[
https://issues.apache.org/jira/browse/SOLR-17478?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=18064195#comment-18064195
]
Chris M. Hostetter commented on SOLR-17478:
-------------------------------------------
At the time this was committed, it was logged once per JVM (because ZK would
only check it once unless/until reset).
Not clear from your comment if that's still true of if you're saying it now
happens once per test (when multiple tests run in the same JVM)
The choice made was to log anytime the value was overridden, in order to ensure
we gave *some* indication to people running tests that the state of their
system had been changed _in a way that would *not* automatically reset at the
end of the test run._
> ZkTestServer bugs can cause 30sec test pauses
> ---------------------------------------------
>
> Key: SOLR-17478
> URL: https://issues.apache.org/jira/browse/SOLR-17478
> Project: Solr
> Issue Type: Test
> Reporter: Chris M. Hostetter
> Assignee: Chris M. Hostetter
> Priority: Major
> Fix For: 9.8, 10.0
>
> Attachments: SOLR-17478.patch
>
>
> Pop Quiz: which of these two (psuedo-code) ZK based test classes will be
> faster...
> {code:java}
> public class Test_X extends SolrTestCase {
> public void test() throws Exception {
> ZkTestServer zkServer = new ZkTestServer(createTempDir());
> zkServer.run();
> zkServer.shutdown();
> }
> }
> public class Test_Y extends SolrTestCaseJ4 {
> public void test() throws Exception {
> ZkTestServer zkServer = new ZkTestServer(createTempDir());
> zkServer.run();
> zkServer.shutdown();
> }
> }
> {code}
> ...if you guessed "Test_X, because SolrTestCase has less overhead then
> SolrTestCaseJ4" then you are *wrong by ~30 seconds.*
> Actually, that's not _always_ true: if you run both tests in the same JVM
> then they will both take the same amount of time:
> * If Test_Y runs first, they will both take ~1 second each
> * If Test_X runs first, they will both take 30+ seconds *_each_*
> The reason for the dependency comes down to:
> * The {{"zookeeper.4lw.commands.whitelist"}} sysprop is set/cleared in
> SolrTestCaseJ4 (BeforeClass/AfterClass), but _NOT_ in SolrTestCase
> * ZkTestServer depends on the ZK helper methods
> {{ClientBase.waitForServerUp()}} which depends on the 4LW {{"stat"}} being
> whitelisted
> ** If "stat" is not whitelisted, then {{waitForServerUp()}} timesout after
> 30 sec and return a failure
> ** Bonus problem: ZkTestServer never checks of the return value of
> {{waitForServerUp()}} !
> * Zookeeper only checks for the 4LW whitelist sysprop the first time it's
> needed by the {{FourLetterCommands}} class _and then caches it staticly_
> ** So if the first testclass to run a ZK server doesn't extend
> SolrTestCaseJ4, then _every_ test that uses ZkTestServer in that JVM will
> stall for 30 seconds
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
