package Test::PostgreSQL; use 5.014; use strict; use warnings; use Moo; use Types::Standard -all; use Function::Parameters qw(:strict); use Try::Tiny; use DBI; use File::Spec; use File::Temp; use File::Which; use POSIX qw(SIGQUIT SIGKILL WNOHANG getuid setuid); use User::pwent; our $VERSION = '1.28'; our $errstr; has dbname => ( is => 'ro', isa => Str, default => 'test', ); has dbowner => ( is => 'ro', isa => Str, default => 'postgres', ); has host => ( is => 'ro', isa => Str, default => '127.0.0.1', ); # Various paths that Postgres gets installed under, sometimes with a version on the end, # in which case take the highest version. We append /bin/ and so forth to the path later. # *Note that these are used only if the program isn't already in the path!* has search_paths => ( is => "ro", isa => ArrayRef, builder => "_search_paths", ); method _search_paths() { my @base_paths = ( # popular installation dir? qw(/usr/local/pgsql), # ubuntu (maybe debian as well, find the newest version) (sort { $b cmp $a } grep { -d $_ } glob "/usr/lib/postgresql/*"), # macport (sort { $b cmp $a } grep { -d $_ } glob "/opt/local/lib/postgresql*"), # Postgresapp.com (sort { $b cmp $a } grep { -d $_ } glob "/Applications/Postgres.app/Contents/Versions/*"), # BSDs end up with it in /usr/local/bin which doesn't appear to be in the path sometimes: "/usr/local", ); # This environment variable is used to override the default, so it gets # prefixed to the start of the search paths. if (defined $ENV{POSTGRES_HOME}) { return [$ENV{POSTGRES_HOME}, @base_paths]; } return \@base_paths; } # We attempt to use this port first, and will increment from there. # The final port ends up in the ->port attribute. has base_port => ( is => "ro", isa => Int, default => 15432, ); has auto_start => ( is => "ro", default => 2, ); has base_dir => ( is => "rw", default => sub { File::Temp->newdir( 'pgtest.XXXXX', CLEANUP => $ENV{TEST_POSTGRESQL_PRESERVE} ? undef : 1, EXLOCK => 0, TMPDIR => 1 ); }, coerce => fun ($newval) { # Ensure base_dir is absolute; usually only the case if the user set it. # Avoid munging objects such as File::Temp ref $newval ? $newval : File::Spec->rel2abs($newval); }, ); has socket_dir => ( is => "ro", isa => Str, lazy => 1, default => method () { File::Spec->catdir( $self->base_dir, 'tmp' ) }, ); has initdb => ( is => "ro", isa => Str, lazy => 1, default => method () { $self->_find_program('initdb') || die $errstr }, ); has initdb_args => ( is => "lazy", isa => Str, ); method _build_initdb_args() { return '-U '. $self->dbowner . ' -A trust ' . $self->extra_initdb_args; } has extra_initdb_args => ( is => "ro", isa => Str, default => "", ); has unix_socket => ( is => "ro", isa => Bool, default => 0, ); has pg_version => ( is => 'ro', isa => Str, lazy => 1, predicate => 1, builder => "_pg_version_builder", ); method _pg_version_builder() { my $ver_cmd = join ' ', ( $self->postmaster, '--version' ); my ($ver) = qx{$ver_cmd} =~ /(\d+(?:\.\d+)?)/; return $ver; } has pg_ctl => ( is => "ro", isa => Maybe[Str], lazy => 1, builder => "_pg_ctl_builder", ); method _pg_ctl_builder() { my $prog = $self->_find_program('pg_ctl'); if ( $prog ) { # we only use pg_ctl if Pg version is >= 9 my $ret = qx/"$prog" --version/; if ( $ret =~ /(\d+)(?:\.