[
https://issues.apache.org/jira/browse/IGNITE-23481?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Alexander Lapin updated IGNITE-23481:
-------------------------------------
Description:
Generally there are two issues.
First of all, the test is useless.
* According to the javadoc table recreation is required "Test scenario when we
have lagged node, and tables with the same name are deleted and created again."
However it's not implemented in code.
*
>From the code, test verifies that it's possible to retrieve data from
>previously retrieved table on node1 while inhibiting node1 watch processing.
>It's senseless because watch processing neither effects previously retrieved
>table retrieval not data retrieval.
* Schema awaiting logic was adjusted since the moment when test was written.
All in all, the test tests nothing and thus should be removed.
Besides that.
Data retrieval within MG inhibited context may fail with Replication is timed
out because of hanging in schema awaiting logic, precisely in
{code:java}
public CompletableFuture<Void> waitForMetadataCompleteness(HybridTimestamp ts) {
return
clusterTime.waitFor(ts.subtractPhysicalTime(delayDurationMs.getAsLong()));
}{code}
It's possible to increase flaky rate by decreasing DD or decrease it by
increasing DD. Locally on my machine with 20 as delayDuraionMs the test will
always fail. The reason for metadata incompleteness is in MG safe time
propagation halting because of WatchListenerInhibitor.
was:
Generally there are two issues.
Data retrieval within MG inhibited context may fail with
> ItTablesApiTest#testGetTableFromLaggedNode may fail with Replication is timed
> out
> ---------------------------------------------------------------------------------
>
> Key: IGNITE-23481
> URL: https://issues.apache.org/jira/browse/IGNITE-23481
> Project: Ignite
> Issue Type: Bug
> Reporter: Alexander Lapin
> Priority: Major
> Labels: ignite-3
>
> Generally there are two issues.
> First of all, the test is useless.
> * According to the javadoc table recreation is required "Test scenario when
> we have lagged node, and tables with the same name are deleted and created
> again." However it's not implemented in code.
> *
> From the code, test verifies that it's possible to retrieve data from
> previously retrieved table on node1 while inhibiting node1 watch processing.
> It's senseless because watch processing neither effects previously retrieved
> table retrieval not data retrieval.
> * Schema awaiting logic was adjusted since the moment when test was written.
> All in all, the test tests nothing and thus should be removed.
>
> Besides that.
> Data retrieval within MG inhibited context may fail with Replication is timed
> out because of hanging in schema awaiting logic, precisely in
> {code:java}
> public CompletableFuture<Void> waitForMetadataCompleteness(HybridTimestamp
> ts) {
> return
> clusterTime.waitFor(ts.subtractPhysicalTime(delayDurationMs.getAsLong()));
> }{code}
> It's possible to increase flaky rate by decreasing DD or decrease it by
> increasing DD. Locally on my machine with 20 as delayDuraionMs the test will
> always fail. The reason for metadata incompleteness is in MG safe time
> propagation halting because of WatchListenerInhibitor.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
