HDDS-5015. localId is not consistent across SCMs when setup a multi node SCM HA cluster. (#2072)
diff --git a/hadoop-hdds/server-scm/src/main/java/org/apache/hadoop/hdds/scm/ha/SequenceIdGenerator.java b/hadoop-hdds/server-scm/src/main/java/org/apache/hadoop/hdds/scm/ha/SequenceIdGenerator.java
index 147cc88..fae1e78 100644
--- a/hadoop-hdds/server-scm/src/main/java/org/apache/hadoop/hdds/scm/ha/SequenceIdGenerator.java
+++ b/hadoop-hdds/server-scm/src/main/java/org/apache/hadoop/hdds/scm/ha/SequenceIdGenerator.java
@@ -33,9 +33,11 @@
import java.io.IOException;
import java.lang.reflect.Proxy;
+import java.time.LocalDate;
import java.util.HashMap;
import java.util.Map;
import java.util.concurrent.ConcurrentHashMap;
+import java.util.concurrent.TimeUnit;
import java.util.concurrent.locks.Lock;
import java.util.concurrent.locks.ReentrantLock;
@@ -334,8 +336,19 @@
Table<String, Long> sequenceIdTable = scmMetadataStore.getSequenceIdTable();
// upgrade localId
+ // Short-term solution: when setup multi SCM from scratch, they need
+ // achieve an agreement upon the initial value of LOCAL_ID.
+ // Long-term solution: the bootstrapped SCM will explicitly download
+ // scm.db from leader SCM, and drop its own scm.db. Thus the upgrade
+ // operations can take effect exactly once in a SCM HA cluster.
if (sequenceIdTable.get(LOCAL_ID) == null) {
- sequenceIdTable.put(LOCAL_ID, UniqueId.next());
+ long millisSinceEpoch = TimeUnit.DAYS.toMillis(
+ LocalDate.of(LocalDate.now().getYear() + 1, 1, 1).toEpochDay());
+
+ long localId = millisSinceEpoch << Short.SIZE;
+ Preconditions.checkArgument(localId > UniqueId.next());
+
+ sequenceIdTable.put(LOCAL_ID, localId);
LOG.info("upgrade {} to {}", LOCAL_ID, sequenceIdTable.get(LOCAL_ID));
}
