--- deliantra/server/lib/cf.pm 2007/01/19 19:42:37 1.183 +++ deliantra/server/lib/cf.pm 2007/02/12 01:25:55 1.209 @@ -10,19 +10,23 @@ use Safe; use Safe::Hole; -use Coro 3.4 (); +use Coro 3.5 (); use Coro::Event; use Coro::Timer; use Coro::Signal; use Coro::Semaphore; use Coro::AIO; +use BDB (); use Data::Dumper; use Digest::MD5; use Fcntl; use IO::AIO 2.32 (); use YAML::Syck (); use Time::HiRes; +use Compress::LZF; + +Compress::LZF::sfreeze_cr { }; # prime Compress::LZF so it does not use require later use Event; $Event::Eval = 1; # no idea why this is required, but it is @@ -50,6 +54,10 @@ our $WRITE_RUNTIME_WATCHER; our $NEXT_TICK; our $NOW; +our $USE_FSYNC = 1; # use fsync to write maps - default off + +our $BDB_POLL_WATCHER; +our $DB_ENV; our %CFG; @@ -60,6 +68,7 @@ our %MAP; # all maps our $LINK_MAP; # the special {link} map, which is always available our $RANDOM_MAPS = cf::localdir . "/random"; +our $BDB_ENV_DIR = cf::localdir . "/db"; our $WAIT_FOR_TICK; $WAIT_FOR_TICK ||= new Coro::Signal; our $WAIT_FOR_TICK_ONE; $WAIT_FOR_TICK_ONE ||= new Coro::Signal; @@ -82,9 +91,12 @@ mkdir cf::localdir . "/" . cf::tmpdir; mkdir cf::localdir . "/" . cf::uniquedir; mkdir $RANDOM_MAPS; +mkdir $BDB_ENV_DIR; our $EMERGENCY_POSITION; +sub cf::map::normalise; + ############################################################################# =head2 GLOBAL VARIABLES @@ -351,10 +363,14 @@ my $fh = aio_open "$runtime~", O_WRONLY | O_CREAT, 0644 or return; - my $value = $cf::RUNTIME + 1 + 10; # 10 is the runtime save interval, for a monotonic clock + my $value = $cf::RUNTIME + 90 + 10; + # 10 is the runtime save interval, for a monotonic clock + # 60 allows for the watchdog to kill the server. + (aio_write $fh, 0, (length $value), $value, 0) <= 0 and return; + # always fsync - this file is important aio_fsync $fh and return; @@ -749,7 +765,7 @@ if (my $fh = aio_open "$filename~", O_WRONLY | O_CREAT, 0600) { chmod SAVE_MODE, $fh; aio_write $fh, 0, (length $$rdata), $$rdata, 0; - aio_fsync $fh; + aio_fsync $fh if $cf::USE_FSYNC; close $fh; if (@$objs) { @@ -757,7 +773,7 @@ chmod SAVE_MODE, $fh; my $data = Storable::nfreeze { version => 1, objs => $objs }; aio_write $fh, 0, (length $data), $data, 0; - aio_fsync $fh; + aio_fsync $fh if $cf::USE_FSYNC; close $fh; aio_rename "$filename.pst~", "$filename.pst"; } @@ -773,6 +789,12 @@ aio_unlink $filename; aio_unlink "$filename.pst"; } + + #d##TODO# nuke non .map-files if exist + if ($filename =~ s/\.map$//) { + aio_unlink $filename; + aio_unlink "$filename.pst"; + } } } @@ -789,6 +811,9 @@ my ($data, $av); + #d#TODO remove .map if file does not exist + aio_stat $filename and $filename =~ s/\.map$//; + (aio_load $filename, $data) >= 0 or return; @@ -1067,6 +1092,8 @@ sub maps($) { my ($pl) = @_; + $pl = ref $pl ? $pl->ob->name : $pl; + my $files = aio_readdir playerdir $pl or return; @@ -1077,8 +1104,7 @@ next if /\.(?:pl|pst)$/; next unless /^$PATH_SEP/o; - s/\.map$//; - push @paths, "~" . $pl->ob->name . "/" . $_; + push @paths, cf::map::normalise "~$pl/$_"; } \@paths @@ -1139,18 +1165,26 @@ ) } +=item cf::map->register ($regex, $prio) + +Register a handler for the map path matching the given regex at the +givne priority (higher is better, built-in handlers have priority 0, the +default). + +=cut + sub register { - my (undef, $regex) = @_; + my (undef, $regex, $prio) = @_; my $pkg = caller; no strict; push @{"$pkg\::ISA"}, __PACKAGE__; - $EXT_MAP{$pkg} = qr<$regex>; + $EXT_MAP{$pkg} = [$prio, qr<$regex>]; } # also paths starting with '/' -$EXT_MAP{"cf::map"} = qr{^(?