Github user StefanRRichter commented on a diff in the pull request:
https://github.com/apache/flink/pull/2249#discussion_r71309594
--- Diff: docs/setup/config.md ---
@@ -272,7 +272,9 @@ For example when running Flink on YARN on an
environment with a restrictive fire
- `recovery.zookeeper.quorum`: Defines the ZooKeeper quorum URL which is
used to connet to the ZooKeeper cluster when the 'zookeeper' recovery mode is
selected
-- `recovery.zookeeper.path.root`: (Default '/flink') Defines the root dir
under which the ZooKeeper recovery mode will create znodes.
+- `recovery.zookeeper.path.root`: (Default '/flink') Defines the root dir
under which the ZooKeeper recovery mode will create namespace directories.
+
+- `recovery.zookeeper.path.namespace`: (Default '/default_ns' in
standalone mode, or the <yarn-application-id> under Yarn) Defines the
subdirectory under the root dir where the ZooKeeper recovery mode will create
znodes. This allows to isolate multiple applications on the same ZooKeeper.
--- End diff --
I think the main problem with default namespaces in standalone mode is that
there is no authority generating reliably unique identifier that we can use to
label clusters (in contrast to e.g. application ids in Yarn). Also the contents
of the masters file could be the same for two different clusters running on the
same machines and collisions could happen. In particular, such collisions could
be very rare and hence even more surprising to the user. Furthermore, I think
that having multiple clusters in this way is exactly when you want to start
Yarn or Mesos.
---
If your project is set up for it, you can reply to this email and have your
reply appear on GitHub as well. If your project does not have this feature
enabled and wishes so, or if the feature is enabled but not working, please
contact infrastructure at infrastruct...@apache.org or file a JIRA ticket
with INFRA.
---
