Skip to content

Latest commit

 

History

History
216 lines (159 loc) · 10.2 KB

INTEGRATION-TESTS.md

File metadata and controls

216 lines (159 loc) · 10.2 KB
Raw

webhdfs integration tests

The following methods are supported for integration testing:

  1. Remote: use a remote Hadoop cluster via ssh (PROVISIONER=remote)
  2. Local: use an Apache Bigtop cluster locally deployed in docker (PROVISIONER=docker) or vagrant (PROVISIONER=vagrant)

The test process itself runs on the host. See comments in itt.sh and itt.sh --help for details.

DISCLAIMER: the integration tests were developed on a Windows 10 machine (using Docker Desktop and both Cygwin and WSL). While the best effort was made to ensure Linux and OS X compatibility, the test scripts have not been tested there (yet), and may contain bugs. Sholud you notice any bugs and incompatibilies, please create an issue in the webhdfs Github project.

Remote method

This the default and the recommended method. It is inherently agnostic to Hadoop version being tested under.

You will need a remote Hadoop cluster available over ssh. SSH tunelling is used for port forwarding, and scp is used to bring local files to the remote system and vice versa.

To run the script under Windows, you need WSL installed on the local computer. In addition, ssh must be installed and properly configured (see below).

In the local configuration file, add the following settings (mandatory unless otherwise specified):

#Remote namenode 1 host:port
R_NN1=nn1:50070

#Remote namenode 2 host:port (optional)
R_NN2=nn2:50070

#Remote host:ports of datanodes
R_DN=(dn0001:50075 dn0002:50075 dn0003:50075) 

#SSH connection string, format: user@host[:port] 
[email protected]

#Adidional SSH arguments (jump hosts, etc) (optional)
SSH_ARGS="-J [email protected]:22"

#Delegation token (optional unless Kerberos authentication is enabled on the cluster)
DTOKEN="XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX"

#Hadoop user name
HADOOP_USER=huser

#remote intermediate directory (optional). This defaults to /test-data, so you would likely want to change that
C_TESTDATA_DIR=/tmp/test-data

In order for the test to succeed, ensure the following:

  1. Any necessary public and private keys are configured so there is no need for ssh password throughout the test script
  2. ssh tunneling is enabled on the server
  3. hdfs command is available for the ssh user at the ssh target machine, and is directly executable (i.e. available in the PATH)
  4. The local filesystem of the ssh target host has free space at least three times the size of the test file.
  5. Namenode/datanode hostnames are resolvabe at the remote end (of course, IP addresses can be used in place of hostnames)

If Kerberos authentication is enabled on the remote cluster:

  1. The ssh user that logs on the remote cluster must have a valid Kerberos TGT issued to the username specified as HADOOP_USER (via e.g. kinit huser).
  2. A valid delegation token must be issued and specified as DTOKEN. The following command sequence may be used to obtain the token:
#Fetch the token into `dt`
hadoop dtutil get hdfs://$CLUSTERNAME dt
#or 
hdfs fetchdt dt
#Then print it. 
hadoop dtutil print dt
#Paste `URL enc token` as the value of DTOKEN

Local methods

Local methods are set up with Apache Bigtop provisioner vagrant (typically 1-node) or docker (typically 3-node) cluster.

Shared folders is used for file exchange, and port mappings are used for port forwarding.

Preparation

  1. Install prerequisites
  • either docker or vagrant
  • For docker, also install ruby.
  • For vagrant, you will need Virtualbox (other providers are untested).
  • Under Windows you will also need Cygwin or WSL. Cygwin is recommended if Docker is used with Bigtop provisioner (because of docker to windows console interop issues), while WSL is recommended for Vagrant.
  1. Check out Bigtop. Bigtop Version 1.3 was used (rel/1.3 tag). Bigtop 1.3 has Hadoop version 2.8.4.
  2. Create a file named itt-config.sh in the webhdfs source root, and make it chmod +x. This is the test configuration script.
  3. Set BIGTOP_ROOT var to the root path of the cloned Bigtop repo. It must be set in the test config script.
  4. Edit Bigtop provisioner configuration to make guest webhdfs ports 50070 and 50075 accessible from the host, and guest folder /test-data mounted to the folder test-data under webhdfs source root. See Sharing ports and folders below for instructions.
  5. Using Bigtop tools, bring the cluster up and make sure the shared folder is mounted and the ports are accessble from the host.
  6. The test script uses Vagrant by default. To use docker, specify PROVISIONER=docker in the test config script.

NOTES:

  1. There is a number of other settings that may be specified in itt-config.sh. See itt.sh for details.
  2. If you use the number of VMs or containers other than default (1 for vagrant and 3 for docker), the exact number of containers must be specified in N_C setting in the test config script, otherwise the test might fail.

