commit 39c007ce9cc3f46db70bde794323b6cbfccfe62a · pyrox.dev/nixpkgs

+73 -3

nixos/modules/services/cluster/hadoop/hdfs.nix

···

       24
       24
        
                 Whether to run the HDFS NameNode

     

       25
       25
        
               '';

     

       26
       26
        
             };

     

       27
       27
       +
             formatOnInit = mkOption {

     

       28
       28
       +
               type = types.bool;

     

       29
       29
       +
               default = false;

     

       30
       30
       +
               description = ''

     

       31
       31
       +
                 Format HDFS namenode on first start. This is useful for quickly spinning up ephemeral HDFS clusters with a single namenode.

     

       32
       32
       +
                 For HA clusters, initialization involves multiple steps across multiple nodes. Follow [this guide](https://hadoop.apache.org/docs/stable/hadoop-project-dist/hadoop-hdfs/HDFSHighAvailabilityWithQJM.html)

     

       33
       33
       +
                 to initialize an HA cluster manually.

     

       34
       34
       +
               '';

     

       35
       35
       +
             };

     

       27
       36
        
             inherit restartIfChanged;

     

       28
       37
        
             openFirewall = mkOption {

     

       29
       38
        
               type = types.bool;

     
···

       50
       59
        
               '';

     

       51
       60
        
             };

     

       52
       61
        
           };

     

       62
       62
       +
           journalnode = {

     

       63
       63
       +
             enabled = mkOption {

     

       64
       64
       +
               type = types.bool;

     

       65
       65
       +
               default = false;

     

       66
       66
       +
               description = ''

     

       67
       67
       +
                 Whether to run the HDFS JournalNode

     

       68
       68
       +
               '';

     

       69
       69
       +
             };

     

       70
       70
       +
             inherit restartIfChanged;

     

       71
       71
       +
             openFirewall = mkOption {

     

       72
       72
       +
               type = types.bool;

     

       73
       73
       +
               default = true;

     

       74
       74
       +
               description = ''

     

       75
       75
       +
                 Open firewall ports for journalnode

     

       76
       76
       +
               '';

     

       77
       77
       +
             };

     

       78
       78
       +
           };

     

       79
       79
       +
           zkfc = {

     

       80
       80
       +
             enabled = mkOption {

     

       81
       81
       +
               type = types.bool;

     

       82
       82
       +
               default = false;

     

       83
       83
       +
               description = ''

     

       84
       84
       +
                 Whether to run the HDFS ZooKeeper failover controller

     

       85
       85
       +
               '';

     

       86
       86
       +
             };

     

       87
       87
       +
             inherit restartIfChanged;

     

       88
       88
       +
           };

     

       53
       89
        
         };

     

       54
       90
        
       

     

       55
       91
        
         config = mkMerge [

     
···

       59
       95
        
               wantedBy = [ "multi-user.target" ];

     

       60
       96
        
               inherit (cfg.hdfs.namenode) restartIfChanged;

     

       61
       97
        
       

     

       62
       62
       -
               preStart = ''

     

       98
       98
       +
               preStart = (mkIf cfg.hdfs.namenode.formatOnInit ''

     

       63
       99
        
                 ${cfg.package}/bin/hdfs --config ${hadoopConf} namenode -format -nonInteractive || true

     

       64
       64
       -
               '';

     

       100
       100
       +
               '');

     

       65
       101
        
       

     

       66
       102
        
               serviceConfig = {

     

       67
       103
        
                 User = "hdfs";

     
···

       74
       110
        
             networking.firewall.allowedTCPPorts = (mkIf cfg.hdfs.namenode.openFirewall [

     

       75
       111
        
               9870 # namenode.http-address

     

       76
       112
        
               8020 # namenode.rpc-address

     

       113
       113
       +
               8022 # namenode. servicerpc-address

     

       77
       114
        
             ]);

     

       78
       115
        
           })

     

       79
       116
        
           (mkIf cfg.hdfs.datanode.enabled {

     
···

       96
       133
        
               9867 # datanode.ipc.address

     

       97
       134
        
             ]);

     

       98
       135
        
           })

     