|devel|beta)/ && $1 >= 9 ) { return $prog; } warn "pg_ctl version earlier than 9"; return; } return; } has pg_config => ( is => 'ro', isa => Str, ); has psql => ( is => 'ro', isa => Str, lazy => 1, default => method () { $self->_find_program('psql') || die $errstr }, ); has psql_args => ( is => 'lazy', isa => Str, ); method _build_psql_args() { return '-U ' . $self->dbowner . ' -d ' . $self->dbname . ' -h '. ($self->unix_socket ? $self->socket_dir : '127.0.0.1') . ' -p ' . $self->port . $self->extra_psql_args; } has extra_psql_args => ( is => 'ro', isa => Str, default => '', ); has run_psql_args => ( is => 'ro', isa => Str, lazy => 1, builder => "_build_run_psql_args", ); method _build_run_psql_args() { my @args = ( '-1', # Single transaction '-X', # Ignore .psqlrc '-q', # Quiet '-v ON_ERROR_STOP=1', # Stop on first error ); # Echo errors, available in psql 9.5+ push @args, '-b' if $self->pg_version >= 9.5; return join ' ', @args; } has seed_scripts => ( is => 'ro', isa => ArrayRef[Str], default => sub { [] }, ); has pid => ( is => "rw", isa => Maybe[Int], ); has port => ( is => "rw", isa => Maybe[Int], ); has uid => ( is => "rw", isa => Maybe[Int], ); # Are we running as root? (Typical when run inside Docker containers) has is_root => ( is => "ro", isa => Bool, default => sub { getuid == 0 } ); has postmaster => ( is => "rw", isa => Str, lazy => 1, default => method () { $self->_find_program("postgres") || $self->_find_program("postmaster") || die $errstr }, ); has postmaster_args => ( is => "lazy", isa => Str, ); method _build_postmaster_args() { return "-h ". ($self->unix_socket ? "''" : "127.0.0.1") . " -F " . $self->extra_postmaster_args; } has extra_postmaster_args => ( is => "ro", isa => Str, default => "", ); has _owner_pid => ( is => "ro", isa => Int, default => sub { $$ }, ); method BUILD($) { # Ensure we have one or the other ways of starting Postgres: try { $self->pg_ctl or $self->postmaster } catch { die $_ }; if (defined $self->uid and $self->uid == 0) { die "uid() must be set to a non-root user id."; } if (not defined($self->uid) and $self->is_root) { my $ent = getpwnam("nobody"); unless (defined $ent) { die "user nobody does not exist, use uid() to specify a non-root user."; } unless ($ent->uid > 0) { die "user nobody has uid 0; confused and exiting. use uid() to specify a non-root user."; } $self->uid($ent->uid); } # Ensure base dir is writable by our target uid, if we were running as root chown $self->uid, -1, $self->base_dir if defined $self->uid; if ($self->auto_start) { $self->setup if $self->auto_start >= 2; $self->start; } } method DEMOLISH($in_global_destruction) { local $?; if (defined $self->pid && $self->_owner_pid == $$) { $self->stop } return; } sub dsn { my %args = shift->_default_args(@_); return 'DBI:Pg:' . join(';', map { "$_=$args{$_}" } sort keys %args); } sub _default_args { my ($self, %args) = @_; # If we're doing socket-only (i.e., not listening on localhost), # then provide the path to the socket if ($self->{unix_socket}) { $args{host} //= $self->socket_dir; } else { $args{host} ||= $self->host; } $args{port} ||= $self->port; $args{user} ||= $self->dbowner; $args{dbname} ||= $self->dbname; return %args; } sub uri { my $self = shift; my %args = $self->_default_args(@_); return sprintf('postgresql://%s@%s:%d/%s', @args{qw/user host port dbname/}); } method start() { if (defined $self->pid) { warn "Apparently already started on " . $self->pid . "; not restarting."; return; } # If the user specified a port, try only that port: if ($self->port) { $self->_try_start($self->port); } else { $self->_find_port_and_launch; } # create "test" database $self->_create_test_database($self->dbname); } # This whole method was mostly cargo-culted from the earlier test-postgresql; # It could probably be made more sane. method _find_port_and_launch() { my $tries = 10; my $port = $self->base_port; srand(); # Re-seed the RNG in case the caller forked the process # try by incrementing port number until PostgreSQL starts while (1) { my $good = try { $self->_try_start($port); 1; } catch { # warn "Postgres failed to start on port $port\n"; unless ($tries--) { die "Failed to start postgres after trying 10 potential ports: $_"; } undef; }; return if $good; # Increment port by a random number to avoid clashes with other Test::Postgresql processes # Keep in mind that this increment is going to be made up to 10 times, so avoid exceeding 64k $port += int(rand(500)) + 1; } } method _try_start($port) { my $logfile = File::Spec->catfile($self->base_dir, 'postgres.log'); if ( $self->pg_ctl ) { my @cmd = ( $self->pg_ctl, 'start', '-w', '-s', '-D', File::Spec->catdir( $self->base_dir, 'data' ), '-l', $logfile, '-o', join( ' ', $self->postmaster_args, '-p', $port, '-k', $self->socket_dir) ); $self->setuid_cmd(\@cmd, 1); my $pid_path = File::Spec->catfile( $self->base_dir, 'data', 'postmaster.pid' ); open( my $pidfh, '<', $pid_path ) or die "Failed to open $pid_path: $!"; # Note that the file contains several lines; we only want the PID from the first. my $pid = <$pidfh>; chomp $pid; $self->pid($pid); close $pidfh; $self->port($port); } else { # old style - open log and fork open my $logfh, '>>', $logfile or die "failed to create log file: $logfile: $!"; my $pid = fork; die "fork(2) failed:$!" unless defined $pid; if ($pid == 0) { open STDOUT, '>>&', $logfh or die "dup(2) failed:$!"; open STDERR, '>>&', $logfh or die "dup(2) failed:$!"; chdir $self->base_dir or die "failed to chdir to:" . $self->base_dir . ":$!"; if (defined $self->uid) { setuid($self->uid) or die "setuid failed: $!"; } my $cmd = join( ' ', $self->postmaster, $self->postmaster_args, '-p', $port, '-D', File::Spec->catdir($self->base_dir, 'data'), '-k', $self->socket_dir, ); exec($cmd); die "failed to launch postmaster:$?"; } close $logfh; # wait until server becomes ready (or dies) for (my $i = 0; $i < 100; $i++) { open $logfh, '<', $logfile or die "failed to create log file: $logfile: $!"; my $lines = do { join '', <$logfh> }; close $logfh; last if $lines =~ /is ready to accept connections/; if (waitpid($pid, WNOHANG) > 0) { # failed die "Failed to start Postgres: $lines\n"; } sleep 1; } # PostgreSQL is ready $self->pid($pid); $self->port($port); } return; } method stop($sig = SIGQUIT) { if ( $self->pg_ctl && defined $self->base_dir ) { my @cmd = ( $self->pg_ctl, 'stop', '-s', '-D', File::Spec->catdir( $self->base_dir, 'data' ), '-m', 'fast' ); $self->setuid_cmd(\@cmd); } else { # old style or $self->base_dir File::Temp obj already DESTROYed return unless defined $self->pid; kill $sig, $self->pid; my $timeout = 10; while ($timeout > 0 and waitpid($self->pid, WNOHANG) == 0) { $timeout -= sleep(1); } if ($timeout <= 0) { warn "Pg refused to die gracefully; killing it violently.\n"; kill SIGKILL, $self->pid; $timeout = 5; while ($timeout > 0 and waitpid($self->pid, WNOHANG) == 0) { $timeout -= sleep(1); } if ($timeout <= 0) { warn "Pg really didn't die.. WTF?\n"; } } } $self->pid(undef); return; } method _create_test_database($dbname) { my $tries = 5; my $dbh; while ($tries) { $tries -= 1; $dbh = DBI->connect($self->dsn(dbname => 'template1'), '', '', { PrintError => 0, RaiseError => 0 }); last if $dbh; # waiting for database to start up if ($DBI::errstr =~ /the database system is starting up/ || $DBI::errstr =~ /Connection refused/) { sleep(1); next; } die $DBI::errstr; } die "Connection to the database failed even after 5 tries" unless ($dbh); if ($dbh->selectrow_arrayref(qq{SELECT COUNT(*) FROM pg_database WHERE datname='$dbname'})->[0] == 0) { $dbh->do("CREATE DATABASE $dbname") or die $dbh->errstr; } my $seed_scripts = $self->seed_scripts || []; $self->run_psql_scripts(@$seed_scripts) if @$seed_scripts; return; } method setup() { # (re)create directory structure mkdir $self->base_dir; chmod 0755, $self->base_dir or die "failed to chmod 0755 dir:" . $self->base_dir . ":$!"; if ($ENV{USER} && $ENV{USER} eq 'root') { chown $self->uid, -1, $self->base_dir or die "failed to chown dir:" . $self->base_dir . ":$!"; } my $tmpdir = $self->socket_dir; if (mkdir $tmpdir) { if ($self->uid) { chown $self->uid, -1, $tmpdir or die "failed to chown dir:$tmpdir:$!"; } } # initdb if (! -d File::Spec->catdir($self->base_dir, 'data')) { if ( $self->pg_ctl ) { my @cmd = ( $self->pg_ctl, 'init', '-s', '-D', File::Spec->catdir($self->base_dir, 'data'), '-o', $self->initdb_args, ); $self->setuid_cmd(\@cmd); } else { # old style pipe my $rfh, my $wfh or die "failed to create pipe:$!"; my $pid = fork; die "fork failed:$!" unless defined $pid; if ($pid == 0) { close $rfh; open STDOUT, '>&', $wfh or die "dup(2) failed:$!"; open STDERR, '>&', $wfh or die "dup(2) failed:$!"; chdir $self->base_dir or die "failed to chdir to:" . $self->base_dir . ":$!"; if (defined $self->uid) { setuid($self->uid) or die "setuid failed:$!"; } my $cmd = join( ' ', $self->initdb, $self->initdb_args, '-D', File::Spec->catdir($self->base_dir, 'data'), ); exec($cmd); die "failed to exec:$cmd:$!"; } close $wfh; my $output = ''; while (my $l = <$rfh>) { $output .= $l; } close $rfh; while (waitpid($pid, 0) <= 0) { } die "*** initdb failed ***\n$output\n" if $? != 0; } my $conf_file = File::Spec->catfile($self->base_dir, 'data', 'postgresql.conf'); if (my $pg_config = $self->pg_config) { open my $fh, '>', $conf_file or die "Can't open $conf_file: $!"; print $fh $pg_config; close $fh; } else { # use postgres hard-coded configuration as some packagers mess # around with postgresql.conf.sample too much: truncate $conf_file, 0; } } } method _find_program($prog) { undef $errstr; my $path = which $prog; return $path if $path; for my $sp (@{$self->search_paths}) { return "$sp/bin/$prog" if -x "$sp/bin/$prog"; return "$sp/$prog" if -x "$sp/$prog"; } $errstr = "could not find $prog, please set appropriate PATH or POSTGRES_HOME"; return; } method setuid_cmd($cmd, $suppress_errors = !1) { my $pid = fork; if ($pid == 0) { chdir $self->base_dir; if (defined $self->uid) { setuid($self->uid) or die "setuid failed: $!"; } close STDERR if $suppress_errors; exec(@$cmd) or die "Failed to exec pg_ctl: $!"; } else { waitpid($pid, 0); } } method run_psql(@psql_args) { my $cmd = join ' ', ( $self->psql, # Default connection settings $self->psql_args, # Extra connection settings or something else $self->extra_psql_args, # run_psql specific arguments $self->run_psql_args, @psql_args, ); # Usually anything less than WARNING is not really helpful # in batch mode. Does it make sense to make this configurable? local $ENV{PGOPTIONS} = '--client-min-messages=warning'; my $psql_out = qx{$cmd 2>&1}; die "Error executing psql: $psql_out" unless $? == 0; } method run_psql_scripts(@script_paths) { my @psql_commands; # psql 9.6+ supports multiple -c and -f commands invoked at once, # older psql does not. Executing psql multiple times breaks single # transaction semantics but is unlikely to cause problems in real world. if ( $self->pg_version > 9.6 ) { push @psql_commands, join ' ', map {; "-f $_" } @script_paths; } else { @psql_commands = map {; "-f $_" } @script_paths; } $self->run_psql($_) for @psql_commands; } 1; __END__ =pod =encoding utf8 =head1 NAME Test::PostgreSQL - PostgreSQL runner for tests =head1 SYNOPSIS use DBI; use Test::PostgreSQL; use Test::More; # optionally # (if not already set at