cannot copy files to hdfs: block under construction

[RafalSkolasinski]

Just tried to deploy hdfs using your helm charts and test it using rclone beta client.

Unfortunately I am unable to copy over files to hdfs. Client hags and in logs I see

2021-01-12 11:57:06,270 INFO org.apache.hadoop.hdfs.StateChange: BLOCK* allocate blk_1073741839_1015{UCState=UNDER_CONSTRUCTION, truncateBlock=null, primaryNodeIndex=-1, replicas=[ReplicaUC[[DISK]DS-1e01efed-b1b9-40bc-adc6-0abf2ae3ca00:NORMAL:10.244.0.25:50010|RBW], ReplicaUC[[DISK]DS-ab65c535-fabb-4e29-ad48-fb15ac3665a9:NORMAL:10.244.0.23:50010|RBW], ReplicaUC[[DISK]DS-52242411-b848-490b-9b28-9de6d5ad63ff:NORMAL:10.244.0.27:50010|RBW]]} for /test-folder/test.txt

hdfs support issue in rclone: rclone/rclone#42

[RafalSkolasinski]

The problem seems to be present only when coping files over to hdfs: folder creation and listing works fine.

To reproduce:

1. helm install hdfs gradiant/hdfs in the k8s cluster
2. Get beta rclone with hdfs support: https://beta.rclone.org/ (I used v1.54.0-beta.5058.35a4de203)
3. configure rclone with

[hdfs-kind]
type = hdfs
namenode = localhost:8020
username = hdfs

4. port-forward: kubectl port-forward svc/hdfs-namenode 8020:8020
5. make folder: /usr/bin/rclone mkdir hdfs-kind:/test-folder.
6. attempt to copy over file: /usr/bin/rclone copy test.txt hdfs-kind:/test-folder/

[RafalSkolasinski]

Happy to provide any more info if required.

[cgiraldo]

hdfs clients must be run inside kubernetes cluster in order to be able to resolve and access datanodes pods (port-forward is not enough though).

Other option to perform operations from outside the kubernetes cluster is to call httpfs rest service.

[RafalSkolasinski]

Thank you @cgiraldo, I will try from inside the cluster and get back to you 👍

[RafalSkolasinski]

Just tested and it works fine from inside the cluster.