       136
       136
       +
           (mkIf cfg.hdfs.journalnode.enabled {

     

       137
       137
       +
             systemd.services.hdfs-journalnode = {

     

       138
       138
       +
               description = "Hadoop HDFS JournalNode";

     

       139
       139
       +
               wantedBy = [ "multi-user.target" ];

     

       140
       140
       +
               inherit (cfg.hdfs.journalnode) restartIfChanged;

     

       141
       141
       +
       

     

       142
       142
       +
               serviceConfig = {

     

       143
       143
       +
                 User = "hdfs";

     

       144
       144
       +
                 SyslogIdentifier = "hdfs-journalnode";

     

       145
       145
       +
                 ExecStart = "${cfg.package}/bin/hdfs --config ${hadoopConf} journalnode";

     

       146
       146
       +
                 Restart = "always";

     

       147
       147
       +
               };

     

       148
       148
       +
             };

     

       149
       149
       +
       

     

       150
       150
       +
             networking.firewall.allowedTCPPorts = (mkIf cfg.hdfs.datanode.openFirewall [

     

       151
       151
       +
               8480 # dfs.journalnode.http-address

     

       152
       152
       +
               8485 # dfs.journalnode.rpc-address

     

       153
       153
       +
             ]);

     

       154
       154
       +
           })

     

       155
       155
       +
           (mkIf cfg.hdfs.zkfc.enabled {

     

       156
       156
       +
             systemd.services.hdfs-zkfc = {

     

       157
       157
       +
               description = "Hadoop HDFS ZooKeeper failover controller";

     

       158
       158
       +
               wantedBy = [ "multi-user.target" ];

     

       159
       159
       +
               inherit (cfg.hdfs.zkfc) restartIfChanged;

     

       160
       160
       +
       

     

       161
       161
       +
               serviceConfig = {

     

       162
       162
       +
                 User = "hdfs";

     

       163
       163
       +
                 SyslogIdentifier = "hdfs-zkfc";

     

       164
       164
       +
                 ExecStart = "${cfg.package}/bin/hdfs --config ${hadoopConf} zkfc";

     

       165
       165
       +
                 Restart = "always";

     

       166
       166
       +
               };

     

       167
       167
       +
             };

     

       168
       168
       +
           })

     

       99
       169
        
           (mkIf (

     

       100
       100
       -
               cfg.hdfs.namenode.enabled || cfg.hdfs.datanode.enabled

     

       170
       170
       +
               cfg.hdfs.namenode.enabled || cfg.hdfs.datanode.enabled || cfg.hdfs.journalnode.enabled || cfg.hdfs.zkfc.enabled

     

       101
       171
        
           ) {

     

       102
       172
        
             users.users.hdfs = {

     

       103
       173
        
               description = "Hadoop HDFS user";

nixos/modules/services/cluster/hadoop/yarn.nix

···

       91
       91
        
               8030 # resourcemanager.scheduler.address

     

       92
       92
        
               8031 # resourcemanager.resource-tracker.address

     

       93
       93
        
               8032 # resourcemanager.address

     

       94
       94
       +
               8033 # resourcemanager.admin.address

     

       94
       95
        
             ]);

     

       95
       96
        
           })

     

       96
       97

+198 -38

nixos/tests/hadoop/hadoop.nix

···

       1
       1
       +
       # This test is very comprehensive. It tests whether all hadoop services work well with each other.

     

       2
       2
       +
       # Run this when updating the Hadoop package or making significant changes to the hadoop module.

     

       3
       3
       +
       # For a more basic test, see hdfs.nix and yarn.nix

     

       1
       4
        
       import ../make-test-python.nix ({pkgs, ...}: {

     

       2
       5
        
       

     

       3
       6
        
         nodes = let

     

       4
       7
        
           package = pkgs.hadoop;

     

       5
       8
        
           coreSite = {

     

       6
       6
       -
             "fs.defaultFS" = "hdfs://master";

     

       9
       9
       +
             "fs.defaultFS" = "hdfs://ns1";

     

       10
       10
       +
           };

     

       11
       11
       +
           hdfsSite = {

     

       12
       12
       +
             "dfs.namenode.rpc-bind-host" = "0.0.0.0";

