nixos/tests/hadoop/hadoop.nix

   1 # This test is very comprehensive. It tests whether all hadoop services work well with each other.
   2 # Run this when updating the Hadoop package or making significant changes to the hadoop module.
   3 # For a more basic test, see hdfs.nix and yarn.nix
   4 import ../make-test-python.nix ({ package, ... }: {
   5   name = "hadoop-combined";
   6
   7   nodes =
   8     let
   9       coreSite = {
  10         "fs.defaultFS" = "hdfs://ns1";
  11       };
  12       hdfsSite = {
  13         # HA Quorum Journal Manager configuration
  14         "dfs.nameservices" = "ns1";
  15         "dfs.ha.namenodes.ns1" = "nn1,nn2";
  16         "dfs.namenode.shared.edits.dir.ns1" = "qjournal://jn1:8485;jn2:8485;jn3:8485/ns1";
  17         "dfs.namenode.rpc-address.ns1.nn1" = "nn1:8020";
  18         "dfs.namenode.rpc-address.ns1.nn2" = "nn2:8020";
  19         "dfs.namenode.servicerpc-address.ns1.nn1" = "nn1:8022";
  20         "dfs.namenode.servicerpc-address.ns1.nn2" = "nn2:8022";
  21         "dfs.namenode.http-address.ns1.nn1" = "nn1:9870";
  22         "dfs.namenode.http-address.ns1.nn2" = "nn2:9870";
  23
  24         # Automatic failover configuration
  25         "dfs.client.failover.proxy.provider.ns1" = "org.apache.hadoop.hdfs.server.namenode.ha.ConfiguredFailoverProxyProvider";
  26         "dfs.ha.automatic-failover.enabled.ns1" = "true";
  27         "dfs.ha.fencing.methods" = "shell(true)";
  28         "ha.zookeeper.quorum" = "zk1:2181";
  29       };
  30       yarnSite = {
  31         "yarn.resourcemanager.zk-address" = "zk1:2181";
  32         "yarn.resourcemanager.ha.enabled" = "true";
  33         "yarn.resourcemanager.ha.rm-ids" = "rm1,rm2";
  34         "yarn.resourcemanager.hostname.rm1" = "rm1";
  35         "yarn.resourcemanager.hostname.rm2" = "rm2";
  36         "yarn.resourcemanager.ha.automatic-failover.enabled" = "true";
  37         "yarn.resourcemanager.cluster-id" = "cluster1";
  38         # 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
  39         # 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
  40         # that causes AM containers to fail otherwise.
  41         "yarn.resourcemanager.webapp.address.rm1" = "rm1:8088";
  42         "yarn.resourcemanager.webapp.address.rm2" = "rm2:8088";
  43       };
  44     in
  45     {
  46       zk1 = { ... }: {
  47         services.zookeeper.enable = true;
  48         networking.firewall.allowedTCPPorts = [ 2181 ];
  49       };
  50
  51       # HDFS cluster
  52       nn1 = { ... }: {
  53         services.hadoop = {
  54           inherit package coreSite hdfsSite;
  55           hdfs.namenode = {
  56             enable = true;
  57             openFirewall = true;
  58           };
  59           hdfs.zkfc.enable = true;
  60         };
  61       };
  62       nn2 = { ... }: {
  63         services.hadoop = {
  64           inherit package coreSite hdfsSite;
  65           hdfs.namenode = {
  66             enable = true;
  67             openFirewall = true;
  68           };
  69           hdfs.zkfc.enable = true;
  70         };
  71       };
  72
  73       jn1 = { ... }: {
  74         services.hadoop = {
  75           inherit package coreSite hdfsSite;
  76           hdfs.journalnode = {
  77             enable = true;
  78             openFirewall = true;
  79           };
  80         };
  81       };
  82       jn2 = { ... }: {
  83         services.hadoop = {
