This is an automated email from the ASF dual-hosted git repository.
noble pushed a commit to branch main
in repository https://gitbox.apache.org/repos/asf/solr.git
The following commit(s) were added to refs/heads/main by this push:
new bc7fceeb7d7 SOLR-17271: PerReplicaState: Shard leader elections still
impact state.json (#2443)
bc7fceeb7d7 is described below
commit bc7fceeb7d7dc6e75b9e79ddf8cddb57eb8f452f
Author: Noble Paul <[email protected]>
AuthorDate: Sat Jun 1 09:52:03 2024 +1000
SOLR-17271: PerReplicaState: Shard leader elections still impact state.json
(#2443)
---
.../apache/solr/cloud/overseer/SliceMutator.java | 4 ++
.../cloud/PerReplicaStatesIntegrationTest.java | 48 ++++++++++++++++++++--
2 files changed, 48 insertions(+), 4 deletions(-)
diff --git
a/solr/core/src/java/org/apache/solr/cloud/overseer/SliceMutator.java
b/solr/core/src/java/org/apache/solr/cloud/overseer/SliceMutator.java
index c016fa5489b..d5170c66474 100644
--- a/solr/core/src/java/org/apache/solr/cloud/overseer/SliceMutator.java
+++ b/solr/core/src/java/org/apache/solr/cloud/overseer/SliceMutator.java
@@ -159,6 +159,10 @@ public class SliceMutator {
log.error("Could not mark shard leader for non existing collection: {}",
collectionName);
return ZkStateWriter.NO_OP;
}
+ if (coll.isPerReplicaState()) {
+ log.debug("Do not mark shard leader for PRS collection: {}",
collectionName);
+ return ZkStateWriter.NO_OP;
+ }
Map<String, Slice> slices = coll.getSlicesMap();
Slice slice = slices.get(sliceName);
diff --git
a/solr/solrj/src/test/org/apache/solr/common/cloud/PerReplicaStatesIntegrationTest.java
b/solr/solrj/src/test/org/apache/solr/common/cloud/PerReplicaStatesIntegrationTest.java
index e75100f1a50..3512e74c6f3 100644
---
a/solr/solrj/src/test/org/apache/solr/common/cloud/PerReplicaStatesIntegrationTest.java
+++
b/solr/solrj/src/test/org/apache/solr/common/cloud/PerReplicaStatesIntegrationTest.java
@@ -40,6 +40,7 @@ import org.slf4j.LoggerFactory;
/** This test would be faster if we simulated the zk state instead. */
@LogLevel(
"org.apache.solr.common.cloud.ZkStateReader=DEBUG;"
+ + "org.apache.solr.cloud.overseer.ZkStateWriter=DEBUG;"
+ "org.apache.solr.handler.admin.CollectionsHandler=DEBUG;"
+ "org.apache.solr.common.cloud.PerReplicaStatesOps=DEBUG;"
+ "org.apache.solr.cloud.Overseer=INFO;"
@@ -315,7 +316,8 @@ public class PerReplicaStatesIntegrationTest extends
SolrCloudTestCase {
CollectionAdminRequest.createCollection(PRS_COLL, "conf", 10, 1)
.setPerReplicaState(Boolean.TRUE)
.process(cluster.getSolrClient());
- stat =
cluster.getZkClient().exists(DocCollection.getCollectionPath(PRS_COLL), null,
true);
+ String PRS_PATH = DocCollection.getCollectionPath(PRS_COLL);
+ stat = cluster.getZkClient().exists(PRS_PATH, null, true);
// +1 after all replica are added with on state.json write to
CreateCollectionCmd.setData()
assertEquals(1, stat.getVersion());
// For each replica:
@@ -330,7 +332,7 @@ public class PerReplicaStatesIntegrationTest extends
SolrCloudTestCase {
CollectionAdminRequest.addReplicaToShard(PRS_COLL, "shard1")
.process(cluster.getSolrClient());
cluster.waitForActiveCollection(PRS_COLL, 10, 11);
- stat =
cluster.getZkClient().exists(DocCollection.getCollectionPath(PRS_COLL), null,
true);
+ stat = cluster.getZkClient().exists(PRS_PATH, null, true);
// For the new replica:
// +2 for state.json overseer writes, even though there's no longer PRS
updates from
// overseer, current code would still do a "TOUCH" on the PRS entry
@@ -350,7 +352,7 @@ public class PerReplicaStatesIntegrationTest extends
SolrCloudTestCase {
CollectionAdminRequest.deleteReplica(PRS_COLL, "shard1",
addedReplica.getName())
.process(cluster.getSolrClient());
cluster.waitForActiveCollection(PRS_COLL, 10, 10);
- stat =
cluster.getZkClient().exists(DocCollection.getCollectionPath(PRS_COLL), null,
true);
+ stat = cluster.getZkClient().exists(PRS_PATH, null, true);
// For replica deletion
// +1 for ZkController#unregister, which delete the PRS entry from data
node
// overseer, current code would still do a "TOUCH" on the PRS entry
@@ -359,11 +361,49 @@ public class PerReplicaStatesIntegrationTest extends
SolrCloudTestCase {
for (JettySolrRunner j : cluster.getJettySolrRunners()) {
j.stop();
j.start(true);
- stat =
cluster.getZkClient().exists(DocCollection.getCollectionPath(PRS_COLL), null,
true);
+ stat = cluster.getZkClient().exists(PRS_PATH, null, true);
// ensure restart does not update the state.json, after
addReplica/deleteReplica, 2 more
// updates hence at version 3 on state.json version
assertEquals(3, stat.getVersion());
}
+
+ // test for leader election
+ Replica leader =
+
cluster.getZkStateReader().clusterState.getCollection(PRS_COLL).getLeader("shard2");
+
+ JettySolrRunner j2 = cluster.startJettySolrRunner();
+ response =
+ CollectionAdminRequest.addReplicaToShard(PRS_COLL, "shard2")
+ .setNode(j2.getNodeName())
+ .process(cluster.getSolrClient());
+
+ // wait for the new replica to be active
+ cluster.waitForActiveCollection(PRS_COLL, 10, 11);
+ stat = cluster.getZkClient().exists(PRS_PATH, null, true);
+ // +1 for a new replica
+ assertEquals(4, stat.getVersion());
+ DocCollection c = cluster.getZkStateReader().getCollection(PRS_COLL);
+ Replica newreplica = c.getReplica((s, replica) ->
replica.node.equals(j2.getNodeName()));
+
+ // let's stop the old leader
+ JettySolrRunner oldJetty = cluster.getReplicaJetty(leader);
+ oldJetty.stop();
+
+ cluster
+ .getZkStateReader()
+ .waitForState(
+ PRS_COLL,
+ 10,
+ TimeUnit.SECONDS,
+ (liveNodes, collectionState) ->
+ PerReplicaStatesOps.fetch(PRS_PATH, cluster.getZkClient(),
null)
+ .states
+ .get(newreplica.name)
+ .isLeader);
+ PerReplicaStates prs = PerReplicaStatesOps.fetch(PRS_PATH,
cluster.getZkClient(), null);
+ stat = cluster.getZkClient().exists(PRS_PATH, null, true);
+ // the version should not have updated
+ assertEquals(4, stat.getVersion());
} finally {
cluster.shutdown();
}