     

       13
       13
       +
             "dfs.namenode.http-bind-host" = "0.0.0.0";

     

       14
       14
       +
             "dfs.namenode.servicerpc-bind-host" = "0.0.0.0";

     

       15
       15
       +
       

     

       16
       16
       +
             # HA Quorum Journal Manager configuration

     

       17
       17
       +
             "dfs.nameservices" = "ns1";

     

       18
       18
       +
             "dfs.ha.namenodes.ns1" = "nn1,nn2";

     

       19
       19
       +
             "dfs.namenode.shared.edits.dir.ns1.nn1" = "qjournal://jn1:8485;jn2:8485;jn3:8485/ns1";

     

       20
       20
       +
             "dfs.namenode.shared.edits.dir.ns1.nn2" = "qjournal://jn1:8485;jn2:8485;jn3:8485/ns1";

     

       21
       21
       +
             "dfs.namenode.rpc-address.ns1.nn1" = "nn1:8020";

     

       22
       22
       +
             "dfs.namenode.rpc-address.ns1.nn2" = "nn2:8020";

     

       23
       23
       +
             "dfs.namenode.servicerpc-address.ns1.nn1" = "nn1:8022";

     

       24
       24
       +
             "dfs.namenode.servicerpc-address.ns1.nn2" = "nn2:8022";

     

       25
       25
       +
             "dfs.namenode.http-address.ns1.nn1" = "nn1:9870";

     

       26
       26
       +
             "dfs.namenode.http-address.ns1.nn2" = "nn2:9870";

     

       27
       27
       +
       

     

       28
       28
       +
             # Automatic failover configuration

     

       29
       29
       +
             "dfs.client.failover.proxy.provider.ns1" = "org.apache.hadoop.hdfs.server.namenode.ha.ConfiguredFailoverProxyProvider";

     

       30
       30
       +
             "dfs.ha.automatic-failover.enabled.ns1" = "true";

     

       31
       31
       +
             "dfs.ha.fencing.methods" = "shell(true)";

     

       32
       32
       +
             "ha.zookeeper.quorum" = "zk1:2181";

     

       33
       33
       +
           };

     

       34
       34
       +
           yarnSiteHA = {

     

       35
       35
       +
             "yarn.resourcemanager.zk-address" = "zk1:2181";

     

       36
       36
       +
             "yarn.resourcemanager.ha.enabled" = "true";

     

       37
       37
       +
             "yarn.resourcemanager.ha.rm-ids" = "rm1,rm2";

     

       38
       38
       +
             "yarn.resourcemanager.hostname.rm1" = "rm1";

     

       39
       39
       +
             "yarn.resourcemanager.hostname.rm2" = "rm2";

     

       40
       40
       +
             "yarn.resourcemanager.ha.automatic-failover.enabled" = "true";

     

       41
       41
       +
             "yarn.resourcemanager.cluster-id" = "cluster1";

     

       42
       42
       +
             # yarn.resourcemanager.webapp.address needs to be defined even though yarn.resourcemanager.hostname is set. This shouldn't be necessary, but there's a bug in

     

       43
       43
       +
             # hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-web-proxy/src/main/java/org/apache/hadoop/yarn/server/webproxy/amfilter/AmFilterInitializer.java:70

     

       44
       44
       +
             # that causes AM containers to fail otherwise.

     

       45
       45
       +
             "yarn.resourcemanager.webapp.address.rm1" = "rm1:8088";

     

       46
       46
       +
             "yarn.resourcemanager.webapp.address.rm2" = "rm2:8088";

     

       7
       47
        
           };

     

       8
       48
        
         in {

     

       9
       9
       -
           master = {pkgs, options, ...}: {

     

       49
       49
       +
           zk1 = { ... }: {

     

       50
       50
       +
             services.zookeeper.enable = true;

     

       51
       51
       +
             networking.firewall.allowedTCPPorts = [ 2181 ];

     

       52
       52
       +
           };

     

       53
       53
       +
       

     

       54
       54
       +
           # HDFS cluster

     

       55
       55
       +
           nn1 = {pkgs, options, ...}: {

     

