Kea  2.1.2-git
Building Kea with Unit Tests

By default, Kea is built without unit-tests as they're used mostly by developers and prospective contributors.

Kea's unit-tests are using gtest framework from Google. Google's approach has changed over the years. For some time, they were very keen on not installing gtest as a normal software would be, but rather provide gtest as sources. This was further complicated with the fact that some Linux distributions packaged gtest and tried to mimic its installation. Kea tries its best to accommodate all typical situations and provides two switches to point to gtest. You can use --with-gtest or --with-gtest-source. Both attempt to locate gtest on their own. However, if neither of them can find it, you can specify the path explicitly. For example, on ubuntu with googletest package installed, you can do the following for Kea to find it:

sudo apt install googletest
./configure --with-gtest-source=/usr/src/googletest

Depending on how you compiled or installed gtest (e.g. from sources or using some package management system) one of those two switches will find gtest. After that you make and run the unit-tests with:

make
make check

As usual, using -jX option will speed up compilation. This parameter is even more useful for unit-tests as there are over 6000 unit-tests and their compilation is significantly slower than just the production Kea sources.

Kea should work with reasonably recent gtest versions. We recently tried with 1.7.0, 1.8.0, 1.8.1 and 1.10.0.

Environment Variables

The following environment variable can affect the unit tests:

  • KEA_LOCKFILE_DIR - Specifies a directory where the logging system should create its lock file. If not specified, it is prefix/var/run/kea, where prefix defaults to /usr/local. This variable must not end with a slash. There is one special value, "none", which instructs Kea to not create a lock file at all. This may cause issues if several processes log to the same file. (Also see the Kea User's Guide, section 15.3.)
  • KEA_LOGGER_DESTINATION - Specifies the logging destination. If not set, logged messages will not be recorded anywhere. There are three special values: stdout, stderr and syslog. Any other value is interpreted as a filename. (Also see Kea User's Guide, section 15.3.)
  • KEA_LOG_CHECK_VERBOSE - Specifies the log check default verbosity. If not set, unit tests using the log utils to verify that logs are generated as expected are by default silent. If set, these unit tests display real and expected logs.
  • KEA_PIDFILE_DIR - Specifies the directory which should be used for PID files as used by dhcp::Daemon or its derivatives. If not specified, the default is prefix/var/run/kea, where prefix defaults to /usr/local. This variable must not end with a slash.
  • KEA_SOCKET_TEST_DIR - If set, it specifies the directory where Unix sockets are created. There is an operating system limitation on how long a Unix socket path can be, typically slightly over 100 characters. By default unit-tests create sockets in temporary folder under /tmp folder. KEA_SOCKET_TEST_DIR can be specified to instruct unit-tests to use a different directory. It must not end with slash.
  • KEA_TEST_DB_WIPE_DATA_ONLY - Unit tests which use a Kea unit test database take steps to ensure they are starting with an empty database of the correct schema version. The first step taken is to simply delete the transient data (such as leases, reservations, etc..), provided the schema exists and is the expected version. If the schema does not exist, is not the expected version, or for some reason the data wipe fails, the schema will be dropped and recreated. Setting this value to "false" will cause the test setup logic to always drop and create the database schema. The default value is "true".
  • KEA_TLS_CHECK_VERBOSE - Specifies the TLS check default verbosity. If not set, TLS unit tests triggering expected failures are by default silent. If set, these TLS unit tests display the error messages which are very dependent on the cryptographic backend and boost library versions.
Note
Setting KEA_TEST_DB_WIPE_DATA_ONLY to false may dramatically increase the time it takes each unit test to execute.
  • GTEST_OUTPUT - Save the test results in XML files. Make it point to a location where a file or directory can be safely created. If there is no file or directory at that location, adding a trailing slash GTEST_OUTPUT=${PWD}/test-results/ will create a directory containing an XML file for each directory being tested. Leaving the slash out will create a single XML file and will put all the test results in it.
  • DEBUG - Set this variable to make shell tests output the commands that are run. They are shown just before they are effectively run. Can be set to anything e.g. DEBUG=true. unset DEBUG to remove this behavior.

Use Sanitizers

GCC and LLVM support some sanitizers which perform additional tests at runtime, for instance the ThreadSanitizer (aka TSan) detects data race in executed C++ code (unfortunately on macOS it intercepts signals and fails to send them to waiting select system calls so some tests always fail when it is used, experiments are run with different versions of Tsan).

The simplest way to enable a sanitizer is to add it to the CXXFLAGS environment variable in .configure by e.g. -fsanitize=thread.

When enabling lcov (code coverage), some gtest functions are detected as not being thread safe. It is recommended to disable lcov when enabling thread sanitizer.

