Change default value for dfs.ha.nn.not-become-active-in-safemode to false

CHANGELOG.md

@@ -11,6 +11,7 @@ All notable changes to this project will be documented in this file.
 ### Changed
 
 - `operator-rs` `0.56.1` -> `0.57.0` ([#433]).
+- Change default value of `dfs.ha.nn.not-become-active-in-safemode` from `true` to `false` ([#458]).
 
 ### Fixed
 
@@ -19,6 +20,7 @@ All notable changes to this project will be documented in this file.
 
 [#433]: https://github.com/stackabletech/hdfs-operator/pull/433
 [#451]: https://github.com/stackabletech/hdfs-operator/pull/451
+[#458]: https://github.com/stackabletech/hdfs-operator/pull/458
 
 ## [23.11.0] - 2023-11-24
 

rust/operator-binary/src/hdfs_controller.rs

@@ -494,6 +494,15 @@ fn rolegroup_config_map(
                 // IMPORTANT: these folders must be under the volume mount point, otherwise they will not
                 // be formatted by the namenode, or used by the other services.
                 // See also: https://github.com/apache-spark-on-k8s/kubernetes-HDFS/commit/aef9586ecc8551ca0f0a468c3b917d8c38f494a0
+                //
+                // Notes on configuration choices
+                // ===============================
+                // We used to set `dfs.ha.nn.not-become-active-in-safemode` to true here due to
+                // badly worded HDFS documentation:
+                // https://hadoop.apache.org/docs/stable/hadoop-project-dist/hadoop-hdfs/HDFSHighAvailabilityWithNFS.html
+                // This caused a deadlock with no namenode becoming active during a startup after
+                // HDFS was completely down for a while.
+
                 hdfs_site_xml = HdfsSiteConfigBuilder::new(hdfs_name.to_string())
                     .dfs_namenode_name_dir()
                     .dfs_datanode_data_dir(merged_config.data_node_resources().map(|r| r.storage))
@@ -508,7 +517,6 @@ fn rolegroup_config_map(
                     .dfs_client_failover_proxy_provider()
                     .security_config(hdfs)
                     .add("dfs.ha.fencing.methods", "shell(/bin/true)")
-                    .add("dfs.ha.nn.not-become-active-in-safemode", "true")
                     .add("dfs.ha.automatic-failover.enabled", "true")
                     .add("dfs.ha.namenode.id", "${env.POD_NAME}")
                     // the extend with config must come last in order to have overrides working!!!