       10
       56
        
             services.hadoop = {

     

       11
       11
       -
               inherit package coreSite;

     

       57
       57
       +
               inherit package coreSite hdfsSite;

     

       12
       58
        
               hdfs.namenode.enabled = true;

     

       13
       13
       -
               yarn.resourcemanager.enabled = true;

     

       59
       59
       +
               hdfs.zkfc.enabled = true;

     

       14
       60
        
             };

     

       15
       15
       -
             virtualisation.memorySize = 1024;

     

       61
       61
       +
           };

     

       62
       62
       +
           nn2 = {pkgs, options, ...}: {

     

       63
       63
       +
             services.hadoop = {

     

       64
       64
       +
               inherit package coreSite hdfsSite;

     

       65
       65
       +
               hdfs.namenode.enabled = true;

     

       66
       66
       +
               hdfs.zkfc.enabled = true;

     

       67
       67
       +
             };

     

       16
       68
        
           };

     

       17
       69
        
       

     

       18
       18
       -
           worker = {pkgs, options, ...}: {

     

       70
       70
       +
           jn1 = {pkgs, options, ...}: {

     

       71
       71
       +
             services.hadoop = {

     

       72
       72
       +
               inherit package coreSite hdfsSite;

     

       73
       73
       +
               hdfs.journalnode.enabled = true;

     

       74
       74
       +
             };

     

       75
       75
       +
           };

     

       76
       76
       +
           jn2 = {pkgs, options, ...}: {

     

       19
       77
        
             services.hadoop = {

     

       20
       20
       -
               inherit package coreSite;

     

       78
       78
       +
               inherit package coreSite hdfsSite;

     

       79
       79
       +
               hdfs.journalnode.enabled = true;

     

       80
       80
       +
             };

     

       81
       81
       +
           };

     

       82
       82
       +
           jn3 = {pkgs, options, ...}: {

     

       83
       83
       +
             services.hadoop = {

     

       84
       84
       +
               inherit package coreSite hdfsSite;

     

       85
       85
       +
               hdfs.journalnode.enabled = true;

     

       86
       86
       +
             };

     

       87
       87
       +
           };

     

       88
       88
       +
       

     

       89
       89
       +
           dn1 = {pkgs, options, ...}: {

     

       90
       90
       +
             services.hadoop = {

     

       91
       91
       +
               inherit package coreSite hdfsSite;

     

       21
       92
        
               hdfs.datanode.enabled = true;

     

       93
       93
       +
             };

     

       94
       94
       +
           };

     

       95
       95
       +
       

     

       96
       96
       +
           # YARN cluster

     

       97
       97
       +
           rm1 = {pkgs, options, ...}: {

     

       98
       98
       +
             virtualisation.memorySize = 1024;

     

       99
       99
       +
             services.hadoop = {

     

       100
       100
       +
               inherit package coreSite hdfsSite;

     

       101
       101
       +
               yarnSite = options.services.hadoop.yarnSite.default // yarnSiteHA;

     

       102
       102
       +
               yarn.resourcemanager.enabled = true;

     

       103
       103
       +
             };

     

       104
       104
       +
           };

     

       105
       105
       +
           rm2 = {pkgs, options, ...}: {

     

       106
       106
       +
             virtualisation.memorySize = 1024;

     

       107
       107
       +
             services.hadoop = {

     

       108
       108
       +
               inherit package coreSite hdfsSite;

     

       109
       109
       +
               yarnSite = options.services.hadoop.yarnSite.default // yarnSiteHA;

     

       110
       110
       +
               yarn.resourcemanager.enabled = true;

     

       111
       111
       +
             };

     

       112
       112
       +
           };

     

       113
       113
       +
           nm1 = {pkgs, options, ...}: {

     

       114
       114
       +
             virtualisation.memorySize = 2048;

     

       115
       115
       +
             services.hadoop = {

     

       116
       116
       +
               inherit package coreSite hdfsSite;

     

       117
       117
       +
               yarnSite = options.services.hadoop.yarnSite.default // yarnSiteHA;

     

       22
       118
        
               yarn.nodemanager.enabled = true;

     

