[
https://issues.apache.org/jira/browse/IGNITE-10798?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16729536#comment-16729536
]
Vladimir Ozerov commented on IGNITE-10798:
------------------------------------------
[~sergi.vladykin], looks good to me in general except of test coverage and
integration with various components. Namely:
# Basic SQL test coverage is needed
# {{DmlStatementsProcessor}} and MVCC remote query execution
({{GridNearTxQueryEnlistRequest}}) - it is not very clear whether scan flag is
propagated further or not; tests will help us answer the question
# In "lazy" mode we may "jump" between different threads during query
execution. This is not the case now, as lazy query is executed in a separate
thread, but will become true in the nearest time once {{IGNITE-9171}} is ready.
In this case we will need to set {{scan}} flag on every "next page" request,
rather than on "execute with timer". Likewise, in MVCC + lazy mode, which is
not implemented yet (IGNITE-10829), we will need to set {{scan}} flag after
every query resume when next locked entry becomes available. As both features
are not in master yet, we cannot fix it in the scope of this ticket, but at the
very least we need to cover this with tests confirming scan mode usage, so that
it is not missed in future
# Probably we also need tests for IO statistics to make sure that read blocks
are still counted properly in {{scan}} mode. See {{IoStatistics*SelfTest}}
classes.
# This flag should be available from JDBC and ODBC drivers as well. I propose
to create separate tickets and link to this one
# We need to better JavaDoc implications of this mode in {{ScanQuery}} and
{{SqlFieldsQuery}} classes, so that user understand when to use it. For scan
query things are more or less clear - we may miss some updates, or have
duplicates. But what are implications for SQL query? If there is nothing
sinister with respect to current SQL guarantees, may be it makes sense to
enable the flag by default?
# Speaking of management perspective - is it possible reflect that direct scan
is used in EXPLAIN output?
# For MVCC mode I would add concurrent tests confirming that no duplicates
appear when scan is performed concurrently with updates
[~agoncharuk], [~gvvinblade], could you please review storage and MVCC parts
respectively?
Vladimir.
> Data page scan for ScanQuery, SqlQuery and SqlFieldsQuery
> ---------------------------------------------------------
>
> Key: IGNITE-10798
> URL: https://issues.apache.org/jira/browse/IGNITE-10798
> Project: Ignite
> Issue Type: Improvement
> Components: cache, sql
> Reporter: Sergi Vladykin
> Assignee: Vladimir Ozerov
> Priority: Major
> Fix For: 2.8
>
>
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