  84           inherit package coreSite hdfsSite;
  85           hdfs.journalnode = {
  86             enable = true;
  87             openFirewall = true;
  88           };
  89         };
  90       };
  91       jn3 = { ... }: {
  92         services.hadoop = {
  93           inherit package coreSite hdfsSite;
  94           hdfs.journalnode = {
  95             enable = true;
  96             openFirewall = true;
  97           };
  98         };
  99       };
 100
 101       dn1 = { ... }: {
 102         virtualisation.diskSize = 4096;
 103         services.hadoop = {
 104           inherit package coreSite hdfsSite;
 105           hdfs.datanode = {
 106             enable = true;
 107             openFirewall = true;
 108           };
 109         };
 110       };
 111
 112       # YARN cluster
 113       rm1 = { options, ... }: {
 114         services.hadoop = {
 115           inherit package coreSite hdfsSite yarnSite;
 116           yarn.resourcemanager = {
 117             enable = true;
 118             openFirewall = true;
 119           };
 120         };
 121       };
 122       rm2 = { options, ... }: {
 123         services.hadoop = {
 124           inherit package coreSite hdfsSite yarnSite;
 125           yarn.resourcemanager = {
 126             enable = true;
 127             openFirewall = true;
 128           };
 129         };
 130       };
 131       nm1 = { options, ... }: {
 132         virtualisation.memorySize = 2048;
 133         services.hadoop = {
 134           inherit package coreSite hdfsSite yarnSite;
 135           yarn.nodemanager = {
 136             enable = true;
 137             openFirewall = true;
 138           };
 139         };
 140       };
 141       client = { options, ... }: {
 142         services.hadoop = {
 143           gatewayRole.enable = true;
 144           inherit package coreSite hdfsSite yarnSite;
 145         };
 146       };
 147   };
 148
 149   testScript = ''
 150     start_all()
 151
 152     #### HDFS tests ####
 153
 154     zk1.wait_for_unit("network.target")
 155     jn1.wait_for_unit("network.target")
 156     jn2.wait_for_unit("network.target")
 157     jn3.wait_for_unit("network.target")
 158     nn1.wait_for_unit("network.target")
 159     nn2.wait_for_unit("network.target")
 160     dn1.wait_for_unit("network.target")
 161
 162     zk1.wait_for_unit("zookeeper")
 163     jn1.wait_for_unit("hdfs-journalnode")
 164     jn2.wait_for_unit("hdfs-journalnode")
 165     jn3.wait_for_unit("hdfs-journalnode")
 166
 167     zk1.wait_for_open_port(2181)
 168     jn1.wait_for_open_port(8480)
 169     jn1.wait_for_open_port(8485)
 170     jn2.wait_for_open_port(8480)
 171     jn2.wait_for_open_port(8485)
 172
 173     # Namenodes must be stopped before initializing the cluster
 174     nn1.succeed("systemctl stop hdfs-namenode")
 175     nn2.succeed("systemctl stop hdfs-namenode")
 176     nn1.succeed("systemctl stop hdfs-zkfc")
 177     nn2.succeed("systemctl stop hdfs-zkfc")
 178
 179     # Initialize zookeeper for failover controller
 180     nn1.succeed("sudo -u hdfs systemd-cat hdfs zkfc -formatZK")
 181
 182     # Format NN1 and start it
 183     nn1.succeed("sudo -u hdfs systemd-cat hadoop namenode -format")
 184     nn1.succeed("systemctl start hdfs-namenode")
 185     nn1.wait_for_open_port(9870)
 186     nn1.wait_for_open_port(8022)
 187     nn1.wait_for_open_port(8020)
 188