       23
       23
       -
               yarnSite = options.services.hadoop.yarnSite.default // {

     

       24
       24
       -
                 "yarn.resourcemanager.hostname" = "master";

     

       25
       25
       -
               };

     

       26
       119
        
             };

     

       27
       27
       -
             virtualisation.memorySize = 2048;

     

       28
       120
        
           };

     

       29
       121
        
         };

     

       30
       122
        
       

     

       31
       123
        
         testScript = ''

     

       32
       124
        
           start_all()

     

       33
       125
        
       

     

       34
       34
       -
           master.wait_for_unit("network.target")

     

       35
       35
       -
           master.wait_for_unit("hdfs-namenode")

     

       126
       126
       +
           #### HDFS tests ####

     

       36
       127
        
       

     

       37
       37
       -
           master.wait_for_open_port(8020)

     

       38
       38
       -
           master.wait_for_open_port(9870)

     

       128
       128
       +
           zk1.wait_for_unit("network.target")

     

       129
       129
       +
           jn1.wait_for_unit("network.target")

     

       130
       130
       +
           jn2.wait_for_unit("network.target")

     

       131
       131
       +
           jn3.wait_for_unit("network.target")

     

       132
       132
       +
           nn1.wait_for_unit("network.target")

     

       133
       133
       +
           nn2.wait_for_unit("network.target")

     

       134
       134
       +
           dn1.wait_for_unit("network.target")

     

       39
       135
        
       

     

       40
       40
       -
           worker.wait_for_unit("network.target")

     

       41
       41
       -
           worker.wait_for_unit("hdfs-datanode")

     

       42
       42
       -
           worker.wait_for_open_port(9864)

     

       43
       43
       -
           worker.wait_for_open_port(9866)

     

       44
       44
       -
           worker.wait_for_open_port(9867)

     

       136
       136
       +
           zk1.wait_for_unit("zookeeper")

     

       137
       137
       +
           jn1.wait_for_unit("hdfs-journalnode")

     

       138
       138
       +
           jn2.wait_for_unit("hdfs-journalnode")

     

       139
       139
       +
           jn3.wait_for_unit("hdfs-journalnode")

     

       45
       140
        
       

     

       46
       46
       -
           master.succeed("curl -f http://worker:9864")

     

       47
       47
       -
           worker.succeed("curl -f http://master:9870")

     

       141
       141
       +
           zk1.wait_for_open_port(2181)

     

       142
       142
       +
           jn1.wait_for_open_port(8480)

     

       143
       143
       +
           jn1.wait_for_open_port(8485)

     

       144
       144
       +
           jn2.wait_for_open_port(8480)

     

       145
       145
       +
           jn2.wait_for_open_port(8485)

     

       48
       146
        
       

     

       49
       49
       -
           worker.succeed("sudo -u hdfs hdfs dfsadmin -safemode wait")

     

       147
       147
       +
           # Namenodes must be stopped before initializing the cluster

     

       148
       148
       +
           nn1.succeed("systemctl stop hdfs-namenode")

     

       149
       149
       +
           nn2.succeed("systemctl stop hdfs-namenode")

     

       150
       150
       +
           nn1.succeed("systemctl stop hdfs-zkfc")

     

       151
       151
       +
           nn2.succeed("systemctl stop hdfs-zkfc")

     

       50
       152
        
       

     

       51
       51
       -
           master.wait_for_unit("yarn-resourcemanager")

     

       153
       153
       +
           # Initialize zookeeper for failover controller

     

       154
       154
       +
           nn1.succeed("sudo -u hdfs hdfs zkfc -formatZK 2>&1 | systemd-cat")

     

       52
       155
        
       

     

       53
       53
       -
           master.wait_for_open_port(8030)

     

       54
       54
       -
           master.wait_for_open_port(8031)

     

       55
       55
       -
           master.wait_for_open_port(8032)

     

       56
       56
       -
           master.wait_for_open_port(8088)

     

       57
       57
       -
           worker.succeed("curl -f http://master:8088")

     

       156
       156
       +
           # Format NN1 and start it

     

       157
       157
       +
           nn1.succeed("sudo -u hdfs hadoop namenode -format 2>&1 | systemd-cat")

