{[project]} Contributing Guidelines.use English; $UID--no-log-timestampcontribpgbackrest-dev{[host-user]}pgbackrest/doc:contrib{[github-url-main]}/test/src{[github-url-main]}/test/src/common{[github-url-main]}/src{[github-url-src]}/commonuse Cwd qw(cwd); cwd()
{[copy-ca-cert]}
# Fix root tty
RUN sed -i 's/^mesg n/tty -s \&\& mesg n/g' /root/.profile && \
# Suppress dpkg interactive output
rm /etc/apt/apt.conf.d/70debconf
# Install base packages
RUN apt-get update && \
apt-get install -y sudo ssh curl vim 2>&1
# Add test user with sudo privileges
RUN adduser --disabled-password --uid={[host-user-id]} --gecos "" {[host-user]} && \
echo '%{[host-user]} ALL=(ALL) NOPASSWD: ALL' >> /etc/sudoers
WORKDIR /home/{[host-user]}
ENTRYPOINT service ssh restart && bash
Table of Contents
Introduction
Building a Development Environment
Coding
Testing
Submitting a Pull Request
Introduction
This documentation is intended to assist contributors to by outlining some basic steps and guidelines for contributing to the project.
Code fixes or new features can be submitted via pull requests. Ideas for new features and improvements to existing functionality or documentation can be submitted as issues. You may want to check the Project Boards to see if your suggestion has already been submitted.
Bug reports should be submitted as issues. Please provide as much information as possible to aid in determining the cause of the problem.
You will always receive credit in the release notes for your contributions.
Coding standards are defined in CODING.md and some important coding details and an example are provided in the Coding section below. At a minimum, unit tests must be written and run and the documentation generated before submitting a Pull Request; see the Testing section below for details.
Building a Development Environment
This example is based on Ubuntu 20.04, but it should work on many versions of Debian and Ubuntu.
Some unit tests and all the integration tests require Docker. Running in containers allows us to simulate multiple hosts, test on different distributions and versions of , and use sudo without affecting the host system.
If using a RHEL-based system, the CPAN XML parser is required to run test.pl and doc.pl. Instructions for installing Docker and the XML parser can be found in the README.md file of the doc directory in the section The following is a sample RHEL 7 configuration that can be used for building the documentation. NOTE that the Install latex (for building PDF) section is not required since testing of the docs need only be run for HTML output.
Coding
The following sections provide information on some important concepts needed for coding within .
Memory Contexts
Memory is allocated inside contexts and can be long lasting (for objects) or temporary (for functions). In general, use OBJ_NEW_BEGIN(MyObj) for objects and MEM_CONTEXT_TEMP_BEGIN() for functions. See memContext.h for more details and the Coding Example below.
Logging
Logging is used for debugging with the built-in macros FUNCTION_LOG_*() and FUNCTION_TEST_*() which are used to trace parameters passed to/returned from functions. FUNCTION_LOG_*() macros are used for production logging whereas FUNCTION_TEST_*() macros will be compiled out of production code. For functions where no parameter is valuable enough to justify the cost of debugging in production, use FUNCTION_TEST_BEGIN()/FUNCTION_TEST_END(), else use FUNCTION_LOG_BEGIN(someLogLevel)/FUNCTION_LOG_END(). See debug.h for more details and the Coding Example below.
Logging is also used for providing information to the user via the LOG_*() macros, such as LOG_INFO("some informational message") and LOG_WARN_FMT("no prior backup exists, %s backup has been changed to full", strZ(cfgOptionDisplay(cfgOptType))) and also via THROW_*() macros for throwing an error. See log.h and error.h for more details and the Coding Example below.
Coding Example
The example below is not structured like an actual implementation and is intended only to provide an understanding of some of the more common coding practices. The comments in the example are only here to explain the example and are not representative of the coding standards. Refer to the Coding Standards document (CODING.md) and sections above for an introduction to the concepts provided here. For an actual implementation, see db.h and db.c.
