[ https://issues.apache.org/jira/browse/IGNITE-5571?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16085618#comment-16085618 ]
Vladimir Ozerov commented on IGNITE-5571: ----------------------------------------- [~al.psc], my comments: 1) {{AffinityTopologyVersion}} is never passed from DML processor. Looks like a bug to me. IMO, we should not pass it to {{querySqlFields}} at all. Instead, we should ask {{GridQueryProcessor}} for backup filter when we understand it is needed. Let's remove {{AffinityTopologyVersion}} argument. 2) Please no {{T*}} classes. This is antipattern. Normal value class should be created instead. 3) {{IgniteH2Indexing:1376}} - incorrect {{distributedJoins}} is set here. I do not know whether it has any adverse affects, but let's make this code cleaner. We do not need {{GridH2QueryContext}} at this point. Let's set only after we calculated final value for {{distributedJoins}} flag. > Make sure that cache-less execution works as good as cache-based > ---------------------------------------------------------------- > > Key: IGNITE-5571 > URL: https://issues.apache.org/jira/browse/IGNITE-5571 > Project: Ignite > Issue Type: Task > Components: sql > Reporter: Vladimir Ozerov > Assignee: Alexander Paschenko > Priority: Critical > Fix For: 2.2 > > > Compare the following two methods: > 1) {{GridQueryProcessor.querySqlFields}} - old good entry point for query > execution; > 2) {{GridQueryProcessor.querySqlFieldsNoCache}} - new method for "cache-less" > execution. > Note how cache context is used in the first method: > 1) First, it helps determine whether query can be converted to "local" > 2) Second, it gets query parallelism of current cache, and if it differs from > {{1}}, then it turns on {{distributedJoins}}. > Neither of this happens in the second implementation. Moreover, I had to > throw an exception for local queries, as I didn't know how to handle them > properly. > We need to investigate and fix these two deficiencies somehow. Probably some > inputs from [~sergi.vladykin] would be required, to understand what is going > on. > Our ultimate goal is to make "cache-less" execution as good as the old one. -- This message was sent by Atlassian JIRA (v6.4.14#64029)