     

       158
       158
       +
           nn1.succeed("systemctl start hdfs-namenode")

     

       159
       159
       +
           nn1.wait_for_open_port(9870)

     

       160
       160
       +
           nn1.wait_for_open_port(8022)

     

       161
       161
       +
           nn1.wait_for_open_port(8020)

     

       58
       162
        
       

     

       59
       59
       -
           worker.wait_for_unit("yarn-nodemanager")

     

       60
       60
       -
           worker.wait_for_open_port(8042)

     

       61
       61
       -
           worker.wait_for_open_port(8040)

     

       62
       62
       -
           master.succeed("curl -f http://worker:8042")

     

       163
       163
       +
           # Bootstrap NN2 from NN1 and start it

     

       164
       164
       +
           nn2.succeed("sudo -u hdfs hdfs namenode -bootstrapStandby 2>&1 | systemd-cat")

     

       165
       165
       +
           nn2.succeed("systemctl start hdfs-namenode")

     

       166
       166
       +
           nn2.wait_for_open_port(9870)

     

       167
       167
       +
           nn2.wait_for_open_port(8022)

     

       168
       168
       +
           nn2.wait_for_open_port(8020)

     

       169
       169
       +
           nn1.succeed("netstat -tulpne | systemd-cat")

     

       63
       170
        
       

     

       64
       64
       -
           assert "Total Nodes:1" in worker.succeed("yarn node -list")

     

       171
       171
       +
           # Start failover controllers

     

       172
       172
       +
           nn1.succeed("systemctl start hdfs-zkfc")

     

       173
       173
       +
           nn2.succeed("systemctl start hdfs-zkfc")

     

       65
       174
        
       

     

       66
       66
       -
           assert "Estimated value of Pi is" in worker.succeed("HADOOP_USER_NAME=hdfs yarn jar $(readlink $(which yarn) | sed -r 's~bin/yarn~lib/hadoop-*/share/hadoop/mapreduce/hadoop-mapreduce-examples-*.jar~g') pi 2 10")

     

       67
       67
       -
           assert "SUCCEEDED" in worker.succeed("yarn application -list -appStates FINISHED")

     

       68
       68
       -
           worker.succeed("sudo -u hdfs hdfs dfs -ls / | systemd-cat")

     

       175
       175
       +
           # DN should have started by now, but confirm anyway

     

       176
       176
       +
           dn1.wait_for_unit("hdfs-datanode")

     

       177
       177
       +
           # Print states of namenodes

     

       178
       178
       +
           dn1.succeed("sudo -u hdfs hdfs haadmin -getAllServiceState | systemd-cat")

     

       179
       179
       +
           # Wait for cluster to exit safemode

     

       180
       180
       +
           dn1.succeed("sudo -u hdfs hdfs dfsadmin -safemode wait")

     

       181
       181
       +
           dn1.succeed("sudo -u hdfs hdfs haadmin -getAllServiceState | systemd-cat")

     

       182
       182
       +
           # test R/W

     

       183
       183
       +
           dn1.succeed("echo testfilecontents | sudo -u hdfs hdfs dfs -put - /testfile")

     

       184
       184
       +
           assert "testfilecontents" in dn1.succeed("sudo -u hdfs hdfs dfs -cat /testfile")

     

       185
       185
       +
       

     

       186
       186
       +
           # Test NN failover

     

       187
       187
       +
           nn1.succeed("systemctl stop hdfs-namenode")

     

       188
       188
       +
           assert "active" in dn1.succeed("sudo -u hdfs hdfs haadmin -getAllServiceState")

     

       189
       189
       +
           dn1.succeed("sudo -u hdfs hdfs haadmin -getAllServiceState | systemd-cat")

     

       190
       190
       +
           assert "testfilecontents" in dn1.succeed("sudo -u hdfs hdfs dfs -cat /testfile")

     

       191
       191
       +
       

     

       192
       192
       +
           nn1.succeed("systemctl start hdfs-namenode")

     

       193
       193
       +
           nn1.wait_for_open_port(9870)

     

       194
       194
       +
           nn1.wait_for_open_port(8022)

     