Databases Configuration for Unit Tests

With the use of databases requiring separate authorisation, there are certain database-specific pre-requisites for successfully running the unit tests. These are listed in the following sections.

Database Users Required for Unit Tests

Unit tests validating database backends require that the keatest database is created. This database should be empty. The unit tests also require that the keatest user is created and that this user is configured to access the database with a password of keatest. Unit tests use these credentials to create database schema, run test cases and drop the schema. Thus, the keatest user must have sufficiently high privileges to create and drop tables, as well as insert and modify the data within those tables.

The database backends which support read only access to the host reservations databases (currently MySQL and PostgreSQL) include unit tests verifying that a database user with read-only privileges can be used to retrieve host reservations. Those tests require another user, keatest_readonly, with SQL SELECT privilege to the keatest database (i.e. without INSERT, UPDATE etc.), is also created. keatest_readonly should also have the password keatest.

The following sections provide step-by-step guidelines how to setup the databases for running unit tests.

MySQL Database

The steps to create the database and users are:

  1. Log into MySQL as root:
      % mysql -u root -p
      Enter password:
         :
      mysql>

  2. Create the test database. This must be called "keatest":
      mysql> CREATE DATABASE keatest;
      mysql>

  3. Create the users under which the test client will connect to the database (the apostrophes around the words keatest, keatest_readonly, and localhost are required):
      mysql> CREATE USER 'keatest'@'localhost' IDENTIFIED BY 'keatest';
      mysql> CREATE USER 'keatest_readonly'@'localhost' IDENTIFIED BY 'keatest';
      mysql>

  4. Grant the created users permissions to access the keatest database (again, the apostrophes around the user names and localhost are required):
      mysql> GRANT ALL ON keatest.* TO 'keatest'@'localhost';
      mysql> GRANT SELECT ON keatest.* TO 'keatest_readonly'@'localhost';
      mysql>

  5. If you get You do not have the SUPER privilege and binary logging is enabled error message, you need to add:
      mysql> SET GLOBAL LOG_BIN_TRUST_FUNCTION_CREATORS = 1;
      mysql>

  6. Exit MySQL:
      mysql> quit
      Bye
      %

The unit tests are run automatically when "make check" is executed (providing that Kea has been build with the –with-mysql switch (see the installation section in the Kea Administrator Reference Manual).

PostgreSQL Database

PostgreSQL set up differs from system to system. Please consult your operating system-specific PostgreSQL documentation. The remainder of that section uses Ubuntu 13.10 x64 (with PostgreSQL 9.0+) as an example.

On Ubuntu, PostgreSQL is installed (with sudo apt-get install postgresql) under user postgres. To create new databases or add new users, initial commands must be issued under this username:

$ sudo -u postgres psql postgres
[sudo] password for thomson:
psql (9.1.12)
Type "help" for help.
postgres=# CREATE USER keatest WITH PASSWORD 'keatest';
CREATE ROLE
postgres=# CREATE DATABASE keatest;
CREATE DATABASE
postgres=# GRANT ALL PRIVILEGES ON DATABASE keatest TO keatest;
GRANT
postgres=# \q

PostgreSQL versions earlier than 9.0 don't provide an SQL statement for granting privileges on all tables in a database. In newer PostgreSQL versions, it is possible to grant specific privileges on all tables within a schema. However, this only affects tables which exist when the privileges are granted. To ensure that the user has specific privileges to tables dynamically created by the unit tests, the default schema privileges must be altered.

The following example demonstrates how to create the user keatest_readonly, which has SELECT privilege to the tables within the keatest database, in Postgres 9.0+. For earlier versions of Postgres, it is recommended to simply grant full privileges to keatest_readonly, using the same steps as for the keatest user.

$ psql -U postgres
Password for user postgres:
psql (9.1.12)
Type "help" for help.

postgres=# CREATE USER keatest_readonly WITH PASSWORD 'keatest';
CREATE ROLE
postgres=# \q

$ psql -U keatest
Password for user keatest:
psql (9.1.12)
Type "help" for help.

keatest=> ALTER DEFAULT PRIVILEGES IN SCHEMA public GRANT SELECT ON TABLES to keatest_readonly;
ALTER DEFAULT PRIVILEGES
keatest=> \q

Note that the keatest user (rather than postgres) is used to grant privileges to the keatest_readonly user. This ensures that the SELECT privilege is granted only on the tables that the keatest user can access within the public schema.