=/)}; +$EXT_MAP{"cf::map"} = [0, qr{^(?=/)}]; sub thawer_merge { my ($self, $merge) = @_; @@ -1167,6 +1201,10 @@ sub normalise { my ($path, $base) = @_; + $path = "$path"; # make sure its a string + + $path =~ s/\.map$//; + # map plan: # # /! non-realised random map exit (special hack!) @@ -1203,8 +1241,8 @@ $path = normalise $path, $base; - for my $pkg (keys %EXT_MAP) { - if ($path =~ $EXT_MAP{$pkg}) { + for my $pkg (sort { $EXT_MAP{$b}[0] <=> $EXT_MAP{$a}[0] } keys %EXT_MAP) { + if ($path =~ $EXT_MAP{$pkg}[1]) { my $self = bless cf::map::new, $pkg; $self->{path} = $path; $self->path ($path); $self->init; # pass $1 etc. @@ -1212,7 +1250,7 @@ } } - Carp::carp "unable to resolve path '$path'."; + Carp::carp "unable to resolve path '$path' (base '$base')."; () } @@ -1237,7 +1275,7 @@ sub load_path { my ($self) = @_; - sprintf "%s/%s/%s", cf::datadir, cf::mapdir, $self->{path} + sprintf "%s/%s/%s.map", cf::datadir, cf::mapdir, $self->{path} } # the temporary/swap location @@ -1245,7 +1283,7 @@ my ($self) = @_; (my $path = $_[0]{path}) =~ s/\//$PATH_SEP/g; - sprintf "%s/%s/%s", cf::localdir, cf::tmpdir, $path + sprintf "%s/%s/%s.map", cf::localdir, cf::tmpdir, $path } # the unique path, undef == no special unique path @@ -1271,14 +1309,19 @@ utf8::encode (my $save = $self->save_path); IO::AIO::aioreq_pri 4; Coro::AIO::aio_unlink $save; IO::AIO::aioreq_pri 4; Coro::AIO::aio_unlink "$save.pst"; + + #d#TODO remove .map and also nuke + $save =~ s/\.map// or return;#d# + IO::AIO::aioreq_pri 4; Coro::AIO::aio_unlink $save;#d# + IO::AIO::aioreq_pri 4; Coro::AIO::aio_unlink "$save.pst";#d# } sub load_header_from($) { my ($self, $path) = @_; utf8::encode $path; - aio_open $path, O_RDONLY, 0 - or return; + #aio_open $path, O_RDONLY, 0 + # or return; $self->_load_header ($path) or return; @@ -1288,35 +1331,46 @@ 1 } -sub load_orig { +sub load_header_orig { my ($self) = @_; $self->load_header_from ($self->load_path) } -sub load_temp { +sub load_header_temp { my ($self) = @_; $self->load_header_from ($self->save_path) } +sub prepare_temp { + my ($self) = @_; + + $self->last_access ((delete $self->{last_access}) + || $cf::RUNTIME); #d# + # safety + $self->{instantiate_time} = $cf::RUNTIME + if $self->{instantiate_time} > $cf::RUNTIME; +} + +sub prepare_orig { + my ($self) = @_; + + $self->{load_original} = 1; + $self->{instantiate_time} = $cf::RUNTIME; + $self->last_access ($cf::RUNTIME); + $self->instantiate; +} + sub load_header { my ($self) = @_; - if ($self->load_temp) { - $self->last_access ((delete $self->{last_access}) - || $cf::RUNTIME); #d# - # safety - $self->{instantiate_time} = $cf::RUNTIME - if $self->{instantiate_time} > $cf::RUNTIME; + if ($self->load_header_temp) { + $self->prepare_temp; } else { - $self->load_orig + $self->load_header_orig or return; - - $self->{load_original} = 1; - $self->{instantiate_time} = $cf::RUNTIME; - $self->last_access ($cf::RUNTIME); - $self->instantiate; + $self->prepare_orig; } 1 @@ -1340,21 +1394,26 @@ $map->load_header or return; - if ($map->should_reset) { - $cf::WAIT_FOR_TICK->wait; + if ($map->should_reset && 0) {#d#TODO# disabled, crashy (locking issue?) + # doing this can freeze the server in a sync job, obviously + #$cf::WAIT_FOR_TICK->wait; $map->reset; undef $guard; - $map = find $path - or return; + return find $path; } $cf::MAP{$path} = $map } } +sub pre_load { } +sub post_load { } + sub load { my ($self) = @_; + local $self->{deny_reset} = 1; # loading can take a long time + my $path = $self->{path}; my $guard = cf::lock_acquire "map_load:$path"; @@ -1363,6 +1422,9 @@ $self->in_memory (cf::MAP_LOADING); $self->alloc; + + $self->pre_load; + $self->_load_objects ($self->{load_path}, 1) or return; @@ -1394,6 +1456,8 @@ $self->activate; } + $self->post_load; + $self->in_memory (cf::MAP_IN_MEMORY); } @@ -1458,7 +1522,6 @@ $MAP_PREFETCHER ||= cf::async { while (%MAP_PREFETCH) { for my $path (keys %MAP_PREFETCH) { - warn "prefetching $path...\n";#d# my $map = find $path or next; $map->load; @@ -1468,6 +1531,7 @@ } undef $MAP_PREFETCHER; }; + $MAP_PREFETCHER->prio (6); () } @@ -1596,8 +1660,7 @@ next if /\.pst$/; next unless /^$PATH_SEP/o; - s/\.map$//; - push @paths, $_; + push @paths, cf::map::normalise $_; } \@paths @@ -1701,7 +1764,7 @@ sub link_map { unless ($LINK_MAP) { $LINK_MAP = cf::map::find "{link}" - or do { warn "FATAL: unable to provide {link} map, exiting."; exit 1 }; + or cf::cleanup "FATAL: unable to provide {link} map, exiting."; $LINK_MAP->load; } @@ -1718,7 +1781,7 @@ $self->{_link_pos} ||= [$self->map->{path}, $self->x, $self->y] if $self->map; - $self->enter_map ($LINK_MAP || link_map, 20, 20); + $self->enter_map ($LINK_MAP || link_map, 10, 10); } sub cf::object::player::leave_link { @@ -1795,13 +1858,12 @@ $self->enter_link; (async { - my $map = cf::map::find $path; - $map = $map->customise_for ($self) if $map; - -# warn "entering ", $map->path, " at ($x, $y)\n" -# if $map; - - $map or $self->message ("The exit to '" . ($path->visible_name) . "' is closed", cf::NDI_UNIQUE | cf::NDI_RED); + my $map = eval { + my $map = cf::map::find $path; + $map = $map->customise_for ($self) if $map; + $map + } or + $self->message ("The exit to '$path' is closed", cf::NDI_UNIQUE | cf::NDI_RED); $self->leave_link ($map, $x, $y); })->prio (1); @@ -1842,7 +1904,7 @@ my $rmp = parse_random_map_params $exit->msg; if ($exit->map) { - $rmp->{region} = $exit->map->region_name; + $rmp->{region} = $exit->region->name; $rmp->{origin_map} = $exit->map->path; $rmp->{origin_x} = $exit->x; $rmp->{origin_y} = $exit->y; @@ -1869,16 +1931,21 @@ return unless $self->type == cf::PLAYER; + if ($exit->slaying eq "/!") { + #TODO: this should de-fi-ni-te-ly not be a sync-job + cf::sync_job { prepare_random_map $exit }; + } + + my $slaying = cf::map::normalise $exit->slaying, $exit->map && $exit->map->path; + my $hp = $exit->stats->hp; + my $sp = $exit->stats->sp; + $self->enter_link; (async { $self->deactivate_recursive; # just to be sure unless (eval { - prepare_random_map $exit - if $exit->slaying eq "/!"; - - my $path = new_from_path cf::map $exit->slaying, $exit->map && $exit->map->path; - $self->goto ($path, $exit->stats->hp, $exit->stats->sp); + $self->goto ($slaying, $hp, $sp); 1; }) { @@ -1949,6 +2016,7 @@ or return; # be conservative, not sure how that can happen, but we saw a crash here (shift @$queue)->[1]->($msg); + return unless $ns->valid; # temporary(?) workaround for callback destroying socket push @{ $ns->{query_queue} }, @$queue; @@ -2119,98 +2187,70 @@ =over 4 -=item $hashref = cf::db_get $family - -Return a hashref for use by the extension C<$family>, which can be -modified. After modifications, you have to call C or -C. - =item $value = cf::db_get $family => $key -Returns a single value from the database - -=item cf::db_put $family => $hashref - -Stores the given family hashref into the database. Updates are delayed, if -you want the data to be synced to disk immediately, use C. +Returns a single value from the database. =item cf::db_put $family => $key => $value -Stores the given C<$value> in the family hash. Updates are delayed, if you -want the data to be synced to disk immediately, use C. - -=item cf::db_dirty - -Marks the database as dirty, to be updated at a later time. - -=item cf::db_sync - -Immediately write the database to disk I. +Stores the given C<$value> in the family. =cut our $DB; -{ - my $path = cf::localdir . "/database.pst"; - - sub db_load() { - $DB = stat $path ? Storable::retrieve $path : { }; - } +unless ($DB) { + $DB = BDB::db_create $DB_ENV; - my $pid; + cf::sync_job { + eval { + $DB->set_flags (BDB::CHKSUM); - sub db_save() { - waitpid $pid, 0 if $pid; - if (0 == ($pid = fork)) { - $DB->{_meta}{version} = 1; - Storable::nstore $DB, "$path~"; - rename "$path~", $path; - cf::_exit 0 if defined $pid; - } - } + BDB::db_open $DB, undef, "db", undef, BDB::BTREE, + BDB::CREATE | BDB::AUTO_COMMIT, 0666; + cf::cleanup "db_open(db): $!" if $!; + }; + cf::cleanup "db_open(db): $@" if $@; + }; - my $dirty; + my $path = cf::localdir . "/database.pst"; + if (stat $path) { + cf::sync_job { + my $pst = Storable::retrieve $path; + + cf::db_put (board => data => $pst->{board}); + cf::db_put (guildrules => data => $pst->{guildrules}); + cf::db_put (rent => balance => $pst->{rent}{balance}); + BDB::db_env_txn_checkpoint $DB_ENV; - sub db_sync() { - db_save if $dirty; - undef $dirty; + unlink $path; + }; } +} - my $idle = Event->idle (min => 10, max => 20, repeat => 0, data => WF_AUTOCANCEL, cb => sub { - db_sync; - }); - - sub db_dirty() { - $dirty = 1; - $idle->start; - } +sub db_get($$) { + my $key = "$_[0]/$_[1]"; - sub db_get($;$) { - @_ >= 2 - ? $DB->{$_[0]}{$_[1]} - : ($DB->{$_[0]} ||= { }) - } + cf::sync_job { + BDB::db_get $DB, undef, $key, my $data; - sub db_put($$;$) { - if (@_ >= 3) { - $DB->{$_[0]}{$_[1]} = $_[2]; - } else { - $DB->{$_[0]} = $_[1]; - } - db_dirty; + $! ? () + : Compress::LZF::sthaw $data } +} - cf::global->attach ( - prio => 10000, - on_cleanup => sub { - db_sync; - }, - ); +sub db_put($$$) { + BDB::dbreq_pri 4; + BDB::db_put $DB, undef, "$_[0]/$_[1]", Compress::LZF::sfreeze_cr $_[2], 0, sub { }; } ############################################################################# -# the server's main() +# the server's init and main functions + +sub load_resources { + load_regions sprintf "%s/%s/regions", cf::datadir, cf::mapdir + or die "unable to load regions file\n";#d# +} sub cfg_load { open my $fh, "<:utf8", cf::confdir . "/config" @@ -2233,6 +2273,10 @@ } } +sub init { + load_resources; +} + sub main { # we must not ever block the main coroutine local $Coro::idle = sub { @@ -2243,7 +2287,6 @@ }; cfg_load; - db_load; load_extensions; $TICK_WATCHER->start; @@ -2257,9 +2300,11 @@ BEGIN { for my $signal (qw(INT HUP TERM)) { Event->signal ( - data => WF_AUTOCANCEL, - signal => $signal, - cb => sub { + reentrant => 0, + data => WF_AUTOCANCEL, + signal => $signal, + prio => 0, + cb => sub { cf::cleanup "SIG$signal"; }, ); @@ -2290,6 +2335,10 @@ $map->save; } warn "end emergency map save\n"; + + warn "begin emergency database checkpoint\n"; + BDB::db_env_txn_checkpoint $DB_ENV; + warn "end emergency database checkpoint\n"; }; warn "leave emergency perl