X-Git-Url: https://git.wh0rd.org/?a=blobdiff_plain;f=update_daemon2.php;h=18d8936b5ed843f83657cd3d74c5a916d01805e0;hb=9a57512c7099fdc91081f7879b155b3cece2e7c7;hp=cf25eeac205e0ede4ef5c1388369b444b15c4778;hpb=abfa57fdf00894a07f8c268c799ce9b225e5bc5b;p=tt-rss.git diff --git a/update_daemon2.php b/update_daemon2.php old mode 100644 new mode 100755 index cf25eeac..18d8936b --- a/update_daemon2.php +++ b/update_daemon2.php @@ -1,19 +1,16 @@ -#!/usr/bin/php +#!/usr/bin/env php MAX_CHILD_RUNTIME) { + _debug("[MASTER] child process $pid seems to be stuck, aborting..."); + posix_kill($pid, SIGKILL); + } + } } function sigchld_handler($signal) { @@ -72,12 +88,28 @@ pcntl_waitpid(-1, $status, WNOHANG); } + function shutdown() { + if (file_exists(LOCK_DIRECTORY . "/update_daemon.lock")) + unlink(LOCK_DIRECTORY . "/update_daemon.lock"); + } + + function task_shutdown() { + $pid = posix_getpid(); + + if (file_exists(LOCK_DIRECTORY . "/update_daemon-$pid.lock")) + unlink(LOCK_DIRECTORY . "/update_daemon-$pid.lock"); + } + function sigint_handler() { - unlink(LOCK_DIRECTORY . "/update_daemon.lock"); + shutdown(); + die("[SIGINT] removing lockfile and exiting.\n"); + } + + function task_sigint_handler() { + task_shutdown(); die("[SIGINT] removing lockfile and exiting.\n"); } - pcntl_signal(SIGALRM, 'sigalrm_handler'); pcntl_signal(SIGCHLD, 'sigchld_handler'); if (file_is_locked("update_daemon.lock")) { @@ -87,7 +119,9 @@ if (!pcntl_fork()) { pcntl_signal(SIGINT, 'sigint_handler'); + register_shutdown_function('shutdown'); + // Try to lock a file in order to avoid concurrent update. $lock_handle = make_lockfile("update_daemon.lock"); if (!$lock_handle) { @@ -100,21 +134,16 @@ // Testing database connection. // It is unnecessary to start the fork loop if database is not ok. - $link = db_connect(DB_HOST, DB_USER, DB_PASS, DB_NAME); + $link = db_connect(DB_HOST, DB_USER, DB_PASS, DB_NAME); - if (!$link) { - if (DB_TYPE == "mysql") { - print mysql_error(); - } - // PG seems to display its own errors just fine by default. - return; - } + if (!init_connection($link)) return; db_close($link); - while (true) { + // Since sleep is interupted by SIGCHLD, we need another way to + // respect the SPAWN_INTERVAL $next_spawn = $last_checkpoint + SPAWN_INTERVAL - time(); if ($next_spawn % 10 == 0) { @@ -124,6 +153,7 @@ if ($last_checkpoint + SPAWN_INTERVAL < time()) { + check_ctimes(); reap_children(); for ($j = count($children); $j < MAX_JOBS; $j++) { @@ -133,151 +163,68 @@ } else if ($pid) { _debug("[MASTER] spawned client $j [PID:$pid]..."); array_push($children, $pid); + $ctimes[$pid] = time(); } else { pcntl_signal(SIGCHLD, SIG_IGN); - pcntl_signal(SIGINT, SIG_DFL); + pcntl_signal(SIGINT, 'task_sigint_handler'); + + register_shutdown_function('task_shutdown'); + + $my_pid = posix_getpid(); + $lock_filename = "update_daemon-$my_pid.lock"; + + $lock_handle = make_lockfile($lock_filename); + + if (!$lock_handle) { + die("error: Can't create lockfile ($lock_filename). ". + "Maybe another daemon is already running.\n"); + } // ****** Updating RSS code ******* // Only run in fork process. $start_timestamp = time(); - $link = db_connect(DB_HOST, DB_USER, DB_PASS, DB_NAME); + $link = db_connect(DB_HOST, DB_USER, DB_PASS, DB_NAME); - if (!$link) { - if (DB_TYPE == "mysql") { - print mysql_error(); - } - // PG seems to display its own errors just fine by default. - return; - } - - if (DB_TYPE == "pgsql") { - pg_query("set client_encoding = 'utf-8'"); - pg_set_client_encoding("UNICODE"); - } else { - if (defined('MYSQL_CHARSET') && MYSQL_CHARSET) { - db_query($link, "SET NAMES " . MYSQL_CHARSET); - // db_query($link, "SET CHARACTER SET " . MYSQL_CHARSET); - } - } + if (!init_connection($link)) return; // We disable stamp file, since it is of no use in a multiprocess update. // not really, tho for the time being -fox if (!make_stampfile('update_daemon.stamp')) { print "warning: unable to create stampfile"; - } - - // $last_purge = 0; - - // if (time() - $last_purge > PURGE_INTERVAL) { - - // FIXME : $last_purge is of no use in a multiprocess update. - // FIXME : We ALWAYS purge old posts. - _debug("Purging old posts (random 30 feeds)..."); - global_purge_old_posts($link, true, 