It seems this no longer works on recent versions of PostgreSQL: if you get a permission problem on SELECT on the schema_version table for eatest_readonly, please try with the schema loaded:

$ psql -h localhost -U keatest -d keatest
Password for user keatest:
psql (11.3 (Debian 11.3-1))
SSL connection (protocol: TLSv1.3, cipher: TLS_AES_256_GCM_SHA384, bits: 256, compression: off)
Type "help" for help.

keatest=> GRANT SELECT ON ALL TABLES IN SCHEMA public TO keatest_readonly;
GRANT
keatest=> \q

Now we should be able to log into the newly created database using both user names:

$ psql -d keatest -U keatest
Password for user keatest:
psql (9.1.12)
Type "help" for help.

keatest=> \q

$ psql -d keatest -U keatest_readonly
Password for user keatest_readonly:
psql (9.1.12)
Type "help" for help.

keatest=>

If instead of seeing keatest=> prompt, your login is refused with an error code about failed peer or Ident authentication failed for user "keatest", it means that PostgreSQL is configured to check unix username and reject login attempts if PostgreSQL names are different. To alter that, the PostgreSQL pg_hba.conf configuration file must be changed. It usually resides at /var/lib/postgresql/data/pg_hba.conf or at /etc/postgresql/${version}/main/pg_hba.conf, but you can find out for sure by running sudo -u postgres psql -t -c 'SHOW hba_file'. Make sure that all the authentication methods are changed to "md5" like this:

local   all             all                                     md5
host    all             all             127.0.0.1/32            md5
host    all             all             ::1/128                 md5

Another possible problem is that you get no password prompt. This is most probably because you have no pg_hba.conf config file and everybody is by default trusted. As it has a very bad effect on the security you should have been warned this is a highly unsafe configuration. The solution is the same, i.e., require password or md5 authentication method.

If you lose the postgres user access you can first add:

local   all             postgres                                trust

to trust only the local postgres user. Note the postgres user can be pgsql on some systems.

Please consult your PostgreSQL user manual before applying those changes as those changes may expose your other databases that you run on the same system. In general case, it is a poor idea to run anything of value on a system that runs tests. Use caution!

The unit tests are run automatically when "make check" is executed (providing that Kea has been build with the –with-pgsql switch (see the installation section in the Kea Administrator Reference Manual).

Cassandra database

Todo:
: Describe steps necessary to set up Cassandra database suitable for running unittests.

It seems this was enough:

  1. Launch cassandra if not running (-f for foreground)
     % cassandra -f

The tool is cqlsh:

  1. Run the tool
     % cqlsh
     Connected to Test Cluster at 127.0.0.1:9042.
     [cqlsh 5.0.1 | Cassandra 3.11.1 | CQL spec 3.4.4 | Native protocol v4]
     Use HELP for help.
     cqlsh> 

Kerberos Configuration for Unit Tests

The GSS-TSIG hook library uses the GSS-API with Kerberos. While there are no doubts that the hook can be safely used with a valid Kerberos configuration in production, unit tests reported problems on some systems.

GSS-TSIG hook unit tests use a setup inherited from bind9 with old crypto settings which are not allowed by default Kerberos system configuration. A simple workaround is to set the KRB5_CONFIG environment variable to a random value that doesn't match a file (e.g. KRB5_CONFIG=).

Writing shell scripts and tests

Shell tests are shellchecked. But there are other writing practices that are good to follow in order to keep, not only shell tests, but shell scripts in general, POSIX-compliant. See below:

