[GitHub] [hudi] n3nash commented on issue #2095: Inserts in partitioned MoR RO view visible without compaction
n3nash commented on issue #2095: URL: https://github.com/apache/hudi/issues/2095#issuecomment-696212548 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
[GitHub] [hudi] n3nash commented on issue #2095: Inserts in partitioned MoR RO view visible without compaction
n3nash commented on issue #2095: URL: https://github.com/apache/hudi/issues/2095#issuecomment-696259058 @vishalpathak1986 Hudi internally maintains an Index to tag the incoming records with the fileId that it maps to. If inserts are written to log files, we require a way from the index to know which log file a particular record was written to. We have indexes such as BloomIndex which are written to parquet files and hence we are able to figure out if a record is present in a parquet file or not, but we don't have such an index for log files. There is no config to turn off writing inserts as parquet, you just have to use an index implementation that can index log files. Currently, only the HbaseIndex can index log files -> https://github.com/apache/hudi/blob/c8e19e2def0c33415bc3945ffb81f524c484c924/hudi-client/src/main/java/org/apache/hudi/index/hbase/HBaseIndex.java#L483. In the future, the record level index I pointed out earlier will be able to index the log files which will eliminate the need of an external K-V store for this feature. 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
[GitHub] [hudi] n3nash commented on issue #2095: Inserts in partitioned MoR RO view visible without compaction
n3nash commented on issue #2095: URL: https://github.com/apache/hudi/issues/2095#issuecomment-696212548 @vishalpathak1986 Currently, Hudi supports writing inserts in columnar file fomat (parquet) for MOR tables. All inserts goto parquet while updates goto AVRO file. This is done for 2 reasons a) If you only have inserts, you don't have to compact again and have your data written in columnar file format to start with b) Absence of an index that can index log file. This feature will soon be supported with -> https://cwiki.apache.org/confluence/display/HUDI/RFC+-+08+%3A+Record+level+indexing+mechanisms+for+Hudi+datasets or you can try the [HbaseIndex](https://github.com/apache/hudi/blob/master/hudi-client/src/main/java/org/apache/hudi/index/hbase/HBaseIndex.java) in the meantime which requires a Hbase cluster. 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