shell): # # $ENV{POSTGRES_HOME} = '/path/to/my/pgsql/installation'; my $pgsql = eval { Test::PostgreSQL->new() } or plan skip_all => $@; plan tests => XXX; my $dbh = DBI->connect($pgsql->dsn); =head1 DESCRIPTION C automatically setups a PostgreSQL instance in a temporary directory, and destroys it when the perl script exits. This module is a fork of Test::postgresql, which was abandoned by its author several years ago. =head1 ATTRIBUTES C object has the following attributes, overridable by passing corresponding argument to constructor: =head2 dbname Database name to use in this C instance. Default is C. =head2 dbowner Database owner user name. Default is C. =head2 host Host name or IP address to use for PostgreSQL instance connections. Default is C<127.0.0.1>. =head2 base_dir Base directory under which the PostgreSQL instance is being created. The property can be passed as a parameter to the constructor, in which case the directory will not be removed at exit. =head2 base_port Connection port number to start with. If the port is already used we will increment the value and try again. Default: C<15432>. =head2 unix_socket Whether to only connect via UNIX sockets; if false (the default), connections can occur via localhost. [This changes the L returned to only give the UNIX socket directory, and avoids any issues with conflicting TCP ports on localhost.] =head2 socket_dir Unix socket directory to use if L is true. Default is C<$basedir/tmp>. =head2 pg_ctl Path to C program which is part of the PostgreSQL distribution. Starting with PostgreSQL version 9.0 C can be used to start/stop postgres without having to use fork/pipe and will be chosen automatically if L is not set but the program is found and the version is recent enough. B do NOT use this with PostgreSQL versions prior to version 9.0. By default we will try to find C in PostgresSQL directory. =head2 initdb Path to C program which is part of the PostreSQL distribution. Default is to try and find it in PostgreSQL directory. =head2 initdb_args Arguments to pass to C program when creating a new PostgreSQL database cluster for Test::PostgreSQL session. Defaults to C<-U postgres -A trust>. See L. =head2 extra_initdb_args Extra args to be appended to L. Default is empty. =head2 pg_config Configuration to place in C<$basedir/data/postgresql.conf>. Use this to override PostgreSQL configuration defaults, e.g. to speed up PostgreSQL database init and seeding one might use something like this: my $pgsql = Test::PostgreSQL->new( pg_config => q| # foo baroo mymse throbbozongo fsync = off synchronous_commit = off full_page_writes = off bgwriter_lru_maxpages = 0 shared_buffers = 512MB effective_cache_size = 512MB work_mem = 100MB |); =head2 postmaster Path to C which is part of the PostgreSQL distribution. If not set, the programs are automatically searched by looking up $PATH and other prefixed directories. Since C is deprecated in newer PostgreSQL versions C is used in preference to C. =head2 postmaster_args Defaults to C<-h 127.0.0.1 -F>. =head2 extra_postmaster_args Extra args to be appended to L. Default is empty. =head2 psql Path to C client which is part of the PostgreSQL distribution. C can be used to run SQL scripts against the temporary database created by L: my $pgsql = Test::PostgreSQL->new(); my $psql = $pgsql->psql; my $out = `$psql -f /path/to/script.sql 2>&1`; die "Error executing script.sql: $out" unless $? == 0; =head2 psql_args Command line arguments necessary for C to connect to the correct PostgreSQL instance. Defaults to C<-U postgres -d test -h 127.0.0.1 -p $self-Eport>. See also L, L, L, L. =head2 extra_psql_args Extra args to be appended to