Example: hypothetical basic object construction
/*
* HEADER FILE - see db.h for a complete implementation example
*/
// Typedef the object declared in the C file
typedef struct MyObj MyObj;
// Constructor, and any functions in the header file, are all declared on one line
MyObj *myObjNew(unsigned int myData, const String *secretName);
// Declare the publicly accessible variables in a structure with Pub appended to the name
typedef struct MyObjPub // First letter upper case
{
unsigned int myData; // Contents of the myData variable
} MyObjPub;
// Declare getters and setters inline for the publicly visible variables
// Only setters require "Set" appended to the name
__attribute__((always_inline)) static inline unsigned int
myObjMyData(const MyObj *const this)
{
return THIS_PUB(MyObj)->myData; // Use the built-in THIS_PUB macro
}
// Destructor
__attribute__((always_inline)) static inline void
myObjFree(MyObj *const this)
{
objFree(this);
}
// TYPE and FORMAT macros for function logging
#define FUNCTION_LOG_MY_OBJ_TYPE \
MyObj *
#define FUNCTION_LOG_MY_OBJ_FORMAT(value, buffer, bufferSize) \
FUNCTION_LOG_STRING_OBJECT_FORMAT(value, myObjToLog, buffer, bufferSize)
/*
* C FILE - see db.c for a more complete and actual implementation example
*/
// Declare the object type
struct MyObj
{
MyObjPub pub; // Publicly accessible variables must be first and named "pub"
const String *name; // Pointer to lightweight string object - see string.h
};
// Object constructor, and any functions in the C file, have the return type and function signature on separate lines
MyObj *
myObjNew(unsigned int myData, const String *secretName)
{
FUNCTION_LOG_BEGIN(logLevelDebug); // Use FUNCTION_LOG_BEGIN with a log level for displaying in production
FUNCTION_LOG_PARAM(UINT, myData); // When log level is debug, myData variable will be logged
FUNCTION_TEST_PARAM(STRING, secretName); // FUNCTION_TEST_PARAM will not display secretName value in production logging
FUNCTION_LOG_END();
ASSERT(secretName != NULL || myData > 0); // Development-only assertions (will be compiled out of production code)
MyObj *this = NULL; // Declare the object in the parent memory context: it will live only as long as the parent
OBJ_NEW_BEGIN(MyObj) // Create a long lasting memory context with the name of the object
{
this = OBJ_NEW_ALLOC(); // Allocate the memory required by the object
*this = (MyObj) // Initialize the object
{
.pub =
{
.myData = myData, // Copy the simple data type to this object
},
.name = strDup(secretName), // Duplicate the String data type to the this object's memory context
};
}
OBJ_NEW_END();
FUNCTION_LOG_RETURN(MyObj, this);
}
// Function using temporary memory context
String *
myObjDisplay(unsigned int myData)
{
FUNCTION_TEST_BEGIN(); // No parameters passed to this function will be logged in production
FUNCTION_TEST_PARAM(UINT, myData);
FUNCTION_TEST_END();
String *result = NULL; // Result is created in the caller's memory context (referred to as "prior context" below)
MEM_CONTEXT_TEMP_BEGIN() // Begin a new temporary context
{
String *resultStr = strNewZ("Hello"); // Allocate a string in the temporary memory context
if (myData > 1)
resultStr = strCatZ(" World"); // Append a value to the string still in the temporary memory context
else
LOG_WARN("Am I not your World?"); // Log a warning to the user
MEM_CONTEXT_PRIOR_BEGIN() // Switch to the prior context so the string duplication is in the caller's context
{
result = strDup(resultStr); // Create a copy of the string in the caller's context
}
MEM_CONTEXT_PRIOR_END(); // Switch back to the temporary context
}
MEM_CONTEXT_TEMP_END(); // Free everything created inside this temporary memory context - i.e resultStr
FUNCTION_TEST_RETURN(STRING, result); // Return result but do not log the value in production
}
// Create the logging function for displaying important information from the object
String *
myObjToLog(const MyObj *this)
{
return strNewFmt(
"{name: %s, myData: %u}", this->name == NULL ? NULL_Z : strZ(this->name), myObjMyData(this));
}
Testing
A list of all possible test combinations can be viewed by running:
pgbackrest/test/test.pl --dry-run
While some files are automatically generated during make, others are generated by running the test harness as follows:
pgbackrest/test/test.pl --gen-only
Prior to any submission, the html version of the documentation should also be run and the output checked by viewing the generated html on the local file system under pgbackrest/doc/output/html. More details can be found in the doc/README.md file.
