[
https://issues.apache.org/jira/browse/IGNITE-29117?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=18124888#comment-18124888
]
Nikolay Izhikov edited comment on IGNITE-29117 at 10/8/26 8:10 AM:
-------------------------------------------------------------------
While preparing this change I checked whether the Calcite module has tests for
a node started *without* {{ignite-indexing}} on the classpath. It does not, so
I wrote one: [^CalciteOnlyNodeIntegrationTest.java] (full patch incl. the Maven
profile and suite registration: [^IGNITE-29117-calcite-only-tests.patch]).
The test starts a plain node with only {{ignite-calcite}} present and exercises
the public SQL entry points. Since the kernel detects H2 with
{{Class.forName}}, the only way to simulate its absence is to drop the jars
from the surefire classpath, hence the new {{calcite-only}} profile in
{{modules/calcite/pom.xml}}. On the regular classpath all tests are skipped by
assumption.
Run command:
{code}
./mvnw test -pl :ignite-calcite -Pcalcite-only,surefire-fork-count-1 \
-Dtest=CalciteOnlyNodeIntegrationTest -Dsurefire.failIfNoSpecifiedTests=false
{code}
Results on current master (f38a8b0412a), H2 excluded: *Tests run: 8, Failures:
3, Errors: 4, Skipped: 0*.
||Test||What it checks||Result on master||
|{{testSqlFieldsQuery}}|{{SqlFieldsQuery}} via cache API and thin JDBC:
{{QUERY_ENGINE()}} is {{calcite}}, {{CREATE TABLE}}, {{INSERT}}, {{SELECT}},
{{DROP TABLE}}|(/) pass|
|{{testSqlQuery}}|legacy {{SqlQuery}} via cache API|(x) {{NullPointerException:
Cannot invoke "GridQueryIndexing.generateFieldsQuery(...)" because "this.idx"
is null}}|
|{{testThinClientSqlQuery}}|legacy {{SqlQuery}} via thin client|(x) same NPE,
surfaced to the client as server error 50000|
|{{testTextQuery}}|{{TextQuery}} fails with a message naming
{{ignite-indexing}}|(x) {{Indexing is disabled.}} - module not named|
|{{testSetStreaming}}|{{SET STREAMING ON}} over thin JDBC fails with a message
naming {{ignite-indexing}}|(x) {{Failed to parse query. Encountered " "ON" "ON
""}} (Calcite parser), and then {{Connection.close()}} hangs forever|
|{{testCopy}}|{{COPY FROM ... FORMAT CSV}} over thin JDBC fails with a message
naming {{ignite-indexing}}|(x) {{Failed to parse query. Non-query expression
encountered in illegal context}} - module not named|
|{{testJdbcQueryEngineProperty}}|{{jdbc:ignite:thin://...?queryEngine=calcite}}
without explicit {{SqlConfiguration.queryEnginesConfiguration}}|(x) {{Handshake
failed ... Not found configuration for query engine: calcite}}|
|{{testQueryEngineHint}}|{{SELECT /\*+ QUERY_ENGINE('calcite') \*/ ...}}
without explicit engines configuration|(x) {{Query engines not configured, but
specified engine: calcite}}|
Without the profile (H2 on the classpath): *Tests run: 8, Skipped: 8*.
Checkstyle and RAT pass for {{:ignite-calcite}}.
One extra finding from {{testSetStreaming}}: a rejected {{SET STREAMING ON}}
leaves the thin JDBC driver with {{streamState}} set but never started, because
{{JdbcThinConnection.executeNative}} assigns the state before {{sendRequest}}.
{{close()}} then goes through {{StreamState.close0() -> executeBatch(true) ->
lastRespFut.get()}} and blocks forever. It is latent today (with H2 the command
never fails) but becomes reachable on a Calcite-only node, so it needs a fix as
part of this work. The test bounds the close with a 10s async wait so it fails
instead of hitting the 5 minute JUnit timeout.
These tests are intentionally submitted without fixes: they document the
current gap and should turn green as the H2-only paths are either ported to
core or rejected with precise errors.
was (Author: nizhikov):
While preparing this change I checked whether the Calcite module has tests for
a node started *without* \{{ignite-indexing}} on the classpath. It does not, so
I wrote one: [^CalciteOnlyNodeIntegrationTest.java] (full patch incl. the Maven
profile and suite registration: [^IGNITE-29117-calcite-only-tests.patch]).