       195
       195
       +
           nn1.wait_for_open_port(8020)

     

       196
       196
       +
           assert "standby" in dn1.succeed("sudo -u hdfs hdfs haadmin -getAllServiceState")

     

       197
       197
       +
           dn1.succeed("sudo -u hdfs hdfs haadmin -getAllServiceState | systemd-cat")

     

       198
       198
       +
       

     

       199
       199
       +
           #### YARN tests ####

     

       200
       200
       +
       

     

       201
       201
       +
           rm1.wait_for_unit("network.target")

     

       202
       202
       +
           rm2.wait_for_unit("network.target")

     

       203
       203
       +
           nm1.wait_for_unit("network.target")

     

       204
       204
       +
       

     

       205
       205
       +
           rm1.wait_for_unit("yarn-resourcemanager")

     

       206
       206
       +
           rm1.wait_for_open_port(8088)

     

       207
       207
       +
           rm2.wait_for_unit("yarn-resourcemanager")

     

       208
       208
       +
           rm2.wait_for_open_port(8088)

     

       209
       209
       +
       

     

       210
       210
       +
           nm1.wait_for_unit("yarn-nodemanager")

     

       211
       211
       +
           nm1.wait_for_open_port(8042)

     

       212
       212
       +
           nm1.wait_for_open_port(8040)

     

       213
       213
       +
           nm1.wait_until_succeeds("yarn node -list | grep Nodes:1")

     

       214
       214
       +
           nm1.succeed("sudo -u yarn yarn rmadmin -getAllServiceState | systemd-cat")

     

       215
       215
       +
           nm1.succeed("sudo -u yarn yarn node -list | systemd-cat")

     

       216
       216
       +
       

     

       217
       217
       +
           # Test RM failover

     

       218
       218
       +
           rm1.succeed("systemctl stop yarn-resourcemanager")

     

       219
       219
       +
           assert "standby" not in nm1.succeed("sudo -u yarn yarn rmadmin -getAllServiceState")

     

       220
       220
       +
           nm1.succeed("sudo -u yarn yarn rmadmin -getAllServiceState | systemd-cat")

     

       221
       221
       +
           rm1.succeed("systemctl start yarn-resourcemanager")

     

       222
       222
       +
           rm1.wait_for_unit("yarn-resourcemanager")

     

       223
       223
       +
           rm1.wait_for_open_port(8088)

     

       224
       224
       +
           assert "standby" in nm1.succeed("sudo -u yarn yarn rmadmin -getAllServiceState")

     

       225
       225
       +
           nm1.succeed("sudo -u yarn yarn rmadmin -getAllServiceState | systemd-cat")

     

       226
       226
       +
       

     

       227
       227
       +
           assert "Estimated value of Pi is" in nm1.succeed("HADOOP_USER_NAME=hdfs yarn jar $(readlink $(which yarn) | sed -r 's~bin/yarn~lib/hadoop-*/share/hadoop/mapreduce/hadoop-mapreduce-examples-*.jar~g') pi 2 10")

     

       228
       228
       +
           assert "SUCCEEDED" in nm1.succeed("yarn application -list -appStates FINISHED")

     

       69
       229
        
         '';

     

       70
       70
       -
        })

     

       230
       230
       +
       })

+9 -10

nixos/tests/hadoop/hdfs.nix

···

       1
       1
       +
       # Test a minimal HDFS cluster with no HA

     

       1
       2
        
       import ../make-test-python.nix ({...}: {

     

       2
       3
        
         nodes = {

     

       3
       4
        
           namenode = {pkgs, ...}: {

     

       4
       5
        
             services.hadoop = {

     

       5
       6
        
               package = pkgs.hadoop;

     

       6
       6
       -
               hdfs.namenode.enabled = true;

     

       7
       7
       +
               hdfs.namenode = {

     

       8
       8
       +
                 enabled = true;

     

       9
       9
       +
                 formatOnInit = true;

     

       10
       10
       +
               };

     

       7
       11
        
               coreSite = {

     

       8
       12
        
                 "fs.defaultFS" = "hdfs://namenode:8020";

     

       9
       13
        
               };