pgbackrest/doc/doc.pl --out=html
ERROR: [028] regarding cache is invalid is OK; it just means there have been changes and the documentation will be built from scratch. In this case, be patient as the build could take 20 minutes or more depending on your system.Running Tests
Examples of test runs are provided in the following sections. There are several important options for running a test:
--dry-run - without any other options, this will list all the available tests--module - identifies the module in which the test is located--test - the actual test set to be run--run - a number identifying the run within a test if testing a single run rather than the entire test--vm-out - displays the test output (helpful for monitoring the progress)--vm - identifies the pre-built container when using Docker, otherwise the setting should be none. See test.yml for a list of valid vm codes noted by param: test.
For more options, run the test or documentation engine with the --help option:
pgbackrest/test/test.pl --help
pgbackrest/doc/doc.pl --help
Without Docker
If Docker is not installed, then the available tests can be listed using --vm=none, and each test must then be run with --vm=none.
List tests that don't require a containerpgbackrest/test/test.pl --vm=none --dry-run{[test-cmd-extra]}[0-9]+ tests selected|DRY RUN COMPLETED SUCCESSFULLYRun a testpgbackrest/test/test.pl --vm=none --vm-out --module=common --test=wait{[test-cmd-extra]}
An entire module can be run by using only the --module option.
Run a modulepgbackrest/test/test.pl --vm=none --module=postgres{[test-cmd-extra]}With Docker
Build a container to run tests. The vm must be pre-configured but a variety are available. A vagrant file is provided in the test directory as an example of running in a virtual environment. The vm names are all three character abbreviations, e.g. u20 for Ubuntu 20.04.
Build a VMpgbackrest/test/test.pl --vm-build --vm=u20{[test-cmd-extra]}to build all the vms, just omit the --vm option above.Run a Specific Test Runpgbackrest/test/test.pl {[dash]}-vm=u20 {[dash]}-module=mock {[dash]}-test=archive {[dash]}-run=2{[test-cmd-extra]}Writing a Unit Test
The goal of unit testing is to have 100 percent code coverage. Two files will usually be involved in this process:
define.yaml - defines the number of tests to be run for each module and test file. There is a comment at the top of the file that provides more information about this file.src/module/somefileTest.c - where somefile is the path and name of the test file where the unit tests are located for the code being updated (e.g. src/module/command/expireTest.c).define.yaml
Each module is separated by a line of asterisks (*) and each test within is separated by a line of dashes (-). In the example below, the module is command and the unit test is check. The number of calls to testBegin() in a unit test file will dictate the number following total:, in this case 4. Under coverage:, the list of files that will be tested.
Unit test files are organized in the test/src/module directory with the same directory structure as the source code being tested. For example, if new code is added to src/command/expire.c then test/src/module/command/expireTest.c will need to be updated.
Assuming that a test file already exists, new unit tests will either go in a new testBegin() section or be added to an existing section. Each such section is a test run. The comment string passed to testBegin() should reflect the function(s) being tested in the test run. Tests within a run should use TEST_TITLE() with a comment string describing the test.
// *****************************************************************************************************************************
if (testBegin("expireBackup()"))
{
// -------------------------------------------------------------------------------------------------------------------------
TEST_TITLE("manifest file removal");
Setting up the command to be run
The harnessConfig.h describes a list of functions that should be used when configuration options are required for a command being tested. Options are set in a StringList which must be defined and passed to the HRN_CFG_LOAD() macro with the command. For example, the following will set up a test to run pgbackrest --repo-path=test/test-0/repo info command on multiple repositories, one of which is encrypted:
StringList *argList = strLstNew(); // Create an empty string list
hrnCfgArgRawZ(argList, cfgOptRepoPath, TEST_PATH "/repo"); // Add the --repo-path option
hrnCfgArgKeyRawZ(argList, cfgOptRepoPath, 2, TEST_PATH "/repo2"); // Add the --repo2-path option
hrnCfgArgKeyRawStrId(argList, cfgOptRepoCipherType, 2, cipherTypeAes256Cbc); // Add the --repo2-cipher-type option
hrnCfgEnvKeyRawZ(cfgOptRepoCipherPass, 2, TEST_CIPHER_PASS); // Set environment variable for the --repo2-cipher-pass option
HRN_CFG_LOAD(cfgCmdInfo, argList); // Load the command and option list into the test harness
Storing a file
Sometimes it is desirable to store or manipulate files before or during a test and then confirm the contents. The harnessStorage.h file contains macros (e.g. HRN_STORAGE_PUT and TEST_STORAGE_GET) for doing this. In addition, HRN_INFO_PUT is convenient for writing out info files (archive.info, backup.info, backup.manifest) since it will automatically add header and checksum information.