 189     # Bootstrap NN2 from NN1 and start it
 190     nn2.succeed("sudo -u hdfs systemd-cat hdfs namenode -bootstrapStandby")
 191     nn2.succeed("systemctl start hdfs-namenode")
 192     nn2.wait_for_open_port(9870)
 193     nn2.wait_for_open_port(8022)
 194     nn2.wait_for_open_port(8020)
 195     nn1.succeed("systemd-cat netstat -tulpne")
 196
 197     # Start failover controllers
 198     nn1.succeed("systemctl start hdfs-zkfc")
 199     nn2.succeed("systemctl start hdfs-zkfc")
 200
 201     # DN should have started by now, but confirm anyway
 202     dn1.wait_for_unit("hdfs-datanode")
 203     # Print states of namenodes
 204     client.succeed("sudo -u hdfs systemd-cat hdfs haadmin -getAllServiceState")
 205     # Wait for cluster to exit safemode
 206     client.succeed("sudo -u hdfs hdfs dfsadmin -safemode wait")
 207     client.succeed("sudo -u hdfs systemd-cat hdfs haadmin -getAllServiceState")
 208     # test R/W
 209     client.succeed("echo testfilecontents | sudo -u hdfs hdfs dfs -put - /testfile")
 210     assert "testfilecontents" in client.succeed("sudo -u hdfs hdfs dfs -cat /testfile")
 211
 212     # Test NN failover
 213     nn1.succeed("systemctl stop hdfs-namenode")
 214     assert "active" in client.succeed("sudo -u hdfs hdfs haadmin -getAllServiceState")
 215     client.succeed("sudo -u hdfs systemd-cat hdfs haadmin -getAllServiceState")
 216     assert "testfilecontents" in client.succeed("sudo -u hdfs hdfs dfs -cat /testfile")
 217
 218     nn1.succeed("systemctl start hdfs-namenode")
 219     nn1.wait_for_open_port(9870)
 220     nn1.wait_for_open_port(8022)
 221     nn1.wait_for_open_port(8020)
 222     assert "standby" in client.succeed("sudo -u hdfs hdfs haadmin -getAllServiceState")
 223     client.succeed("sudo -u hdfs systemd-cat hdfs haadmin -getAllServiceState")
 224
 225     #### YARN tests ####
 226
 227     rm1.wait_for_unit("network.target")
 228     rm2.wait_for_unit("network.target")
 229     nm1.wait_for_unit("network.target")
 230
 231     rm1.wait_for_unit("yarn-resourcemanager")
 232     rm1.wait_for_open_port(8088)
 233     rm2.wait_for_unit("yarn-resourcemanager")
 234     rm2.wait_for_open_port(8088)
 235
 236     nm1.wait_for_unit("yarn-nodemanager")
 237     nm1.wait_for_open_port(8042)
 238     nm1.wait_for_open_port(8040)
 239     client.wait_until_succeeds("yarn node -list | grep Nodes:1")
 240     client.succeed("sudo -u yarn systemd-cat yarn rmadmin -getAllServiceState")
 241     client.succeed("sudo -u yarn systemd-cat yarn node -list")
 242
 243     # Test RM failover
 244     rm1.succeed("systemctl stop yarn-resourcemanager")
 245     assert "standby" not in client.succeed("sudo -u yarn yarn rmadmin -getAllServiceState")
 246     client.succeed("sudo -u yarn systemd-cat yarn rmadmin -getAllServiceState")
 247     rm1.succeed("systemctl start yarn-resourcemanager")
 248     rm1.wait_for_unit("yarn-resourcemanager")
 249     rm1.wait_for_open_port(8088)
 250     assert "standby" in client.succeed("sudo -u yarn yarn rmadmin -getAllServiceState")
 251     client.succeed("sudo -u yarn systemd-cat yarn rmadmin -getAllServiceState")
 252
 253     assert "Estimated value of Pi is" in client.succeed("HADOOP_USER_NAME=hdfs yarn jar $(readlink $(which yarn) | sed -r 's~bin/yarn~share/hadoop/mapreduce/hadoop-mapreduce-examples-*.jar~g') pi 2 10")
 254     assert "SUCCEEDED" in client.succeed("yarn application -list -appStates FINISHED")
 255   '';
 256 })