KeeProMise commented on code in PR #5990:
URL: https://github.com/apache/hadoop/pull/5990#discussion_r1306285491
##########
hadoop-hdfs-project/hadoop-hdfs-rbf/src/main/java/org/apache/hadoop/hdfs/server/federation/resolver/MembershipNamenodeResolver.java:
##########
@@ -478,4 +476,27 @@ private List<MembershipState>
getRecentRegistrationForQuery(
public void setRouterId(String router) {
this.routerId = router;
}
+
+ /**
+ * Shuffle cache, to ensure that the current nn will not be accessed first
next time.
+ *
+ *
+ * @param nsId name service id
+ * @param namenode namenode contexts
+ */
+ @Override
+ public synchronized void shuffleCache(String nsId, FederationNamenodeContext
namenode) {
+ cacheNS.compute(Pair.of(nsId, false), (ns, namenodeContexts) -> {
+ if (namenodeContexts != null
+ && namenodeContexts.size() > 0
Review Comment:
The reason for not judging outside here is that we should ensure that get
cache and modify cache are atomic, considering the following situation:
1. After **Thread1** judges that the cache is not empty, it gets the ns no
active in the cache
2. **Thread2** (StateStoreCacheUpdateService loadCache )clear the cache
3. **Thread3** processes the client request and finds that the cache is
empty（MembershipNamenodeResolver.getNamenodesForNameserviceId()）, and updates
the cache. At this time, the ns - in the cache has active nn
4. **Thread1** rotates the previously acquired data of non-active nn and
writes it into the cache, causing ns in the cache to have no active nn
so, guaranteed atomicity can ensure that the above 2 and 3 are recognized
before modifying the cache
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