Tests are run and results confirmed via macros that are described in harnessTest.h. With the exception of TEST_ERROR, the third parameter is a short description of the test. Some of the more common macros are:
TEST_RESULT_STR - Test the actual value of the string returned by the function.TEST_RESULT_UINT / TEST_RESULT_INT - Test for an unsigned integer / integer.TEST_RESULT_BOOL - Test a boolean value.TEST_RESULT_PTR / TEST_RESULT_PTR_NE - Test a pointer: useful for testing if the pointer is NULL or not equal (NE) to NULL.TEST_RESULT_VOID - The function being tested returns a void. This is then usually followed by tests that ensure other actions occurred (e.g. a file was written to disk).TEST_ERROR / TEST_ERROR_FMT - Test that a specific error code was raised with specific wording.HRN_* macros should be used only for test setup and cleanup. TEST_* macros must be used for testing results.Testing a log message
If a function being tested logs something with LOG_WARN, LOG_INFO or other LOG_*() macro, then the logged message must be cleared before the end of the test by using the TEST_RESULT_LOG()/TEST_RESULT_LOG_FMT() macros.
TEST_RESULT_LOG(
"P00 WARN: WAL segment '000000010000000100000001' was not pushed due to error [25] and was manually skipped: error");
In the above, Pxx indicates the process (P) and the process number (xx), e.g. P00, P01.
Testing using child process
Sometimes it is useful to use a child process for testing. Below is a simple example. See harnessFork.h for more details.
HRN_FORK_BEGIN()
{
HRN_FORK_CHILD_BEGIN()
{
TEST_RESULT_INT_NE(
lockAcquire(cfgOptionStr(cfgOptLockPath), STRDEF("stanza1"), STRDEF("999-ffffffff"), lockTypeBackup, 0, true),
-1, "create backup/expire lock");
// Notify parent that lock has been acquired
HRN_FORK_CHILD_NOTIFY_PUT();
// Wait for parent to allow release lock
HRN_FORK_CHILD_NOTIFY_GET();
lockRelease(true);
}
HRN_FORK_CHILD_END();
HRN_FORK_PARENT_BEGIN()
{
// Wait for child to acquire lock
HRN_FORK_PARENT_NOTIFY_GET(0);
HRN_CFG_LOAD(cfgCmdInfo, argListText);
TEST_RESULT_STR_Z(
infoRender(),
"stanza: stanza1\n"
" status: error (no valid backups, backup/expire running)\n"
" cipher: none\n"
"\n"
" db (current)\n"
" wal archive min/max (9.4): none present\n",
"text - single stanza, no valid backups, backup/expire lock detected");
// Notify child to release lock
HRN_FORK_PARENT_NOTIFY_PUT(0);
}
HRN_FORK_PARENT_END();
}
HRN_FORK_END();
Testing using a shim
A libpq shim is provided to simulate interactions with . Below is a simple example. See harnessPq.h for more details.
// Set up two standbys but no primary
harnessPqScriptSet((HarnessPq [])
{
HRNPQ_MACRO_OPEN_GE_92(1, "dbname='postgres' port=5432", PG_VERSION_92, "/pgdata", true, NULL, NULL),
HRNPQ_MACRO_OPEN_GE_92(8, "dbname='postgres' port=5433", PG_VERSION_92, "/pgdata", true, NULL, NULL),
// Close the "inner" session first (8) then the outer (1)
HRNPQ_MACRO_CLOSE(8),
HRNPQ_MACRO_CLOSE(1),
HRNPQ_MACRO_DONE()
});
TEST_ERROR(cmdCheck(), ConfigError, "primary database not found\nHINT: check indexed pg-path/pg-host configurations");
Running a Unit Test
Code Coverage
Unit tests are run for all files that are listed in define.yaml and a coverage report generated for each file listed under the tag coverage:. Note that some files are listed in multiple coverage: sections for a module; in this case, each test for the file being modified should be specified for the module in which the file exists (e.g. --module=storage --test=posix --test=gcs, etc.) or, alternatively, simply run the module without the --test option. It is recommended that a --vm be specified since running the same test for multiple vms is unnecessary for coverage. The following example would run the test set from the define.yaml section detailed above.