The test starts a plain node with only \{{ignite-calcite}} present and
exercises the public SQL entry points. Since the kernel detects H2 with
\{{Class.forName}}, the only way to simulate its absence is to drop the jars
from the surefire classpath, hence the new \{{calcite-only}} profile in
\{{modules/calcite/pom.xml}}. On the regular classpath all tests are skipped by
assumption.
Run command:
{code}
./mvnw test -pl :ignite-calcite -Pcalcite-only,surefire-fork-count-1 \
-Dtest=CalciteOnlyNodeIntegrationTest -Dsurefire.failIfNoSpecifiedTests=false
{code}
Results on current master (f38a8b0412a), H2 excluded: *Tests run: 8, Failures:
3, Errors: 4, Skipped: 0*.
||Test||What it checks||Result on master||
|\{{testSqlFieldsQuery}}|\{{SqlFieldsQuery}} via cache API and thin JDBC:
\{{QUERY_ENGINE()}} is \{{calcite}}, \{{CREATE TABLE}}, \{{INSERT}},
\{{SELECT}}, \{{DROP TABLE}}|(/) pass|
|\{{testSqlQuery}}|legacy \{{SqlQuery}} via cache API|(x)
\{{NullPointerException: Cannot invoke
"GridQueryIndexing.generateFieldsQuery(...)" because "this.idx" is null}}|
|\{{testThinClientSqlQuery}}|legacy \{{SqlQuery}} via thin client|(x) same NPE,
surfaced to the client as server error 50000|
|\{{testTextQuery}}|\{{TextQuery}} fails with a message naming
\{{ignite-indexing}}|(x) \{{Indexing is disabled.}} - module not named|
|\{{testSetStreaming}}|\{{SET STREAMING ON}} over thin JDBC fails with a
message naming \{{ignite-indexing}}|(x) \{{Failed to parse query. Encountered "
"ON" "ON ""}} (Calcite parser), and then \{{Connection.close()}} hangs forever|
|\{{testCopy}}|\{{COPY FROM ... FORMAT CSV}} over thin JDBC fails with a
message naming \{{ignite-indexing}}|(x) \{{Failed to parse query. Non-query
expression encountered in illegal context}} - module not named|
|\{{testJdbcQueryEngineProperty}}|\{{jdbc:ignite:thin://...?queryEngine=calcite}}
without explicit \{{SqlConfiguration.queryEnginesConfiguration}}|(x)
\{{Handshake failed ... Not found configuration for query engine: calcite}}|
|\{{testQueryEngineHint}}|\{{SELECT /\*+ QUERY_ENGINE('calcite') \*/ ...}}
without explicit engines configuration|(x) \{{Query engines not configured, but
specified engine: calcite}}|
Without the profile (H2 on the classpath): *Tests run: 8, Skipped: 8*.
Checkstyle and RAT pass for \{{:ignite-calcite}}.
One extra finding from \{{testSetStreaming}}: a rejected \{{SET STREAMING ON}}
leaves the thin JDBC driver with \{{streamState}} set but never started,
because \{{JdbcThinConnection.executeNative}} assigns the state before
\{{sendRequest}}. \{{close()}} then goes through \{{StreamState.close0() ->
executeBatch(true) -> lastRespFut.get()}} and blocks forever. It is latent
today (with H2 the command never fails) but becomes reachable on a Calcite-only
node, so it needs a fix as part of this work. The test bounds the close with a
10s async wait so it fails instead of hitting the 5 minute JUnit timeout.
These tests are intentionally submitted without fixes: they document the
current gap and should turn green as the H2-only paths are either ported to
core or rejected with precise errors.
> Switch default SQL engine from H2 to Calcite
> --------------------------------------------
>
> Key: IGNITE-29117
> URL: https://issues.apache.org/jira/browse/IGNITE-29117
> Project: Ignite
> Issue Type: Improvement
> Reporter: Nikolay Izhikov
> Assignee: Nikolay Izhikov
> Priority: Major
> Fix For: 2.19
>
>
> Calcite based SQL engine are read to become default.
> H2 module must be moved to optional.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