     
···

       13
       17
        
                 "dfs.namenode.http-bind-host" = "0.0.0.0";

     

       14
       18
        
               };

     

       15
       19
        
             };

     

       16
       16
       -
             networking.firewall.allowedTCPPorts = [

     

       17
       17
       -
               9870 # namenode.http-address

     

       18
       18
       -
               8020 # namenode.rpc-address

     

       19
       19
       -
             ];

     

       20
       20
        
           };

     

       21
       21
        
           datanode = {pkgs, ...}: {

     

       22
       22
        
             services.hadoop = {

     
···

       26
       26
        
                 "fs.defaultFS" = "hdfs://namenode:8020";

     

       27
       27
        
               };

     

       28
       28
        
             };

     

       29
       29
       -
             networking.firewall.allowedTCPPorts = [

     

       30
       30
       -
               9864 # datanode.http.address

     

       31
       31
       -
               9866 # datanode.address

     

       32
       32
       -
               9867 # datanode.ipc.address

     

       33
       33
       -
             ];

     

       34
       29
        
           };

     

       35
       30
        
         };

     

       36
       31
        
       

     
···

       50
       45
        
       

     

       51
       46
        
           namenode.succeed("curl -f http://namenode:9870")

     

       52
       47
        
           datanode.succeed("curl -f http://datanode:9864")

     

       48
       48
       +
       

     

       49
       49
       +
           datanode.succeed("sudo -u hdfs hdfs dfsadmin -safemode wait")

     

       50
       50
       +
           datanode.succeed("echo testfilecontents | sudo -u hdfs hdfs dfs -put - /testfile")

     

       51
       51
       +
           assert "testfilecontents" in datanode.succeed("sudo -u hdfs hdfs dfs -cat /testfile")

     

       53
       52
        
         '';

     

       54
       53
        
       })

+1 -10

nixos/tests/hadoop/yarn.nix

···

       1
       1
       +
       # This only tests if YARN is able to start its services

     

       1
       2
        
       import ../make-test-python.nix ({...}: {

     

       2
       3
        
         nodes = {

     

       3
       4
        
           resourcemanager = {pkgs, ...}: {

     
···

       6
       7
        
             services.hadoop.yarnSite = {

     

       7
       8
        
               "yarn.resourcemanager.scheduler.class" = "org.apache.hadoop.yarn.server.resourcemanager.scheduler.fifo.FifoScheduler";

     

       8
       9
        
             };

     

       9
       9
       -
             networking.firewall.allowedTCPPorts = [

     

       10
       10
       -
               8088 # resourcemanager.webapp.address

     

       11
       11
       -
               8031 # resourcemanager.resource-tracker.address

     

       12
       12
       -
             ];

     

       13
       10
        
           };

     

       14
       11
        
           nodemanager = {pkgs, ...}: {

     

       15
       12
        
             services.hadoop.package = pkgs.hadoop;

     
···

       17
       14
        
             services.hadoop.yarnSite = {

     

       18
       15
        
               "yarn.resourcemanager.hostname" = "resourcemanager";

     

       19
       16
        
               "yarn.nodemanager.log-dirs" = "/tmp/userlogs";

     

       20
       20
       -
               "yarn.nodemanager.address" = "0.0.0.0:8041";

     

       21
       17
        
             };

     

       22
       22
       -
             networking.firewall.allowedTCPPorts = [

     

       23
       23
       -
               8042 # nodemanager.webapp.address

     

       24
       24
       -
               8041 # nodemanager.address

     

       25
       25
       -
             ];

     

       26
       18
        
           };

     

       27
       19
        
       

     

       28
       20
        
         };

     
···

       38
       30
        
           nodemanager.wait_for_unit("yarn-nodemanager")

     

       39
       31
        
           nodemanager.wait_for_unit("network.target")

     

       40
       32
        
           nodemanager.wait_for_open_port(8042)

     

       41
       41
       -
           nodemanager.wait_for_open_port(8041)

     

       42
       33
        
       

     

       43
       34
        
           resourcemanager.succeed("curl -f http://localhost:8088")

     

       44
       35
        
           nodemanager.succeed("curl -f http://localhost:8042")