pgbackrest/test/test.pl --vm-out --module=command --test=check --vm=u20
Not all systems perform at the same speed, so if a test is timing out, try rerunning with another vm.
Because a test run has not been specified, a coverage report will be generated and written to the local file system under the directory test/result/coverage/lcov/index.html and a file with only the highlighted code that has not been covered will be written to test/result/coverage/coverage.html.
If 100 percent code coverage has not been achieved, an error message will be displayed, for example: ERROR: [125]: c module command/check/check is not fully covered
Debugging with files
Sometimes it is useful to look at files that were generated during the test. The default for running any test is that, at the start/end of the test, the test harness will clean up all files and directories created. To override this behavior, a single test run must be specified and the option --no-cleanup provided. Again, continuing with the check command, from define.yaml above, there are four tests. Below, test one will be run and nothing will be cleaned up so that the files and directories in test/test-0 can be inspected.
pgbackrest/test/test.pl --vm-out --module=command --test=check --run=1 --no-cleanup
Understanding Test Output
The following is a small sample of a typical test output.
run 8 - expireTimeBasedBackup()
run 8/1 ------------- L2285 no current backups
000.002s L2298 empty backup.info
000.009s 000.007s L2300 no backups to expire
run 8 - expireTimeBasedBackup() - indicates the run number (8) within the module and the parameter provided to testBegin, e.g. testBegin("expireTimeBasedBackup()")
run 8/1 ------------- L2285 no current backups - this is the first test (1) in run 8 which is the TEST_TITLE("no current backups"); at line number 2285.
000.002s L2298 empty backup.info - the first number, 000.002s, is the time in seconds that the test started from the beginning of the run. L2298 is the line number of the test and empty backup.info is the test comment.
000.009s 000.007s L2300 no backups to expire - again, 000.009s, is the time in seconds that the test started from the beginning of the run. The second number, 000.007s, is the run time of the previous test (i.e. empty backup.info test took 000.007 seconds to execute). L2300 is the line number of the test and no backups to expire is the test comment.
Adding an Option
Options can be added to a command or multiple commands. Options can be configuration file only, command-line only or valid for both. Once an option is successfully added, the config.auto.h and parse.auto.c.inc files will automatically be generated by the build system.
To add an option, two files need be to be modified:
These files are discussed in the following sections along with how to verify the help command output.
config.yaml
There are detailed comment blocks above each section that explain the rules for defining commands and options. Regarding options, there are two types: 1) command line only, and 2) configuration file. With the exception of secrets, all configuration file options can be passed on the command line. To configure an option for the configuration file, the section: key must be present.
The option: section is broken into sub-sections by a simple comment divider (e.g. # Repository options) under which the options are organized alphabetically by option name. To better explain this section, two hypothetical examples will be discussed. For more details, see config.yaml.
