Meson is a modern build system that has seen a rise in adoption and some
version of it is available in almost every platform supported.
Compared to automake, meson has the following advantages:
* Meson provides a significant boost to the build and configuration time
by better exploiting parallelism.
* Meson is subjectively considered to be better in readability.
These merits alone justify experimenting with meson as a way of
improving development time and ergonomics. However, there are some
compromises to ensure the transition goes relatively smooth:
* The system tests currently rely on various files within the source
directory. Changing this requirement is a non-trivial task that can't
be currently justified. Currently the last compiled build directory
writes into the source tree which is in turn used by pytest.
* The minimum version supported has been fixed at 0.61. Increasing this
value will require choosing a baseline of distributions that can
package with meson. On the contrary, there will likely be an attempt
to decrease this value to ensure almost universal support for building
BIND 9 with meson.
Instead of calling dst_lib_init() and dst_lib_destroy() explicitly by
all the programs, create a separate memory context for the DST subsystem
and use the library constructor and destructor to initialize the DST
internals.
To avoid any escaping issues or messing with a language-specific format
when the variable has to be parsed, create a dedicated file for each
variable that is obtained from autoconf.
Remove conf.sh.in and move the environment variables into isctest/vars
python package. This enabled the removal of an ugly pytest hack which
loaded and parsed these variables from the environment.
Pytest provides JUnit output and uses different exit codes from
Automake. Use the conversion script to interpret the JUnit test results
from python rather than relying on the status code.
The _common directory is a special case directory which contains shared
files for other system test directories. Make sure it's tracked in git
and not deleted during temporary directory cleanup.
While temporary directories are useful for test execution to keep
everything clean, they are difficult to work with manually. Create a
symlink for each test artifact directory with a stable and predictable
path. The symlink always either points to the latest artifacts, or is
missing in case the last run succeeded.
Ensure these symlinked directories aren't detected as test suites by the
pytest runner.
Utilize developers' muscle memory to incentivize using the pytest runner
instead of the legacy one. The script also serves as basic examples of
how to run the pyest command to achieve the same results as the legacy
runner.
Invoking pytest directly should be the end goal, since it offers many
potentially useful options (refer to pytest --help).
"parallel.sh" script was used on Windows to run system tests in
parallel. Since Windows support was removed from BIND 9, the script is
not needed anymore.
This is almost minimal prototype to show how to use python-hypothesis
library in a system test. It does not fully replace existing shell-based
system test for wildcards.
The pytest "cacheprovider" plugin produces a .cache/v/cache/lastfailed
file, which holds a Python dictionary structure with failed tests.
However, on Ubuntu 16.04 (Xenial) the file is created even though the
test passed and the file contains just an empty dictionary ("{}").
Given that we are not interested in this feature, disabling the
"cacheprovider" plugin globally and removing per-test removals of the
.cache directory seems like the best course of action.
This commit changes the taskmgr to run the individual tasks on the
netmgr internal workers. While an effort has been put into keeping the
taskmgr interface intact, couple of changes have been made:
* The taskmgr has no concept of universal privileged mode - rather the
tasks are either privileged or unprivileged (normal). The privileged
tasks are run as a first thing when the netmgr is unpaused. There
are now four different queues in in the netmgr:
1. priority queue - netievent on the priority queue are run even when
the taskmgr enter exclusive mode and netmgr is paused. This is
needed to properly start listening on the interfaces, free
resources and resume.
2. privileged task queue - only privileged tasks are queued here and
this is the first queue that gets processed when network manager
is unpaused using isc_nm_resume(). All netmgr workers need to
clean the privileged task queue before they all proceed normal
operation. Both task queues are processed when the workers are
finished.
3. task queue - only (traditional) task are scheduled here and this
queue along with privileged task queues are process when the
netmgr workers are finishing. This is needed to process the task
shutdown events.
4. normal queue - this is the queue with netmgr events, e.g. reading,
sending, callbacks and pretty much everything is processed here.
* The isc_taskmgr_create() now requires initialized netmgr (isc_nm_t)
object.
* The isc_nm_destroy() function now waits for indefinite time, but it
will print out the active objects when in tracing mode
(-DNETMGR_TRACE=1 and -DNETMGR_TRACE_VERBOSE=1), the netmgr has been
made a little bit more asynchronous and it might take longer time to
shutdown all the active networking connections.
* Previously, the isc_nm_stoplistening() was a synchronous operation.
This has been changed and the isc_nm_stoplistening() just schedules
the child sockets to stop listening and exits. This was needed to
prevent a deadlock as the the (traditional) tasks are now executed on
the netmgr threads.
* The socket selection logic in isc__nm_udp_send() was flawed, but
fortunatelly, it was broken, so we never hit the problem where we
created uvreq_t on a socket from nmhandle_t, but then a different
socket could be picked up and then we were trying to run the send
callback on a socket that had different threadid than currently
running.
The get_ports.sh script is used for determining the range of ports a
given system test should use. It first determines the start of the port
range to return (the base port); it can either be specified explicitly
by the caller or chosen randomly. Subsequent ports are picked
sequentially, starting from the base port. To ensure no single port is
used by multiple tests, a state file (get_ports.state) containing the
last assigned port is maintained by the script. Concurrent access to
the state file is protected by a lock file (get_ports.lock); if one
instance of the script holds the lock file while another instance tries
to acquire it, the latter retries its attempt to acquire the lock file
after sleeping for 1 second; this retry process can be repeated up to 10
times before the script returns an error.
