Support Questions

robert_jonczy · ‎12-20-2017

Hi,

I have a problem with rebalancing HDFS after adding new DataNode to cluter. In my configuration I had 4 DataNodes and added new one (5th).

Below is report from dfsadmin

[hdfs@snr-prod-master0 ~]$ hdfs dfsadmin -report
Configured Capacity: 21563228579840 (19.61 TB)
Present Capacity: 20460562895805 (18.61 TB)
DFS Remaining: 20290148094909 (18.45 TB)
DFS Used: 170414800896 (158.71 GB)
DFS Used%: 0.83%
Under replicated blocks: 0
Blocks with corrupt replicas: 0
Missing blocks: 0
Missing blocks (with replication factor 1): 0


-------------------------------------------------
Live datanodes (5):


Name: 172.17.2.61:50010 (snr-prod-slave1)
Hostname: snr-prod-slave1
Decommission Status : Normal
Configured Capacity: 4312645715968 (3.92 TB)
DFS Used: 35358969856 (32.93 GB)
Non DFS Used: 0 (0 B)
DFS Remaining: 4056646234773 (3.69 TB)
DFS Used%: 0.82%
DFS Remaining%: 94.06%
Configured Cache Capacity: 0 (0 B)
Cache Used: 0 (0 B)
Cache Remaining: 0 (0 B)
Cache Used%: 100.00%
Cache Remaining%: 0.00%
Xceivers: 12
Last contact: Wed Dec 20 20:52:23 UTC 2017




Name: 172.17.2.64:50010 (snr-prod-slave4)
Hostname: snr-prod-slave4
Decommission Status : Normal
Configured Capacity: 4312645715968 (3.92 TB)
DFS Used: 47864344576 (44.58 GB)
Non DFS Used: 0 (0 B)
DFS Remaining: 4044275077691 (3.68 TB)
DFS Used%: 1.11%
DFS Remaining%: 93.78%
Configured Cache Capacity: 0 (0 B)
Cache Used: 0 (0 B)
Cache Remaining: 0 (0 B)
Cache Used%: 100.00%
Cache Remaining%: 0.00%
Xceivers: 10
Last contact: Wed Dec 20 20:52:26 UTC 2017




Name: 172.17.2.62:50010 (snr-prod-slave2)
Hostname: snr-prod-slave2
Decommission Status : Normal
Configured Capacity: 4312645715968 (3.92 TB)
DFS Used: 221184 (216 KB)
Non DFS Used: 0 (0 B)
DFS Remaining: 4092407638196 (3.72 TB)
DFS Used%: 0.00%
DFS Remaining%: 94.89%
Configured Cache Capacity: 0 (0 B)
Cache Used: 0 (0 B)
Cache Remaining: 0 (0 B)
Cache Used%: 100.00%
Cache Remaining%: 0.00%
Xceivers: 6
Last contact: Wed Dec 20 20:52:26 UTC 2017




Name: 172.17.2.65:50010 (snr-prod-slave5)
Hostname: snr-prod-slave5
Decommission Status : Normal
Configured Capacity: 4312645715968 (3.92 TB)
DFS Used: 44406976512 (41.36 GB)
Non DFS Used: 0 (0 B)
DFS Remaining: 4047866664447 (3.68 TB)
DFS Used%: 1.03%
DFS Remaining%: 93.86%
Configured Cache Capacity: 0 (0 B)
Cache Used: 0 (0 B)
Cache Remaining: 0 (0 B)
Cache Used%: 100.00%
Cache Remaining%: 0.00%
Xceivers: 8
Last contact: Wed Dec 20 20:52:23 UTC 2017




Name: 172.17.2.60:50010 (snr-prod-slave0)
Hostname: snr-prod-slave0
Decommission Status : Normal
Configured Capacity: 4312645715968 (3.92 TB)
DFS Used: 42784288768 (39.85 GB)
Non DFS Used: 0 (0 B)
DFS Remaining: 4048952479802 (3.68 TB)
DFS Used%: 0.99%
DFS Remaining%: 93.89%
Configured Cache Capacity: 0 (0 B)
Cache Used: 0 (0 B)
Cache Remaining: 0 (0 B)
Cache Used%: 100.00%
Cache Remaining%: 0.00%
Xceivers: 16
Last contact: Wed Dec 20 20:52:23 UTC 2017

And after adding new node to cluster i have run rebalance operation, to distribute data equally, but it says it is balanced (The cluster is balanced. Exiting...)

