md.query_id)) { String snippet = IOUtils.toString(is); Dataset ds = spark.sql(snippet); String key = md.product; res.put(key, res.get(key) == null ? ds : ds.union(res.get(key))); } catch (IOException ex) { Logger.getLogger(SparkMainApp.class.getName()).log(Level.SEVERE, null, ex); } }); String name = dbName + "." + table; res.values().stream() .forEach( result -> { result.repartition( result.col(PRODUCT.toString()), result.col(PROTOCOL.toString())).write() .mode(SaveMode.Overwrite).insertInto(name); } ); } {code} > Spark long delay on close, possible problem with killing executors > ------------------------------------------------------------------ > > Key: SPARK-27228 > URL: https://issues.apache.org/jira/browse/SPARK-27228 > Project: Spark > Issue Type: Bug > Components: Block Manager > Affects Versions: 2.3.0 > Reporter: Lukas Waldmann > Priority: Major > Attachments: log.html > > > When using dynamic allocations after all jobs finishes spark delays for > several minutes before finally finishes. Log suggest that executors are not > cleared up properly. > See the attachment for log > -- This message was sent by Atlassian JIRA (v7.6.3#76005) --------------------------------------------------------------------- To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org For additional commands, e-mail: issues-h...@spark.apache.org