EXAMPLE 1 hypothetical command line only option
set:
type: string
command:
backup:
depend:
option: stanza
required: false
restore:
default: latest
command-role:
main: {}
Note that section: is not present thereby making this a command-line only option defined as follows:
set - the name of the optiontype - the type of the option. Valid values for types are: boolean, hash, integer, list, path, size, string, and timecommand - list each command for which the option is valid. If a command is not listed, then the option is not valid for the command and an error will be thrown if it is attempted to be used for that command. In this case the valid commands are backup and restore.backup - details the requirements for the --set option for the backup command. It is dependent on the option --stanza, meaning it is only allowed to be specified for the backup command if the --stanza option has been specified. And required: false indicates that the --set option is never required, even with the dependency.restore - details the requirements for the --set option for the restore command. Since required: is omitted, it is not required to be set by the user but it is required by the command and will default to latest if it has not been specified by the user.command-role - defines the processes for which the option is valid. main indicates the option will be used by the main process and not be passed on to other local/remote processes.EXAMPLE 2 hypothetical configuration file option
repo-test-type:
section: global
type: string
group: repo
default: full
allow-list:
- full
- diff
- incr
command:
backup: {}
restore: {}
command-role:
main: {}
repo-test-type - the name of the optionsection - the section of the configuration file where this option is valid (omitted for command line only options, see Example 1 above)type - the type of the option. Valid values for types are: boolean, hash, integer, list, path, size, string, and timegroup - indicates that this option is part of the repo group of indexed options and therefore will follow the indexing rules e.g. repo1-test-type.default - sets a default for the option if the option is not provided when the command is run. The default can be global (as it is here) or it can be specified for a specific command in the command section (as in Example 1 above).allow-list - lists the allowable values for the option for all commands for which the option is valid.command - list each command for which the option is valid. If a command is not listed, then the option is not valid for the command and an error will be thrown if it is attempted to be used for that command. In this case the valid commands are backup and restore.command-role - defines the processes for which the option is valid. main indicates the option will be used by the main process and not be passed on to other local/remote processes.
At compile time, the config.auto.h file will be generated to contain the constants used for options in the code. For the C enums, any dashes in the option name will be removed, camel-cased and prefixed with cfgOpt, e.g. repo-path becomes cfgOptRepoPath.
help.xml
All options must be documented or the system will error during the build. To add an option, find the command section identified by command id="COMMAND" section where COMMAND is the name of the command (e.g. expire) or, if the option is used by more than one command and the definition for the option is the same for all of the commands, the operation-general title="General Options" section.
To add an option, add the following to the <option-list> section; if it does not exist, then wrap the following in <option-list></option-list>. This example uses the boolean option force of the restore command. Simply replace that with your new option and the appropriate summary, text and example.
<option id="force" name="Force">
<summary>Force a restore.</summary>
<text>By itself this option forces the <postgres/> data and tablespace paths to be completely overwritten. In combination with <br-option>--delta</br-option> a timestamp/size delta will be performed instead of using checksums.</text>
<example>y</example>
</option>
A period (.) is required to end the summary section.Testing the help
It is important to run the help command unit test after adding an option in case a change is required:
To verify the help command output, build the executable:
pgbackrest/test/test.pl --vm=none --build-only
Use the executable to test the help output:
test/bin/none/pgbackrest help backup repo-type
Testing the documentation
To quickly view the HTML documentation, the --no-exe option can be passed to the documentation generator in order to bypass executing the code elements:
pgbackrest/doc/doc.pl --out=html --no-exe
The generated HTML files will be placed in the doc/output/html directory where they can be viewed locally in a browser.
If Docker is installed, it will be used by the documentation generator to execute the code elements while building the documentation, therefore, the --no-exe should be omitted, (i.e. pgbackrest/doc/doc.pl --output=html). --no-cache may be used to force a full build even when no code elements have changed since the last build. --pre will reuse the container definitions from the prior build and saves time during development.
The containers created for documentation builds can be useful for manually testing or trying out new code or features. The following demonstrates building through just the quickstart section of the user-guide without encryption.
The resulting Docker containers can be listed with docker ps and the container can be entered with docker exec doc-pg-primary bash. Additionally, the -u option can be added for entering the container as a specific user (e.g. postgres).
Submitting a Pull Request
Before submitting a Pull Request:
Does it meet the coding standards?Have Unit Tests been written and run with 100% coverage?If your submission includes changes to the help or online documentation, have the help and documentation tests been run?Has it passed continuous integration testing? Simply renaming your branch with the appendix -cig and pushing it to your GitHub account will initiate GitHub Actions to run CI tests.
When submitting a Pull Request:
Provide a short submission title.Write a detailed comment to describe the purpose of your submission and any issue(s), if any, it is resolving; a link to the GitHub issue is also helpful.Select the integration branch as the base for your PR, do not select main nor any other branch.
After submitting a Pull Request:
One or more reviewers will be assigned.Respond to any issues (conversations) in GitHub but do not resolve the conversation; the reviewer is responsible for ensuring the issue raised has been resolved and marking the conversation resolved. It is helpful to supply the commit in your reply if one was submitted to fix the issue.