[
https://issues.apache.org/jira/browse/PHOENIX-4712?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16459980#comment-16459980
]
James Taylor commented on PHOENIX-4712:
---------------------------------------
My current patch breaks non IT unit tests (that don't go against an actual
cluster), because they rely on the adding of the tables to the cache (which is
the source of truth in these tests). One way to solve this would be to have a
ConnectionQueryServices method to maintain the cache (or at least a
ConnectionQueryServices method that determines whether we add to the cache or
not). Needs a bit more thinking. My inclination is that this isn't a 4.14
showstopper since it works fine when the query comes through a different
connection than the one in which the index was added. WDYT, [~brfrn169]?
> When creating an index on a table, meta data cache of views related to the
> table isn't updated
> ----------------------------------------------------------------------------------------------
>
> Key: PHOENIX-4712
> URL: https://issues.apache.org/jira/browse/PHOENIX-4712
> Project: Phoenix
> Issue Type: Bug
> Reporter: Toshihiro Suzuki
> Assignee: Toshihiro Suzuki
> Priority: Major
> Attachments: PHOENIX-4712-v2.patch, PHOENIX-4712.patch,
> PHOENIX-4712_v3.patch
>
>
> Steps to reproduce are as follows:
> 1. Create a table
> {code}
> create table tbl (col1 varchar primary key, col2 varchar);
> {code}
> 2. Create a view on the table
> {code}
> create view vw (col3 varchar) as select * from tbl;
> {code}
> 3. Create a index on the table
> {code}
> create index idx ON tbl (col2);
> {code}
> After those, when issuing a explain query like the following, it seems like
> the query doesn't use the index, although the index should be used:
> {code}
> 0: jdbc:phoenix:> explain select /*+ INDEX(vw idx) */ * from vw where col2 =
> 'aaa';
> +---------------------------------------------------------------+
> | PLAN |
> +---------------------------------------------------------------+
> | CLIENT 1-CHUNK PARALLEL 1-WAY ROUND ROBIN FULL SCAN OVER TBL |
> | SERVER FILTER BY COL2 = 'aaa' |
> +---------------------------------------------------------------+
> {code}
> However, after restarting sqlline, the explain output is changed, and the
> index is used.
> {code}
> 0: jdbc:phoenix:> explain select /*+ INDEX(vw idx) */ * from vw where col2 =
> 'aaa';
> +--------------------------------------------------------------------------------+
> | PLAN
> |
> +--------------------------------------------------------------------------------+
> | CLIENT 1-CHUNK PARALLEL 1-WAY ROUND ROBIN FULL SCAN OVER TBL
> |
> | SKIP-SCAN-JOIN TABLE 0
> |
> | CLIENT 1-CHUNK PARALLEL 1-WAY ROUND ROBIN RANGE SCAN OVER IDX
> ['aaa'] |
> | SERVER FILTER BY FIRST KEY ONLY
> |
> | DYNAMIC SERVER FILTER BY "VW.COL1" IN ($3.$5)
> |
> +--------------------------------------------------------------------------------+
> {code}
> I think when creating an index on a table, meta data cache of views related
> to the table isn't updated, so the index isn't used for that query. However
> after restarting sqlline, the meta data cache is refreshed, so the index is
> used.
> When creating an index on a table, we should update meta data cache of views
> related to the table.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