save\n"; @@ -2312,19 +2361,20 @@ eval { # if anything goes wrong in here, we should simply crash as we already saved - warn "syncing database to disk"; - cf::db_sync; - warn "cancelling all WF_AUTOCANCEL watchers"; for (Event::all_watchers) { $_->cancel if $_->data & WF_AUTOCANCEL; } + warn "syncing database to disk"; + BDB::db_env_txn_checkpoint $DB_ENV, 0, 0, 0, sub { }; + warn "flushing outstanding aio requests"; for (;;) { + BDB::flush; IO::AIO::flush; Coro::cede; - last unless IO::AIO::nreqs; + last unless IO::AIO::nreqs || BDB::nreqs; warn "iterate..."; } @@ -2389,7 +2439,6 @@ warn "loading config and database again"; cf::cfg_load; - cf::db_load; warn "loading extensions"; cf::load_extensions; @@ -2399,6 +2448,9 @@ warn "reattaching attachments to maps"; reattach $_ for values %MAP; + warn "loading reloadable resources"; + load_resources; + warn "restarting server ticker"; $TICK_WATCHER->start; @@ -2424,10 +2476,15 @@ # doing reload synchronously and two reloads happen back-to-back, # coro crashes during coro_state_free->destroy here. - $RELOAD_WATCHER ||= Event->timer (after => 0, data => WF_AUTOCANCEL, cb => sub { - reload; - undef $RELOAD_WATCHER; - }); + $RELOAD_WATCHER ||= Event->timer ( + reentrant => 0, + after => 0, + data => WF_AUTOCANCEL, + cb => sub { + reload; + undef $RELOAD_WATCHER; + }, + ); } }; @@ -2457,6 +2514,12 @@ $WAIT_FOR_TICK->broadcast; $WAIT_FOR_TICK_ONE->send if $WAIT_FOR_TICK_ONE->awaited; + Event::sweep; + Coro::cede_notself; + +# my $AFTER = Event::time; +# warn $AFTER - $NOW;#d# + # if we are delayed by four ticks or more, skip them all $NEXT_TICK = Event::time if Event::time >= $NEXT_TICK + $TICK * 4; @@ -2465,21 +2528,78 @@ }, ); -IO::AIO::max_poll_time $TICK * 0.2; +{ + BDB::max_poll_time $TICK * 0.1; + $BDB_POLL_WATCHER = Event->io ( + reentrant => 0, + fd => BDB::poll_fileno, + poll => 'r', + prio => 0, + data => WF_AUTOCANCEL, + cb => \&BDB::poll_cb, + ); + BDB::min_parallel 8; -$AIO_POLL_WATCHER = Event->io ( - fd => IO::AIO::poll_fileno, - poll => 'r', - prio => 5, - data => WF_AUTOCANCEL, - cb => \&IO::AIO::poll_cb, -); + BDB::set_sync_prepare { + my $status; + my $current = $Coro::current; + ( + sub { + $status = $!; + $current->ready; undef $current; + }, + sub { + Coro::schedule while defined $current; + $! = $status; + }, + ) + }; + + unless ($DB_ENV) { + $DB_ENV = BDB::db_env_create; + + cf::sync_job { + eval { + BDB::db_env_open + $DB_ENV, + $BDB_ENV_DIR, + BDB::INIT_LOCK | BDB::INIT_LOG | BDB::INIT_MPOOL | BDB::INIT_TXN + | BDB::RECOVER | BDB::REGISTER | BDB::USE_ENVIRON | BDB::CREATE, + 0666; + + cf::cleanup "db_env_open($BDB_ENV_DIR): $!" if $!; + + $DB_ENV->set_flags (BDB::AUTO_COMMIT | BDB::REGION_INIT | BDB::TXN_NOSYNC, 1); + $DB_ENV->set_lk_detect; + }; + + cf::cleanup "db_env_open(db): $@" if $@; + }; + } +} + +{ + IO::AIO::min_parallel 8; + + undef $Coro::AIO::WATCHER; + IO::AIO::max_poll_time $TICK * 0.1; + $AIO_POLL_WATCHER = Event->io ( + reentrant => 0, + fd => IO::AIO::poll_fileno, + poll => 'r', + prio => 6, + data => WF_AUTOCANCEL, + cb => \&IO::AIO::poll_cb, + ); +} $WRITE_RUNTIME_WATCHER = Event->timer ( - data => WF_AUTOCANCEL, - after => 1, - interval => 10, - cb => Coro::unblock_sub { + reentrant => 0, + data => WF_AUTOCANCEL, + after => 1, + interval => 10, + prio => 6, # keep it lowest so it acts like a watchdog + cb => Coro::unblock_sub { write_runtime or warn "ERROR: unable to write runtime file: $!"; },