ZookeeperClient should retry instantly on zk session expiry (#2450)
Descriptions of the changes in this PR:
### Motivation
Currently, we wait for `sessionTimeoutMs` before initiating a new zk cnxn. Session expiry is recevied only
when zk quprum is in a healthy state and can acquire a quorum to expire a session.
All other connection loss retries happen at zk client library transparently.
There is not much benefit waiting at BK client level.
### Changes
Updated the values for backoff and maxbackoff to 0 seconds.
* ZookeeperClient should retry instantly on zk session expiry
Currently, we wait for `sessionTimeoutMs` before initiating a new zk cnxn. Session expiry is recevied only
when zk quprum is in a healthy state and can acquire a quorum to expire a session.
All other connection loss retries happen at zk client library transparently.
* Added comments
diff --git a/bookkeeper-server/src/main/java/org/apache/bookkeeper/zookeeper/ZooKeeperClient.java b/bookkeeper-server/src/main/java/org/apache/bookkeeper/zookeeper/ZooKeeperClient.java
index 3d97e6f..cb401dc 100644
--- a/bookkeeper-server/src/main/java/org/apache/bookkeeper/zookeeper/ZooKeeperClient.java
+++ b/bookkeeper-server/src/main/java/org/apache/bookkeeper/zookeeper/ZooKeeperClient.java
@@ -228,8 +228,11 @@
checkArgument(retryExecThreadCount > 0);
if (null == connectRetryPolicy) {
+ // Session expiry event is received by client only when zk quorum is well established.
+ // All other connection loss retries happen at zk client library transparently.
+ // Hence, we don't need to wait before retrying.
connectRetryPolicy =
- new BoundExponentialBackoffRetryPolicy(sessionTimeoutMs, sessionTimeoutMs, Integer.MAX_VALUE);
+ new BoundExponentialBackoffRetryPolicy(0, 0, Integer.MAX_VALUE);
}
if (null == operationRetryPolicy) {
operationRetryPolicy =
