Sei sulla pagina 1di 4

hadv1:namenode,datanode,jobtracker,tasktracker and secondarynamenode

v1/mrv1:
storage related daemons: NN,DN,SNN
Processing related daemons: JT,TT
Cloudera: HDFS service-Roles:NN,DN,SNN
Mapreduce service-Roles:JT,TT
Hortonworks: HDFS service-components:NN,DN,SNN
Mapreduce service-components:JT,TT
hadv2:namenode,datanode,Resourcemanager,nodemanager,Appmaster and secondarynamen
ode(yes if federation)
,no(if HA)
v2/mrv2:
storage related daemons: NN,DN,SNN
Processing related daemons: RM,NM
Similarly like hadv1 as mentioned above
Note** if you dont mention the paths, each daemons uses system /tmp directory to
store the data.
m1-nn
/abc/name
m2-JT
m3-SNN,DN,TT
/abc/snnname
/abc/data1
m4-DN,TT
/abc/data2
m5-DN,TT
/abc/data3
===================
Download hadoop on all machines, edit the config files.
Before u start ur cluster,formatting of NN is mandatory
m1-NN (10 am)---we format our NN
/abc/name---in disk of nn
edits---to capture changes (4 bytes)
fsimage---consistent image of HDFS ( 112 bytes)---this was status of what we hav
e on hdfs
start ur cluster...
all ur daemons running..
NN<================DN DN DN DN...(starts sending HB every 3 sec)
JT<=================TT,TT,TT,TT...sending HB
10:05am
start writing data to cluster
hadoop.txt(file of 128mb): based on blksize of 64mb-2 blks created
edits----1048750 bytes (showing some change to hdfs has happened)
fsimage---112 bytes--as of 10:05am
NN<================DN DN DN DN(send block report every 10 seconds)
(metadata in namenode ram is getting built)
dns available
files written
files-blks
blks-which dns
file-permission
10:10am
edits---4 bytes(all changes from edits are applied to fsimage)????? who does thi
s
fsmimage---300 bytes---as of 10:10am(latest consistent status of ur hdfs)
SNN does this...
NN--has metadata in 2 places
in ram- bitmap with dn info,file info,file-blks info, blks-dn info, file permiss
ion
in disk-edits and fsimage (imp for nn to come up)
???who does this--SNN is doing checkpointing(by default : every 1 hr which is ch
angeable)
Note**
when u do a formatting...it creates name directory and name directory has the fo
llowing
/abc/name
edits
fsimage
fstime
VERSION-------------------->namespace ID(12345678)
after u start ur cluster,dn start talking to nn
/abc/data1
VERSION-------------------->namespace ID(12345678)
in hadoop2
/abc/name
edits_04inprogress
edits01
fsimage01
edits02
fsimage02
edits03
fsimage03....
when u do a formatting...it creates name directory and name directory has the fo
llowing
hadoop 2 without federation or HA
/abc/name
edits_04inprogress
edits01
fsimage01
edits02
fsimage02
edits03
fsimage03....
VERSION-------------------->namespace ID(12345678) + clusterid(xx)+bpoolid(for n
n)

after u start ur cluster,dn start talking to nn


/abc/data1
VERSION-------------------->clusterid(xx)
=========================
hadoop 2 with federation
ie multiple namenodes
nn1
/abc/name--bin/hadoop namenode -format
edits_04inprogress
edits01
fsimage01
edits02
fsimage02
edits03
fsimage03....
VERSION-------------------->namespace ID(12345678) + clusterid(xx)+bpoolid(for n
n)
nn2--bin/hadoop namenode -format -clusterid xx
/dezyre/name
edits_04inprogress
edits01
fsimage01
edits02
fsimage02
edits03
fsimage03....
VERSION-------------------->namespace ID(12345678-2) + clusterid(xx)+bpoolid(for
nn-2)
nn3---bin/hadoop namenode -format -clusterid xx
/dezyre/name
edits_04inprogress
edits01
fsimage01
edits02
fsimage02
edits03
fsimage03....
VERSION-------------------->namespace ID(12345678-3) + clusterid(xx)+bpoolid(for
nn-3)
after u start ur cluster,dn start talking to nn
/abc/data1
VERSION-------------------->clusterid(xx)
=============================
high availability
nn1
nn2
/abc/name
edits_04inprogress
edits01
fsimage01
edits02
fsimage02
edits03
fsimage03....
VERSION-------------------->namespace ID(12345678) + clusterid(xx)+bpoolid(for n
n)
nn2
nn2
/abc/name
edits_04inprogress
edits01
fsimage01
edits02
fsimage02
edits03
fsimage03....
VERSION-------------------->namespace ID(12345678) + clusterid(xx)+bpoolid(for n
n)
after u start ur cluster,dn start talking to nn
/dezyre/data1
VERSION-------------------->clusterid(xx)

Potrebbero piacerti anche