Sharing ports and folders: Docker

Go to the provisioner/docker folder under Bigtop source root. Edit docker-container.yaml to add the following (edit the absolute volume part):

   ports:
   - "50070"
   - "50075"
   volumes:
   - //c/path/to/test-data : /test-data

Sharing ports and folders: Vagrant

  1. Edit bigtop's Vagrantfile in the provisioner/vagrant folder under Bigtop source root and add the following in the VM configuration loop:
      #=== Intergation test environment settings ========================================================
      if i == 1 then
        config.vm.network "forwarded_port", guest: 50070, guest_ip: "10.10.10.11", host: 51070
        bigtop.vm.synced_folder "C:/devel/src/webhdfs-rs/test-data", "/test-data"
      end
      config.vm.network "forwarded_port", guest: 50075, host: 50075 + i * 1000
      #=== END: Intergation test environment settings ===================================================

Sample diff:

--- Vagrantfile.orig    2019-07-22 17:02:05.929272200 +0300
+++ Vagrantfile 2019-09-14 14:47:58.212960600 +0300
@@ -118,6 +118,14 @@
       # two levels up is the bigtop "home" directory.
       # the current directory has puppet recipes which we need for provisioning.
       bigtop.vm.synced_folder "../../", "/bigtop-home"
+
+      #=== Intergation test environment settings ========================================================
+      if i == 1 then
+        config.vm.network "forwarded_port", guest: 50070, guest_ip: "10.10.10.11", host: 51070
+        bigtop.vm.synced_folder "C:/devel/src/webhdfs-rs/test-data", "/test-data"
+      end
+      config.vm.network "forwarded_port", guest: 50075, host: 50075 + i * 1000
+      #=== END: Intergation test environment settings ===================================================

       # We also add the bigtop-home output/ dir, so that locally built rpms will be available.
       puts "Adding output/ repo ? #{enable_local_repo}"

How the test works

Read test

The read test consists of a sequence of reads and seeks against a testfile, set up by READSCRIPT setting. The social networking graph data file soc-pokec-relationships.txt, available from SNAP, is used as a testflle (any file, large enough, may be used; recommended size is 200-400M). During preparation, SHA-512 checksums are pre-calculated for each of r: chunks set up by READSCRIPT (chunks are extracted by dd). The program under test is expected to execute the READSCRIPT below (actually, READSCRIPTOUT) by doing seeks and reads as requested. Upon each read, the program reads from the testfile (HDFS) then writes the content read to a file specified by 3rd part of READSCRIPTOUT item. Finally, checksums for newly written chunks are validated against pre-calculated checksums.

Hooks

  1. place all local settings in ./itt-config.sh and make it chmod a+x
  2. if using a test file other than standard, or using the standard one but pre-downloaded, or using other source, place the test file materialzation command(s) in ./test-data./create-source-script and make it chmod a+x. Note that the script is launched in ./test-data. Note that if the test file is already in ./test-data, it is used as-is and left intact (if downloaded, the test file is deleted from ./test-data after preparation).

Running the test

  1. Make sure the cluster is up and running. For local methods, bring the Bigtop cluster up.
  2. Remote method only: in a separate terminal, run ./itt.sh --c-ssh. This establishes the ssh link to the remote system, enabling port forwarding via ssh tunnels. The link should be up during the entire test.
  3. Run itt.sh --run.
  4. Look for the following terminating line in the output from itt.sh --run:
==================== TEST SUCCESSFUL ====================

Possible problems and workarounds

  1. Note that both Bigtop provisioner variants seemingly do not support VM/container reboot. If a cluster's VM is powered down or a container terminated, the cluster shall be entirely deleted and re-created from scratch (yet a Vagrant VM can be suspended).

  2. 'IOException: Failed to replace a bad datanode on the existing pipeline due to no more good datanodes being available to try' errors during write test: Add the following to /etc/hadoop/conf/hdfs-site.xml and restart the datanode services(s):

<property>
  <name>dfs.client.block.write.replace-datanode-on-failure.policy</name>
  <value>NEVER</value>
</property>
  1. If port forwarding does not work (Vagrant): this might be consequences of VM's reboot. The following are possible temporary workarounds.

a) a possible cause is iptables in the VM. Check if it is active; if yes, execute

sudo iptables -A IN_public_allow -p tcp --dport 50070 -j ACCEPT
sudo iptables -A IN_public_allow -p tcp --dport 50075 -j ACCEPT

b) (NN only) the other possible cause is NN binding address mismatch. Check what IP is the NN bound to and specify that address in port forwarding settings.

c) Look at /etc/hosts, should look as the following. In particular, remove 10.0.. entries from the NAT adapter or give this IP a name other than hostname.

::1         localhost localhost.localdomain localhost6 localhost6.localdomain6

## vagrant-hostmanager-start
10.10.10.11     bigtop1.vagrant
10.10.10.11     bigtop1.vagrant bigtop1
## vagrant-hostmanager-end