[jira] [Commented] (PHOENIX-6983) Add hint to disable server merges for uncovered index queries
[ https://issues.apache.org/jira/browse/PHOENIX-6983?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=17741473#comment-17741473 ] Istvan Toth commented on PHOENIX-6983: -- No, [~larsh] . Previously, we've used the skip-scan-join execution plan for hinted uncovered queries. Recently, we've added the new sever merge execution plan for it (the same as the new uncovered indexes use), but that is not optimal in some cases, and also has bugs, some of which have been fixed, but I am still working on one known bug, which is blocking a customer. Specifying this hint will cause Phoenix to use the old query plan, which still uses the index, but in a different way. > Add hint to disable server merges for uncovered index queries > - > > Key: PHOENIX-6983 > URL: https://issues.apache.org/jira/browse/PHOENIX-6983 > Project: Phoenix > Issue Type: Improvement > Components: core >Affects Versions: 5.2.0, 5.1.3 >Reporter: Istvan Toth >Assignee: Istvan Toth >Priority: Major > Fix For: 5.2.0, 5.1.4 > > > In certain cases, the new server merge code is far less efficient than the > old skip-scan-merge code path. > Specifically, when a large number of rows is matched on the index table, then > each of those rows has to be resolved from the data table, and the filtering > must be done on the index RS. > With the old code path, these filters were pushed to the data table, and > processed in parallel, with much better performance. -- This message was sent by Atlassian Jira (v8.20.10#820010)
[jira] [Commented] (PHOENIX-6983) Add hint to disable server merges for uncovered index queries
[ https://issues.apache.org/jira/browse/PHOENIX-6983?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=17737302#comment-17737302 ] Lars Hofhansl commented on PHOENIX-6983: isn't this the same as hinting not to use the index in question, or hinting a full scan? > Add hint to disable server merges for uncovered index queries > - > > Key: PHOENIX-6983 > URL: https://issues.apache.org/jira/browse/PHOENIX-6983 > Project: Phoenix > Issue Type: Improvement > Components: core >Affects Versions: 5.2.0, 5.1.3 >Reporter: Istvan Toth >Assignee: Istvan Toth >Priority: Major > Fix For: 5.2.0, 5.1.4 > > > In certain cases, the new server merge code is far less efficient than the > old skip-scan-merge code path. > Specifically, when a large number of rows is matched on the index table, then > each of those rows has to be resolved from the data table, and the filtering > must be done on the index RS. > With the old code path, these filters were pushed to the data table, and > processed in parallel, with much better performance. -- This message was sent by Atlassian Jira (v8.20.10#820010)
[jira] [Commented] (PHOENIX-6983) Add hint to disable server merges for uncovered index queries
[ https://issues.apache.org/jira/browse/PHOENIX-6983?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=17734075#comment-17734075 ] Istvan Toth commented on PHOENIX-6983: -- One possible improvement could be pushing these server filters to the data table scans, but it's much easier and faster to provide a fallback to the old execution plan in the short term. > Add hint to disable server merges for uncovered index queries > - > > Key: PHOENIX-6983 > URL: https://issues.apache.org/jira/browse/PHOENIX-6983 > Project: Phoenix > Issue Type: Improvement > Components: core >Affects Versions: 5.2.0, 5.1.3 >Reporter: Istvan Toth >Priority: Major > > In certain cases, the new server merge code is far less efficient than the > old skip-scan-merge code path. > Specifically, when a large number of rows is matched on the index table, then > each of those rows has to be resolved from the data table, and the filtering > must be done on the index RS. > With the old code path, these filters were pushed to the data table, and > processed in parallel, with much better performance. -- This message was sent by Atlassian Jira (v8.20.10#820010)