[hdfs@snr-prod-master0 ~]$ hdfs balancer -threshold 5
17/12/20 20:57:36 INFO balancer.Balancer: Using a threshold of 5.0
17/12/20 20:57:36 INFO balancer.Balancer: namenodes  = [hdfs://snr-prod-master0:8020]
17/12/20 20:57:36 INFO balancer.Balancer: parameters = Balancer.BalancerParameters [BalancingPolicy.Node, threshold = 5.0, max idle iteration = 5, #excluded nodes = 0, #included nodes = 0, #source nodes = 0, #blockpools = 0, run during upgrade = false]
17/12/20 20:57:36 INFO balancer.Balancer: included nodes = []
17/12/20 20:57:36 INFO balancer.Balancer: excluded nodes = []
17/12/20 20:57:36 INFO balancer.Balancer: source nodes = []
Time Stamp               Iteration#  Bytes Already Moved  Bytes Left To Move  Bytes Being Moved
17/12/20 20:57:37 INFO balancer.KeyManager: Block token params received from NN: update interval=10hrs, 0sec, token lifetime=10hrs, 0sec
17/12/20 20:57:38 INFO block.BlockTokenSecretManager: Setting block keys
17/12/20 20:57:38 INFO balancer.KeyManager: Update block keys every 2hrs, 30mins, 0sec
17/12/20 20:57:38 INFO balancer.Balancer: dfs.balancer.movedWinWidth = 5400000 (default=5400000)
17/12/20 20:57:38 INFO balancer.Balancer: dfs.balancer.moverThreads = 1000 (default=1000)
17/12/20 20:57:38 INFO balancer.Balancer: dfs.balancer.dispatcherThreads = 200 (default=200)
17/12/20 20:57:38 INFO balancer.Balancer: dfs.datanode.balance.max.concurrent.moves = 5 (default=5)
17/12/20 20:57:38 INFO balancer.Balancer: dfs.balancer.getBlocks.size = 2147483648 (default=2147483648)
17/12/20 20:57:38 INFO balancer.Balancer: dfs.balancer.getBlocks.min-block-size = 10485760 (default=10485760)
17/12/20 20:57:38 INFO block.BlockTokenSecretManager: Setting block keys
17/12/20 20:57:38 INFO balancer.Balancer: dfs.balancer.max-size-to-move = 10737418240 (default=10737418240)
17/12/20 20:57:38 INFO balancer.Balancer: dfs.blocksize = 134217728 (default=134217728)
17/12/20 20:57:38 INFO net.NetworkTopology: Adding a new node: /default-rack/172.17.2.61:50010
17/12/20 20:57:38 INFO net.NetworkTopology: Adding a new node: /default-rack/172.17.2.60:50010
17/12/20 20:57:38 INFO net.NetworkTopology: Adding a new node: /default-rack/172.17.2.64:50010
17/12/20 20:57:38 INFO net.NetworkTopology: Adding a new node: /default-rack/172.17.2.62:50010
17/12/20 20:57:38 INFO net.NetworkTopology: Adding a new node: /default-rack/172.17.2.65:50010
17/12/20 20:57:38 INFO balancer.Balancer: 0 over-utilized: []
17/12/20 20:57:38 INFO balancer.Balancer: 0 underutilized: []
The cluster is balanced. Exiting...
Dec 20, 2017 8:57:38 PM           0                  0 B                 0 B                0 B
Dec 20, 2017 8:57:38 PM  Balancing took 1.714 seconds

What am i missing?

Thanks for reply!

Robert

bkosaraju · ‎12-20-2017

Hi @Robert Jonczy,

The report you got is accurate, as I would like to stress on the parameter you have used "threshold"

-threshold 		<threshold>Percentage of disk capacity.

this is the value that balancer considered to have + or - of the percent of the "average DFS usage" to be moved

which is : % of DFS Used / total capacity

In your scenario it is < almost 1%, the threshold you specified (5%) which only works if there is a difference of 10%( +/- 5%) [not in your case ], hence it is not balancing anymore the data.

Hope this clarifies !!

View solution in original post

namaheshwari · ‎12-20-2017

Today the HDFS balancer doesn't balance disks within a DataNode

This is a pretty known and talked about issue with HDFS balancer.

See apache jira - https://issues.apache.org/jira/browse/HDFS-1312

The apache documentation is - https://hadoop.apache.org/docs/current/hadoop-project-dist/hadoop-hdfs/HDFSDiskbalancer.html

This tracks and resolve this issue.

bkosaraju · ‎12-20-2017