  • For portability, all shell scripts should have a shebang.
    #!/bin/sh
    The sh shell can differ on various operating systems. On most systems it is GNU sh. Notable exceptions are Alpine which links it to ash, FreeBSD which has the primordial non-GNU sh, Ubuntu which links it to dash. These four shells should all be tested against, when adding shell scripts or making changes to them.
  • Reference variables with curly brackets.
    ${var} # better
    $var
    For consistency with cases where you need advanced features from the variables which make the curly brackets mandatory. Such cases are:
    # Retrieving variable/string length...
    ${#var}
    # Defaulting to a given value when the variable is undefined...
    ${var-default}
    # Substituting the variable with a given value when the variable is defined...
    ${var+value}
    # Concatenating the value of a variable with an alphanumeric constant...
    ${var}constant
  • Always use printf instead of echo. There are times when a newline is not desired such as when you want to print on a single line from multiple points in your script or when you want to get the character count from an expression:
    var1='not '
    var2=' you want to ignore'
    # Prints the number of characters.
    printf '%s' "${var1}something${var2}" | wc -c
    # Result:
    32
    # This one prints a plus one i.e. the inherent newline.
    echo "${var1}something${var2}" | wc -c
    # Result:
    33
    # `echo` does have `-n` to suppress newline, but...
    # SC2039: In POSIX sh, echo flags are undefined.
    echo -n "${var1}something${var2}" | wc -c
    # Result:
    32 # sometimes, other times an error
    printf also has the benefit of separating the format from the actual variables which has many use cases. One such use case is coloring output with ANSI escape sequence codes, see the test_finish function in src/lib/testutils/dhcp_test_lib.sh.in, which is not possible with POSIX echo.
  • set -e should be enabled at all times to immediately fail when a command returns a non-zero exit code. There are times when you expect a non-zero exit code in your tests. This is what the run_command function in src/lib/testutils/dhcp_test_lib.sh.in is for. It momentarily disables the -e flag to capture the output and exit code and enables it again afterwards. The variables used are ${EXIT_CODE} and ${OUTPUT}. /dev/stderr is not captured. run_command also doesn't work with pipes and redirections. When these mechanisms are needed, you can always wrap your complex expression in a function and then call run_command wrapping_function. Alternatively, if you only care about checking for zero exit code, you can use if conditions.
    # The non-zero exit code does not stop script execution, but we can still adjust
    # behavior based on it.
    if maybe-failing-command; then
    f
    else
    g
    fi
    There are times when your piped or redirected command that is expected to return non-zero is so small or has so few instantiations that it doesn't deserve a separate function. Such an example could be grepping for something in a variable. grep returns a non-zero exit code if it doesn't find anything. In that case, you can add || true at the end to signal the fact that you allow finding nothing like so:
    printf '%s' "${var}" | grep -F 'search-criterion' || true
  • set -u should be enabled at all times to immediately signal an undefined variable. If you're a stickler for the legacy behavior of defaulting to an empty space then you can reference all your variables with:
    # Default variable is an empty space.
    ${var-}
    # Or like this if you prefer to quote the empty space.
    ${var-''}
  • SC2086: Double quote to prevent globbing and word splitting. Even though covered by shellcheck, it's worth mentioning because shellcheck doesn't always warn you because of what might be a systematic deduction of when quoting is not needed. Globbing is a pattern matching mechanism. It's used a lot with the * wildcard character e.g. ls *.txt. Sometimes, you want to glob intentionally. In that case, you can omit quoting, but it is preferable to take the wildcard characters outside the variable so that you are able to quote to prevent other globbing and word splitting e.g.:
    # Globbing done right
    ls "${var}"*.txt
    # Word splitting problem
    path='/home/my user'
    ls ${path}
    # Result:
    ls: cannot access '/home/my': No such file or directory
    ls: cannot access 'user': No such file or directory
    # Word splitting avoided
    path='/home/my user'
    ls "${path}"
    # Result:
    Desktop
    Documents
    Downloads
    If you have an expression composed of multiple variables don't just quote the variables. It's correct, but not readable. Quote the entire expression.
    # no
    "${var1}"some-fixed-contiguous-value"${var2}"
    # yes
    "${var1}some-fixed-contiguous-value${var2}"
  • Single quote expressions when no variables are inside. This is to avoid the need to escape special shell characters like $.
  • All shell tests are created from .in autoconf template files. They initially contain template variables like @prefix@ which are then substituted with the configured values. All of these should be double quoted, not single-quoted since they themselves can contain shell variables that need to be expanded.
  • Use $(...) notation instead of legacy backticks. One important advantage is that the $(...) notation allows for nested executions.
    # SC2006 Use `$(...)` notation instead of legacy backticked `...`.
    hostname=`cat /etc/hostname`
    # Better
    hostname=$(cat /etc/hostname)
    # Nested executions
    is_ssh_open=$(nc -vz $(cat /etc/hostname).lab.isc.org 22)
    # Results in confusing "command not found" messages.
    is_ssh_open=`nc -vz `cat /etc/hostname`.lab.isc.org 22`
  • When using test and [, == is just a convenience alias for =. Use = because it's more widely supported. If using, [[, then indeed == has extra features like glob matching. But don't use [[, it's not part of the POSIX standard.
  • Capturing parameters in functions or scripts simply cannot be done without breaking POSIX compliance. In POSIX, pass the quoted parameters "${@}" as positional parameters to all the function and scripts invocations. if it gets too unmanageable or you need custom positional arguments then break your script into multiple scripts or handle all possible parameters and don't accept any ad-hoc parameters.
    # Neither of these preserve original quoting.
    parameters="${*}"
    parameters="${@}"
    # In advanced shells this could be done with lists.
    parameters=( "${@}" )
    do-something --some --other --optional --parameters "${parameters[@]}"
    # Proper POSIX way
    do-something --some --other --optional --parameters "${@}"
  • Never use eval. It doesn't preserve original quoting. Have faith that there are always good alternatives.