L. =head2 run_psql_args Arguments specific for L invocation, used mostly to set up and seed database schema after PostgreSQL instance is launched and configured. Default is C<-1Xqb -v ON_ERROR_STOP=1>. This means: =over 4 =item * 1: Run all SQL statements in passed scripts as single transaction =item * X: Skip C<.psqlrc> files =item * q: Run quietly, print only notices and errors on stderr (if any) =item * b: Echo SQL statements that cause PostgreSQL exceptions (version 9.5+) =item * -v ON_ERROR_STOP=1: Stop processing SQL statements after the first error =back =head2 seed_scripts Arrayref with the list of SQL scripts to run after the database was instanced and set up. Default is C<[]>. B that C older than 9.6 does not support multiple C<-c> and C<-f> switches in arguments so C will be executed one by one. This implies multiple transactions instead of just one; if you need all seed statements to apply within a single transaction, combine them into one seed script. =head2 auto_start Integer value that controls whether PostgreSQL server is started and setup after creating C instance. Possible values: =over 4 =item C<0> Do not start PostgreSQL. =item C<1> Start PostgreSQL but do not run L. =item C<2> Start PostgreSQL and run L. Default is C<2>. =back =head1 METHODS =head2 new Create and run a PostgreSQL instance. The instance is terminated when the returned object is being DESTROYed. If required programs (initdb and postmaster) were not found, the function returns undef and sets appropriate message to $Test::PostgreSQL::errstr. =head2 dsn Builds and returns dsn by using given parameters (if any). Default username is C, and dbname is C (an empty database). =head2 uri Builds and returns a connection URI using the given parameters (if any). See L for details about the format. Default username is C, and dbname is C (an empty database). =head2 pid Returns process id of PostgreSQL (or undef if not running). =head2 port Returns TCP port number on which postmaster is accepting connections (or undef if not running). =head2 start Starts postmaster. =head2 stop Stops postmaster. =head2 setup Setups the PostgreSQL instance. Note that this method should be invoked I L. =head2 run_psql Execute C program with the given list of arguments. Usually this would be something like: $pgsql->run_psql('-c', q|'INSERT INTO foo (bar) VALUES (42)'|); Or: $pgsql->run_psql('-f', '/path/to/script.sql'); Note that when using literal SQL statements with C<-c> parameter you will need to escape them manually like shown above. C will not quote them for you. The actual command line to execute C will be concatenated from L, L, and L. B that C older than 9.6 does not support multiple C<-c> and/or C<-f> switches in arguments. =head2 run_psql_scripts Given a list of script file paths, invoke L once with C<-f 'script'> for every path in PostgreSQL 9.6+, or once per C<-f 'script'> for older PostgreSQL versions. =head1 ENVIRONMENT =head2 POSTGRES_HOME If your postgres installation is not located in a well known path, or you have many versions installed and want to run your tests against particular one, set this environment variable to the desired path. For example: export POSTGRES_HOME='/usr/local/pgsql94beta' This is the same idea and variable name which is used by the installer of L. =head1 AUTHOR Toby Corkindale, Kazuho Oku, Peter Mottram, Alex Tokarev, plus various contributors. =head1 COPYRIGHT Current version copyright © 2012-2015 Toby Corkindale. Previous versions copyright (C) 2009 Cybozu Labs, Inc. =head1 LICENSE This module is free software, released under the Perl Artistic License 2.0. See L for more information. =cut