30); - - // $last_purge = time(); - // } - - // Process all other feeds using last_updated and interval parameters - - $random_qpart = sql_random_function(); - - if (DAEMON_UPDATE_LOGIN_LIMIT > 0) { - if (DB_TYPE == "pgsql") { - $login_thresh_qpart = "AND ttrss_users.last_login >= NOW() - INTERVAL '".DAEMON_UPDATE_LOGIN_LIMIT." days'"; - } else { - $login_thresh_qpart = "AND ttrss_users.last_login >= DATE_SUB(NOW(), INTERVAL ".DAEMON_UPDATE_LOGIN_LIMIT." DAY)"; - } - } else { - $login_thresh_qpart = ""; - } - - //if (DB_TYPE == "pgsql") { - // $update_limit_qpart = "AND ttrss_feeds.last_updated < NOW() - INTERVAL '".(DAEMON_SLEEP_INTERVAL*2)." seconds'"; - //} else { - // $update_limit_qpart = "AND ttrss_feeds.last_updated < DATE_SUB(NOW(), INTERVAL ".(DAEMON_SLEEP_INTERVAL*2)." SECOND)"; - //} - - if (DB_TYPE == "pgsql") { - $update_limit_qpart = "AND ttrss_feeds.last_updated < NOW() - INTERVAL '".(DAEMON_SLEEP_INTERVAL*2)." seconds'"; - $update_limit_qpart = "AND (( - ttrss_feeds.update_interval = 0 - AND ttrss_feeds.last_updated < NOW() - INTERVAL ttrss_user_prefs.value || ' minutes' - ) OR ( - ttrss_feeds.update_interval > 0 - AND ttrss_feeds.last_updated < NOW() - INTERVAL ttrss_feeds.update_interval || ' minutes' - ))"; - } else { - $update_limit_qpart = "AND (( - ttrss_feeds.update_interval = 0 - AND ttrss_feeds.last_updated < DATE_SUB(NOW(), INTERVAL CONVERT(ttrss_user_prefs.value, SIGNED INTEGER) MINUTE) - ) OR ( - ttrss_feeds.update_interval > 0 - AND ttrss_feeds.last_updated < DATE_SUB(NOW(), INTERVAL ttrss_feeds.update_interval MINUTE) - ))"; } + // Call to the feed batch update function + // or regenerate feedbrowser cache - if (DB_TYPE == "pgsql") { - $updstart_thresh_qpart = "AND (ttrss_feeds.last_update_started IS NULL OR ttrss_feeds.last_update_started < NOW() - INTERVAL '120 seconds')"; + if (rand(0,100) > 30) { + update_daemon_common($link); } else { - $updstart_thresh_qpart = "AND (ttrss_feeds.last_update_started IS NULL OR ttrss_feeds.last_update_started < DATE_SUB(NOW(), INTERVAL 120 SECOND))"; - } + $count = update_feedbrowser_cache($link); + _debug("Feedbrowser updated, $count feeds processed."); - $result = db_query($link, "SELECT ttrss_feeds.feed_url,ttrss_feeds.id, ttrss_feeds.owner_uid, - SUBSTRING(ttrss_feeds.last_updated,1,19) AS last_updated, - ttrss_feeds.update_interval - FROM - ttrss_feeds, ttrss_users, ttrss_user_prefs - WHERE - ttrss_feeds.owner_uid = ttrss_users.id - AND ttrss_users.id = ttrss_user_prefs.owner_uid - AND ttrss_user_prefs.pref_name='DEFAULT_UPDATE_INTERVAL' - $login_thresh_qpart $update_limit_qpart - $updstart_thresh_qpart - ORDER BY $random_qpart DESC LIMIT " . DAEMON_FEED_LIMIT); - - $user_prefs_cache = array(); - - _debug(sprintf("Scheduled %d feeds to update...\n", db_num_rows($result))); - - // Here is a little cache magic in order to minimize risk of double feed updates. - $feeds_to_update = array(); - while ($line = db_fetch_assoc($result)) { - $feeds_to_update[$line['id']] = $line; - } - - // We update the feed last update started date before anything else. - // There is no lag due to feed contents downloads - // It prevent an other process to update the same feed. - $feed_ids = array_keys($feeds_to_update); - if($feed_ids) { - db_query($link, sprintf("UPDATE ttrss_feeds SET last_update_started = NOW() - WHERE id IN (%s)", implode(',', $feed_ids))); - } + purge_orphans($link, true); - while ($line = array_pop($feeds_to_update)) { + $rc = cleanup_tags($link, 14, 50000); - _debug("Feed: " . $line["feed_url"] . ", " . $line["last_updated"]); + _debug("Cleaned $rc cached tags."); - pcntl_alarm(300); - update_rss_feed($link, $line["feed_url"], $line["id"], true); - pcntl_alarm(0); + _debug("Updating linked feeds..."); + get_linked_feeds($link); - sleep(1); // prevent flood (FIXME make this an option?) } - if (DAEMON_SENDS_DIGESTS) send_headlines_digests($link); - _debug("Elapsed time: " . (time() - $start_timestamp) . " second(s)"); - + db_close($link); // We are in a fork. // We wait a little before exiting to avoid to be faster than our parent process. sleep(1); + + unlink(LOCK_DIRECTORY . "/$lock_filename"); + // We exit in order to avoid fork bombing. exit(0); }