database/doris
2
0
Fork 0
Code Issues Pull Requests Actions 3 Packages Projects Releases Wiki Activity

[Bug][Colocation Join] Fix colocation balance endless loop bug (#4471)

Browse Source 
1. Only one available backend.
2. All backends are checked but this round is not changed. For example, all backends are on the same host.
This commit is contained in:
wyb
2020-08-28 09:27:57 +08:00
committed by GitHub
parent 174c9f89ea
commit ec64789e89
2 changed files with 48 additions and 3 deletions
Download Patch File Download Diff File Expand all files Collapse all files

18
fe/fe-core/src/main/java/org/apache/doris/clone/ColocateTableBalancer.java
View File

@ -546,6 +546,11 @@ public class ColocateTableBalancer extends MasterDaemon {
// sort backends with replica num
List<Map.Entry<Long, Long>> backendWithReplicaNum =
getSortedBackendReplicaNumPairs(allAvailBackendIds, flatBackendsPerBucketSeq);
// if there is only one available backend, end the outer loop
if (backendWithReplicaNum.size() == 1) {
LOG.info("there is only one available backend, end the outer loop in colocate group {}", groupId);
break;
}
int i = 0;
int j = backendWithReplicaNum.size() - 1;
@ -600,11 +605,18 @@ public class ColocateTableBalancer extends MasterDaemon {
}
if (!isThisRoundChanged) {
// select another load backend and try again
LOG.info("unable to replace backend {} with backend {} in colocate group {}",
srcBeId, destBeId, groupId);
j--;
continue;
if (--j == i) {
// if all backends are checked but this round is not changed,
// we should end the outer loop to avoid endless loops
LOG.info("all backends are checked but this round is not changed, " +
"end outer loop in colocate group {}", groupId);
break OUT;
} else {
// select another load backend and try again
continue;
}
}
break;

33
fe/fe-core/src/test/java/org/apache/doris/clone/ColocateTableBalancerTest.java
View File

@ -151,4 +151,37 @@ public class ColocateTableBalancerTest {
Assert.assertFalse(changed);
Assert.assertTrue(balancedBackendsPerBucketSeq.isEmpty());
}
@Test
public void testFixBalanceEndlessLoop() {
GroupId groupId = new GroupId(10000, 10001);
List<Column> distributionCols = Lists.newArrayList();
distributionCols.add(new Column("k1", PrimitiveType.INT));
ColocateGroupSchema groupSchema = new ColocateGroupSchema(groupId, distributionCols, 5, (short) 1);
Map<GroupId, ColocateGroupSchema> group2Schema = Maps.newHashMap();
group2Schema.put(groupId, groupSchema);
// 1. only one available backend
// [[7], [7], [7], [7], [7]]
ColocateTableIndex colocateTableIndex = createColocateIndex(groupId, Lists.newArrayList(7L, 7L, 7L, 7L, 7L));
Deencapsulation.setField(colocateTableIndex, "group2Schema", group2Schema);
List<List<Long>> balancedBackendsPerBucketSeq = Lists.newArrayList();
List<Long> allAvailBackendIds = Lists.newArrayList(7L);
boolean changed = Deencapsulation.invoke(balancer, "balance", groupId, allAvailBackendIds,
colocateTableIndex, infoService, balancedBackendsPerBucketSeq);
Assert.assertFalse(changed);
// 2. all backends are checked but this round is not changed
// [[7], [7], [7], [7], [7]]
// and add new backends 8, 9 that are on the same host with 7
colocateTableIndex = createColocateIndex(groupId, Lists.newArrayList(7L, 7L, 7L, 7L, 7L));
Deencapsulation.setField(colocateTableIndex, "group2Schema", group2Schema);
balancedBackendsPerBucketSeq = Lists.newArrayList();
allAvailBackendIds = Lists.newArrayList(7L, 8L, 9L);
changed = Deencapsulation.invoke(balancer, "balance", groupId, allAvailBackendIds,
colocateTableIndex, infoService, balancedBackendsPerBucketSeq);
Assert.assertFalse(changed);
}
}