Hi Josef, Does the client connect to impalads via a load-balancing proxy? If so, you might need to set SYNC_DDL=1 before triggering any DDLs.
BTW, 5 minutes seems too long. Do you see any warnings or errors in statestored's log? Which Impala version do you use? After Impala-2.9 (also distributed in CDH5.12), you can set up coordinator/executor roles for impalads ( https://www.cloudera.com/documentation/enterprise/5-12-x/topics/impala_dedicated_coordinator.html). Leave a few impalads to be coordinator and the others to be executor may help. Quanlong On Mon, Oct 1, 2018 at 10:46 PM <josef.zahn...@swisscom.com> wrote: > Hi guys > > > > I’m trying to do an ALTER TABLE for a Kudu range partitioned table. The > ALTER TABLE itself works fine, my issue is that most of the time I get a > delay of about 5min until the “show range partitions xyz” shows the > current/updated range partitions and we need this show command to > automatically update the table ranges and cut the range boundaries... > > > > Sometimes the delay is there and sometimes not - at least after a restart > of impala. I’ve tried several things, removed the kudu table, dropped the > table in impala and re-added it plus other things, but I don’t get a > reproducible result. sometimes it’s fine, but most of the time it’s not. > > > > If the issue persists, the two following commands doesn’t help to get the > table updated: `REFRESH` and `INVALIDATE METADAT` > > > > We have 33 Impala nodes. > > > > Any other ideas to solve the issue? Or is it well known. Thanks in advance > > > > Josef > > > > > > > > *Example:* > > ################################################## > > ## Kudu Partition Schema: > > ################################################## > > > > HASH (flowEndDate, uniqueID) PARTITIONS 16, > > RANGE (flowEndDate) ( > > PARTITION VALUES < 1537860190, > > PARTITION 1537860190 <= VALUES < 1537863790, > > PARTITION 1537863790 <= VALUES < 1537867390, > > PARTITION 1537867390 <= VALUES < 1537870990, > > PARTITION 1537870990 <= VALUES < 1537874590, > > PARTITION 1537874590 <= VALUES < 1537878190, > > PARTITION VALUES >= 1537878190 > > ) > > > > ################################################## > > ## show partitions > > ################################################## > > > > [:21000] > show range partitions test_sql_drop; > > Query: show range partitions test_sql_drop > > +-----------------------------------+ > > | RANGE (flowenddate) | > > +-----------------------------------+ > > | VALUES < 1537860190 | > > | 1537860190 <= VALUES < 1537863790 | > > | 1537863790 <= VALUES < 1537867390 | > > | 1537867390 <= VALUES < 1537870990 | > > | 1537870990 <= VALUES < 1537874590 | > > | 1537874590 <= VALUES < 1537878190 | > > | VALUES >= 1537878190 | > > +-----------------------------------+ > > Fetched 7 row(s) in 0.02s > > > > > > ################################################## > > ## drop a range partition > > ################################################## > > > > ALTER TABLE test_sql_drop DROP RANGE PARTITION VALUES < 1537860190; > > > > > > ################################################## > > ## between 0-5 minutes after "alter table". Kudu webgui updates the > Partition Schema instantly > > ################################################## > > > > [:21000] > show range partitions test_sql_drop; > > Query: show range partitions test_sql_drop > > +-----------------------------------+ > > | RANGE (flowenddate) | > > +-----------------------------------+ > > | VALUES < 1537860190 | > > | 1537860190 <= VALUES < 1537863790 | > > | 1537863790 <= VALUES < 1537867390 | > > | 1537867390 <= VALUES < 1537870990 | > > | 1537870990 <= VALUES < 1537874590 | > > | 1537874590 <= VALUES < 1537878190 | > > | VALUES >= 1537878190 | > > +-----------------------------------+ > > Fetched 7 row(s) in 0.02s > > > > ################################################## > > ## more than 5 minutes after "alter table". > > ################################################## > > > > the " show range partitions test_sql_drop;" is updated and correct >
