question Re: Why Standalone NiFi (PutHDFS) is slow ingesting small files in HDFS in Archives of Support Questions (Read Only)

Why Standalone NiFi (PutHDFS) is slow ingesting small files in HDFS

melvinmendoza — Sun, 18 Aug 2019 02:34:30 GMT

Having Problem with the slowness of PutHDFS to HDFS.

Re: Why Standalone NiFi (PutHDFS) is slow ingesting small files in HDFS

bbende — Thu, 04 May 2017 22:25:21 GMT

Have you tried installing the hadoop client on the same machine as NiFi and then sending a similar file to HDFS from the command line? In most cases the issue here is the network or disk.

Re: Why Standalone NiFi (PutHDFS) is slow ingesting small files in HDFS

melvinmendoza — Fri, 05 May 2017 11:22:16 GMT

@Bryan Bende

Btw, most of the file is small around 10~50 mb,

Upon testing:

350 Files, 10 mb per file, 1Gbps Network

Client to Server (SSH) - 36 secs

[admin@client1 data]$ cat transfer.logs Fri May 5 09:42:40 +08 2017 Fri May 5 09:43:16 +08 2017 36 secs

Server to HDFS - 4,479 secs or 1.27 Hrs

[hadoop@master1 data]$ ./puthdfs-stdf.sh Fri May 5 09:55:53 +08 2017 Fri May 5 11:10:32 +08 2017 4479 sec

[client@master1 data]$ cat puthdfs-stdf.sh
start=`date +%s`
echo `date`
hdfs dfs -put *.txt /data/test/testing/
echo `date`
end=`date +%s`
runtime=$((end-start))
echo $runtime

I haven't install Hadoop client in NIFI Server (a windows server)

Re: Why Standalone NiFi (PutHDFS) is slow ingesting small files in HDFS

melvinmendoza — Tue, 13 Jun 2017 10:45:39 GMT

@Bryan Bende Thanks!

Slow HDFS Put cause by a slow BlockReceiver.

2017-06-05 13:17:34,210 WARN  datanode.DataNode (BlockReceiver.java:receivePacket(571)) - Slow BlockReceiver write packet to mirror took 6316ms (threshold=300ms)