xiarixiaoyao commented on a change in pull request #2722: URL: https://github.com/apache/hudi/pull/2722#discussion_r615508116
########## File path: hudi-hadoop-mr/src/main/java/org/apache/hudi/hadoop/realtime/HoodieParquetRealtimeInputFormat.java ########## @@ -85,12 +85,14 @@ void addProjectionToJobConf(final RealtimeSplit realtimeSplit, final JobConf job // risk of experiencing race conditions. Hence, we synchronize on the JobConf object here. There is negligible // latency incurred here due to the synchronization since get record reader is called once per spilt before the // actual heavy lifting of reading the parquet files happen. - if (jobConf.get(HoodieInputFormatUtils.HOODIE_READ_COLUMNS_PROP) == null) { + if (jobConf.get(HoodieInputFormatUtils.HOODIE_READ_COLUMNS_PROP) == null + || (!realtimeSplit.getDeltaLogPaths().isEmpty() && !HoodieRealtimeInputFormatUtils.requiredProjectionFieldsExistInConf(jobConf))) { synchronized (jobConf) { LOG.info( "Before adding Hoodie columns, Projections :" + jobConf.get(ColumnProjectionUtils.READ_COLUMN_NAMES_CONF_STR) + ", Ids :" + jobConf.get(ColumnProjectionUtils.READ_COLUMN_IDS_CONF_STR)); - if (jobConf.get(HoodieInputFormatUtils.HOODIE_READ_COLUMNS_PROP) == null) { + if (jobConf.get(HoodieInputFormatUtils.HOODIE_READ_COLUMNS_PROP) == null Review comment: Clone the Configuration object can be very expensive。 To avoid unexpected performance regressions for workloads, we should not isolation the jobconf for different recordreader i also agree with that revert the https://github.com/apache/hudi/pull/2190/files. however if current query does not involve any log files, adding hoodie additional projection columns will lead unnecessary io,since we have scanned hoodie additional projection columns . -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org