It only supports on table stats but not on per partitions stats (incremental stats), it says in your link: "For non-incremental COMPUTE STATS statement, the columns for which statistics are computed can be specified with an optional comma-separated list of columns." So it looks like column specific is only on a table without partitions (non-incremental)   It really strange that it works only in this way ... View more

@EricL @eMazarakis We have tables with lots of non-filtered columns, so I know we don't want to collect statistics on them. Impala docs say that: "For a table with a huge number of partitions and many columns, the approximately 400 bytes of metadata per column per partition can add up to significant memory overhead, as it must be cached on the CatalogD host and on every ImpalaD host that is eligible to be a coordinator. If this metadata for all tables combined exceeds 2 GB, you might experience service downtime."   so for me, it's strange user don't have options to minimize the statistics on tables.  Hive has this option but if I use it it won't sync to Impala: "If you run the Hive statement ANALYZE TABLE COMPUTE STATISTICS FOR COLUMNS, Impala can only use the resulting column statistics if the table is unpartitioned. Impala cannot use Hive-generated column statistics for a partitioned table." ... View more

Thanks @eMazarakis  but I mean stats on specific partitions AND specific columns (BOTH). If I'll run as you suggest It will collect statistics on all the columns which we don't want, So how can I collect stats on specific partitions AND specific columns?  ... View more

Hi,   It's an internal table with 4 levels of partitions, But still, when I remove the partition the data and metadata is deleted but not the folder.   Thanks ... View more

Hi @Tim Armstrong , This is the output of SHOW FILES on the specific partition the query failed on (it failed on) hdfs://HadoopCluster/user/database/table_name/partition_value=KS5021/part-m-00000.snappy 2.74GB partition_value=KS5021 hdfs://HadoopCluster/user/database/table_name/partition_value=KS5021/part-m-00001.snappy 3.20GB partition_value=KS5021 hdfs://HadoopCluster/user/database/table_name/partition_value=KS5021/part-m-00002.snappy 3.55GB partition_value=KS5021 hdfs://HadoopCluster/user/database/table_name/partition_value=KS5021/part-m-00003.snappy 3.19GB partition_value=KS5021   This is the version: impalad version 2.12.0-cdh5.15.1 RELEASE (build 64f4e19bf59fab8664ebff7e80fc70570dcd8cb8) Built on Thu Aug  9 09:21:02 PDT 2018     Thanks ... View more

We don't have any critical issues. We just saw in other systems (Cassandra, Kafka etc) that G1GC brought better performance and fewer problems so we thought to use it also for CDH, but I see from your answer it is not a big change. Thanks! ... View more