[
https://issues.apache.org/jira/browse/IGNITE-28229?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Vyacheslav Koptilin updated IGNITE-28229:
-----------------------------------------
Description:
Let's take a look at `org.apache.ignite.internal.Cluster`:
{code:java}
/**
* Starts a cluster node.
*
* @param testInfo Test info.
* @param nodeIndex Index of the node to start.
* @param nodeBootstrapConfigTemplate Bootstrap config template to use for this
node.
* @param nodeBootstrapConfigUpdater Boot configuration updater before starting
the node.
* @return Started server and its registration future.
*/
public ServerRegistration startEmbeddedNode(
@Nullable TestInfo testInfo,
int nodeIndex,
String nodeBootstrapConfigTemplate,
NodeBootstrapConfigUpdater nodeBootstrapConfigUpdater
) {
...
IgniteServer node = clusterConfiguration.usePreConfiguredStorageProfiles()
? TestIgnitionManager.start(
nodeName,
config,
clusterConfiguration.workDir().resolve(clusterConfiguration.clusterName()).resolve(nodeName))
: TestIgnitionManager.startWithoutPreConfiguredStorageProfiles(
nodeName,
config,
clusterConfiguration.workDir().resolve(clusterConfiguration.clusterName()).resolve(nodeName));
synchronized (igniteServers) {
setListAtIndex(igniteServers, nodeIndex, node);
}
CompletableFuture<Void> registrationFuture =
node.waitForInitAsync().thenRun(() -> {
synchronized (nodes) {
setListAtIndex(nodes, nodeIndex, node.api());
}
if (stopped) {
// Make sure we stop even a node that finished starting after the
cluster has been stopped.
node.shutdown();
}
});
return new ServerRegistration(node, registrationFuture);
}{code}
Let's assume that `TestIgnitionManager.start` hangs for some reason, and the
current thread (which executes `startEmbeddedNode) is interrupted. In this
case, the starting node is not registered as a part of `igniteServers` and
`nodes`, however, it does not interrupt the process of starting itself.
Therefore, we can end up with the stale node, which in turn may affect other
tests.
was:
Let's take a look at `org.apache.ignite.internal.Cluster`:
{code:java}
/**
* Starts a cluster node.
*
* @param testInfo Test info.
* @param nodeIndex Index of the node to start.
* @param nodeBootstrapConfigTemplate Bootstrap config template to use for this
node.
* @param nodeBootstrapConfigUpdater Boot configuration updater before starting
the node.
* @return Started server and its registration future.
*/
public ServerRegistration startEmbeddedNode(
@Nullable TestInfo testInfo,
int nodeIndex,
String nodeBootstrapConfigTemplate,
NodeBootstrapConfigUpdater nodeBootstrapConfigUpdater
) {
...
IgniteServer node = clusterConfiguration.usePreConfiguredStorageProfiles()
? TestIgnitionManager.start(
nodeName,
config,
clusterConfiguration.workDir().resolve(clusterConfiguration.clusterName()).resolve(nodeName))
: TestIgnitionManager.startWithoutPreConfiguredStorageProfiles(
nodeName,
config,
clusterConfiguration.workDir().resolve(clusterConfiguration.clusterName()).resolve(nodeName));
synchronized (igniteServers) {
setListAtIndex(igniteServers, nodeIndex, node);
}
CompletableFuture<Void> registrationFuture =
node.waitForInitAsync().thenRun(() -> {
synchronized (nodes) {
setListAtIndex(nodes, nodeIndex, node.api());
}
if (stopped) {
// Make sure we stop even a node that finished starting after the
cluster has been stopped.
node.shutdown();
}
});
return new ServerRegistration(node, registrationFuture);
}{code}
Let's assume that `TestIgnitionManager.start` hangs for some reason, and the
current thread (which executes `startEmbeddedNode) is interrupted. In this
case, the starting node is not registered as a part of `igniteServers` and
`nodes`, however, it does not prevent the process of starting itself.
Therefore, we can end up with the stale node, that in turn may affect other
tests.
> TestCluster might skip registration an Ignite node
> --------------------------------------------------
>
> Key: IGNITE-28229
> URL: https://issues.apache.org/jira/browse/IGNITE-28229
> Project: Ignite
> Issue Type: Bug
> Reporter: Vyacheslav Koptilin
> Priority: Major
> Labels: MakeTeamcityGreenAgain, ignite-3
>
> Let's take a look at `org.apache.ignite.internal.Cluster`:
> {code:java}
> /**
> * Starts a cluster node.
> *
> * @param testInfo Test info.
> * @param nodeIndex Index of the node to start.
> * @param nodeBootstrapConfigTemplate Bootstrap config template to use for
> this node.
> * @param nodeBootstrapConfigUpdater Boot configuration updater before
> starting the node.
> * @return Started server and its registration future.
> */
> public ServerRegistration startEmbeddedNode(
> @Nullable TestInfo testInfo,
> int nodeIndex,
> String nodeBootstrapConfigTemplate,
> NodeBootstrapConfigUpdater nodeBootstrapConfigUpdater
> ) {
> ...
> IgniteServer node = clusterConfiguration.usePreConfiguredStorageProfiles()
> ? TestIgnitionManager.start(
> nodeName,
> config,
>
> clusterConfiguration.workDir().resolve(clusterConfiguration.clusterName()).resolve(nodeName))
> : TestIgnitionManager.startWithoutPreConfiguredStorageProfiles(
> nodeName,
> config,
>
> clusterConfiguration.workDir().resolve(clusterConfiguration.clusterName()).resolve(nodeName));
> synchronized (igniteServers) {
> setListAtIndex(igniteServers, nodeIndex, node);
> }
> CompletableFuture<Void> registrationFuture =
> node.waitForInitAsync().thenRun(() -> {
> synchronized (nodes) {
> setListAtIndex(nodes, nodeIndex, node.api());
> }
> if (stopped) {
> // Make sure we stop even a node that finished starting after the
> cluster has been stopped.
> node.shutdown();
> }
> });
> return new ServerRegistration(node, registrationFuture);
> }{code}
> Let's assume that `TestIgnitionManager.start` hangs for some reason, and the
> current thread (which executes `startEmbeddedNode) is interrupted. In this
> case, the starting node is not registered as a part of `igniteServers` and
> `nodes`, however, it does not interrupt the process of starting itself.
> Therefore, we can end up with the stale node, which in turn may affect other
> tests.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