There are some problems with this approach:
- the sleep period in case of failure to acquire the lock file is
fixed, which leads to a "thundering herd" type of problem, where
(depending on how processes are scheduled by the operating system)
multiple system tests try to acquire the lock file at the same time
and subsequently sleep for 1 second, only for the same situation to
likely happen the next time around,
- the lock file is being locked and then unlocked for every single
port assignment made, not just once for the entire range of ports a
system test should use; in other words, the lock file is currently
locked and unlocked 13 times per system test; this increases the
odds of the "thundering herd" problem described above preventing a
system test from getting one or more ports assigned before the
maximum retry count is reached (assuming multiple system tests are
run in parallel); it also enables the range of ports used by a given
system test to be non-sequential (which is a rather cosmetic issue,
but one that can make log interpretation harder than necessary when
test failures are diagnosed),
- both issues described above cause unnecessary delays when multiple
system tests are started in parallel (due to high lock file
contention among the system tests being started),
- maintaining a state file requires ensuring proper locking, which
complicates the script's source code.
Rework the get_ports.sh script so that it assigns non-overlapping port
ranges to its callers without using a state file or a lock file:
- add a new command line switch, "-t", which takes the name of the
system test to assign ports for,
- ensure every instance of get_ports.sh knows how many ports all
system tests which form the test suite are going to need in total
(based on the number of subdirectories found in bin/tests/system/),
- in order to ensure all instances of get_ports.sh work on the same
global port range (so that no port range collisions happen), a
stable (throughout the expected run time of a single system test
suite) base port selection method is used instead of the random one;
specifically, the base port, unless specified explicitly using the
"-p" command line switch, is derived from the number of hours which
passed since the Unix Epoch time,
- use the name of the system test to assign ports for (passed via the
new "-t" command line switch) as a unique index into the global
system test range, to ensure all system tests use disjoint port
ranges.
The system tests currently uses patchwork of shell scripts which doesn't
offer proper error handling.
This commit introduced option to write new tests in pytest framework
that also allows easier manipulation of DNS traffic (using dnspython),
native XML and JSON manipulation and proper error reporting.
The current script used ephemeral port range which clashed with the
ports used by the tools (dig, ...), and the range always started with
the first port and there was 100 ports allocated for each system test.
In this commit, the first port has been randomized, the get_ports.sh
script outputs the variables (the output has to be eval'ed from run.sh)
and there's less waste in the port range.
There are several improvements over the default/previous behaviour of
the test log driver and log compiler:
* The system-test-driver.sh was dropped (it was used incorrectly)
* The run.sh script is now both log compiler and cli script to run
individual tests
* The custom-test-driver was added as extended version of the automake
test-driver with capability to tee the test output to stdout when
`--verbose yes` is passed to it (you can use LOG_DRIVER_FLAGS to
add the option by default)
* Makefile.am has been extended to honor V=1 for the system tests
test-driver (e.g. V=1 adds `--verbose yes` to AM_LOG_DRIVER_FLAGS)
The rewrite of BIND 9 build system is a large work and cannot be reasonable
split into separate merge requests. Addition of the automake has a positive
effect on the readability and maintainability of the build system as it is more
declarative, it allows conditional and we are able to drop all of the custom
make code that BIND 9 developed over the years to overcome the deficiencies of
autoconf + custom Makefile.in files.
This squashed commit contains following changes:
- conversion (or rather fresh rewrite) of all Makefile.in files to Makefile.am
by using automake
- the libtool is now properly integrated with automake (the way we used it
was rather hackish as the only official way how to use libtool is via
automake
- the dynamic module loading was rewritten from a custom patchwork to libtool's
libltdl (which includes the patchwork to support module loading on different
systems internally)
- conversion of the unit test executor from kyua to automake parallel driver
- conversion of the system test executor from custom make/shell to automake
parallel driver
- The GSSAPI has been refactored, the custom SPNEGO on the basis that
all major KRB5/GSSAPI (mit-krb5, heimdal and Windows) implementations
support SPNEGO mechanism.
- The various defunct tests from bin/tests have been removed:
bin/tests/optional and bin/tests/pkcs11
- The text files generated from the MD files have been removed, the
MarkDown has been designed to be readable by both humans and computers
- The xsl header is now generated by a simple sed command instead of
perl helper
- The <irs/platform.h> header has been removed
- cleanups of configure.ac script to make it more simpler, addition of multiple
macros (there's still work to be done though)
- the tarball can now be prepared with `make dist`
- the system tests are partially able to run in oot build
Here's a list of unfinished work that needs to be completed in subsequent merge
requests:
- `make distcheck` doesn't yet work (because of system tests oot run is not yet
finished)
- documentation is not yet built, there's a different merge request with docbook
to sphinx-build rst conversion that needs to be rebased and adapted on top of
the automake
- msvc build is non functional yet and we need to decide whether we will just
cross-compile bind9 using mingw-w64 or fix the msvc build
- contributed dlz modules are not included neither in the autoconf nor automake
Some clean.sh scripts contain overly broad file deletion wildcards which
cause the test.output file (used by the system test framework for
collecting output) in a given system test's directory to be erroneously
removed immediately after the test is started (due to setup.sh scripts
calling clean.sh at the beginning). This prevents the test's output
from being placed in bin/tests/system/systests.output at the end of a
test suite run and thus can lead to test failures being ignored. Fix by
storing each test's output in a test.output.<test-name> file in
bin/tests/system/, which prevents clean.sh scripts from removing it (as
they should only ever affect files contained in a given system test's
directory).
When running all the system tests, output from a test is sent to a
test.output file in the test directory. These are combined in to
systests.output when the run finishes.
