Boris Gitline created IMPALA-8301:
-------------------------------------
Summary: Eliminate need for SYNC_DDL in local catalog mode
Key: IMPALA-8301
URL: https://issues.apache.org/jira/browse/IMPALA-8301
Project: IMPALA
Issue Type: Improvement
Reporter: Boris Gitline
In the following scenario looks like the INSERT on coordinator 2 is gated
behind a long-running DDL on coordinator 1. That scenario still requires
SYNC_DDL even in metadata v2. We want to change the metadata handling design so
that coordinator 3 does not have to wait for the long-running DDL to complete –
it would render the correct result on the target table t1 reference.
Step1. coordinator 1
##*say*, the following compute stats runs about 100 seconds
compute stats tao_ddl_contention;
[steps 2 and 3 are performed while COMPUTE STATS is running]
Step2. coordinator 2
create another new table.
create table t1(c1 int);
insert into t1 select 1 ;
select * from t1;
[can see the inserted rows]
Step3. coordinator 3
query the newly inserted rows in t1 while the COMPUTE STATS is still running:
select * from t1;
[see no rows]
##the query in [Step3] won't show the row inserted by step2 until the first
step "compute stats" completed, unless:
- SYNC_DDL is set before the INSERT on coordinator 2, or
* when the step1 compute stats is completed, or
* you can see the data from this impala session, or
* a manual refresh of the t1 table.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
