[
https://issues.apache.org/jira/browse/CAMEL-25882?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=18143658#comment-18143658
]
Aurélien Pupier commented on CAMEL-25882:
-----------------------------------------
h3. Analysis
I measured a full {{mvn verify -Pit -pl components/camel-cassandraql}} run (80
tests, one Cassandra container per test class). Of the ~280s of test-class
time, only ~189s is the tests themselves; the remaining ~90s is setup/teardown.
||Cause||Cost||Evidence||
|Cassandra driver Netty graceful shutdown quiet period on every
{{CqlSession.close()}} (default {{advanced.netty.io-group.shutdown.quiet-period
= 2}} and same for {{admin-group}})|~2.0s per test (~160s total over 80
tests)|Every trivial test took 2.0-2.1s; timing
{{BaseCassandra.doPostTearDown()}} gave {{session-close ms=2005..2013}}|
|One Cassandra container started per test class ({{static @RegisterExtension}}
in {{BaseCassandra}} using
{{CassandraServiceFactory.createLocalService("initScript.cql")}})|~4-7s per
class, 13 classes (~60-70s)|Container start to "Cassandra server running" is
~8s for the first class and ~4s for the following ones|
|Session creation + {{TRUNCATE}} + {{BasicDataSet.cql}} in
{{@BeforeEach}}|~100ms per test|Measured, negligible|
h3. Done in the first PR (https://github.com/apache/camel/pull/29704)
Test-only {{application.conf}} setting the Netty shutdown quiet period to 0.
Module ITs go from 4:51 to 1:52 min, all 80 tests pass
({{CassandraKeyValueRepositoryIT}} 66s -> 17s, {{CassandraComponentProducerIT}}
32s -> 5.7s).
h3. Potential further improvements (not done)
# *Share a single Cassandra container across all test classes* (potential
saving ~50-60s). {{CassandraServiceFactory.createSingletonService()}} already
exists but {{BaseCassandra}} uses {{createLocalService("initScript.cql")}}
instead. This requires applying the init script once and keeping the test
classes isolated from each other (as {{CassandraKeyValueRepositoryIT}} already
does with unique keys / dedicated tables), and checking the singleton container
lifecycle with Failsafe.
# Reuse one {{CqlSession}} per test class instead of one per test (~50ms per
test, minor).
# Tune the Cassandra container startup (e.g.
{{-Dcassandra.skip_wait_for_gossip_to_settle=0}}, smaller heap) to shorten the
3-7s start. Riskier, not tried.
_Note: {{CassandraKeyValueRepositoryIT.testDeleteWithExpectedValueMatching}}
failed once in the first (unmodified) run, which looks like the flakiness
already tracked in CAMEL-24290 / CAMEL-26486; not investigated here._
_IBM Bob 2.2.1 on behalf of Aurélien Pupier_
> Improve test time of Cassandra CQL
> ----------------------------------
>
> Key: CAMEL-25882
> URL: https://issues.apache.org/jira/browse/CAMEL-25882
> Project: Camel
> Issue Type: Test
> Components: camel-cassandraql
> Affects Versions: 4.24.0
> Reporter: Aurélien Pupier
> Assignee: Aurélien Pupier
> Priority: Minor
>
> it is taking almost [8' on
> jenkins|https://develocity.apache.org/s/qaf4axq6qn5ss/timeline?details=bw5egmw7tor2k&page=22]
> (but in // so not the most dramatic)
--
This message was sent by Atlassian Jira
(v10.3.26#10030026)
