Merge pull request #4522 from elliefm/v39/proc-info-for-daemons

register DAEMON and EVENTS processes so cyr_info proc can list them

Makefile.am

@@ -642,6 +642,7 @@ cunit_TESTS = \
 	cunit/msgid.testc \
 	cunit/parseaddr.testc \
 	cunit/parse.testc \
+	cunit/proc.testc \
 	cunit/procinfo.testc \
 	cunit/prot.testc \
 	cunit/ptrarray.testc \
@@ -793,6 +794,7 @@ include_HEADERS = \
 	lib/murmurhash2.h \
 	lib/nonblock.h \
 	lib/parseaddr.h \
+	lib/proc.h \
 	lib/procinfo.h \
 	lib/retry.h \
 	lib/rfc822tok.h \
@@ -1065,8 +1067,6 @@ imap_libcyrus_imap_la_SOURCES = \
 	imap/notify.h \
 	imap/partlist.c \
 	imap/partlist.h \
-	imap/proc.c \
-	imap/proc.h \
 	imap/prometheus.c \
 	imap/prometheus.h \
 	imap/protocol.h \
@@ -1087,7 +1087,6 @@ imap_libcyrus_imap_la_SOURCES = \
 	imap/search_sort.h \
 	imap/seen.h \
 	imap/seen_db.c \
-	imap/setproctitle.c \
 	imap/sievedir.c \
 	imap/sievedir.h \
 	imap/smtpclient.c \
@@ -1573,7 +1572,9 @@ lib_libcyrus_min_la_SOURCES = \
 	lib/hashu64.c \
 	lib/libconfig.c \
 	lib/mpool.c \
+	lib/proc.c \
 	lib/retry.c \
+	lib/setproctitle.c \
 	lib/smallarrayu64.c \
 	lib/strarray.c \
 	lib/strhash.c \

backup/backupd.c

@@ -56,14 +56,14 @@
 #include "lib/bsearch.h"
 #include "lib/imparse.h"
 #include "lib/map.h"
+#include "lib/proc.h"
 #include "lib/signals.h"
 #include "lib/strarray.h"
 #include "lib/util.h"
 #include "lib/xmalloc.h"
 
 #include "imap/global.h"
 #include "imap/imap_err.h"
-#include "imap/proc.h"
 #include "imap/sync_support.h"
 #include "imap/telemetry.h"
 #include "imap/tls.h"
@@ -84,6 +84,7 @@ static sasl_conn_t *backupd_saslconn = NULL;
 static int backupd_starttls_done = 0;
 static int backupd_compress_done = 0;
 static int backupd_logfd = -1;
+static struct proc_handle *proc_handle = NULL;
 
 struct open_backup {
     char *name;
@@ -159,7 +160,7 @@ EXPORTED void fatal(const char* s, int code)
 
     if (recurse_code) {
         /* We were called recursively. Just give up */
-        proc_cleanup();
+        proc_cleanup(&proc_handle);
         exit(recurse_code);
     }
     recurse_code = code;
@@ -184,7 +185,7 @@ EXPORTED int service_init(int argc __attribute__((unused)),
 {
     // FIXME should this be calling fatal? fatal exits directly
     if (geteuid() == 0) fatal("must run as the Cyrus user", EX_USAGE);
-    setproctitle_init(argc, argv, envp);
+    proc_settitle_init(argc, argv, envp);
 
     /* set signal handlers */
     signals_set_shutdown(&shut_down);
@@ -222,7 +223,7 @@ EXPORTED int service_main(int argc __attribute__((unused)),
 {
     const char *localip, *remoteip;
     sasl_security_properties_t *secprops = NULL;
-    int timeout;
+    int r, timeout;
 
     signals_poll();
 
@@ -268,7 +269,10 @@ EXPORTED int service_main(int argc __attribute__((unused)),
         tcp_disable_nagle(1); /* XXX magic fd */
     }
 
-    proc_register(config_ident, backupd_clienthost, NULL, NULL, NULL);
+    r = proc_register(&proc_handle, 0,
+                      config_ident, backupd_clienthost, NULL, NULL, NULL);
+    if (r) fatal("unable to register process", EX_IOERR);
+    proc_settitle(config_ident, backupd_clienthost, NULL, NULL, NULL);
 
     /* Set inactivity timer */
     timeout = config_getduration(IMAPOPT_SYNC_TIMEOUT, 's');
@@ -295,7 +299,7 @@ static void backupd_reset(void)
 {
     open_backups_list_close(&backupd_open_backups, 0);
 
-    proc_cleanup();
+    proc_cleanup(&proc_handle);
 
     if (backupd_in) {
         prot_NONBLOCK(backupd_in);
@@ -823,7 +827,11 @@ static void cmd_authenticate(char *mech, char *resp)
     }
 
     backupd_userid = xstrdup((const char *) val);
-    proc_register(config_ident, backupd_clienthost, backupd_userid, NULL, NULL);
+    r = proc_register(&proc_handle, 0,
+                      config_ident, backupd_clienthost, backupd_userid,
+                      NULL, NULL);
+    if (r) fatal("unable to register process", EX_IOERR);
+    proc_settitle(config_ident, backupd_clienthost, backupd_userid, NULL, NULL);
 
     syslog(LOG_NOTICE, "login: %s %s %s%s %s", backupd_clienthost, backupd_userid,
            mech, backupd_starttls_done ? "+TLS" : "", "User logged in");

cassandane/Cassandane/Cyrus/Annotator.pm

@@ -78,14 +78,14 @@ sub start_my_instances
 {
     my ($self) = @_;
 
-    $self->{instance}->add_generic_daemon(
+    $self->{instance}->add_generic_listener(
         name => 'annotator',
         port => $self->{instance}->{config}->get('annotation_callout'),
         argv => sub {
-            my ($daemon) = @_;
+            my ($listener) = @_;
             return (
                 abs_path('utils/annotator.pl'),
-                '--port', $daemon->port(),
+                '--port', $listener->port(),
                 '--pidfile', '@basedir@/run/annotator.pid',
                 );
         });

cassandane/Cassandane/Cyrus/Info.pm

@@ -40,7 +40,9 @@
 package Cassandane::Cyrus::Info;
 use strict;
 use warnings;
+use Cwd qw(realpath);
 use Data::Dumper;
+use Date::Format qw(time2str);
 
 use lib '.';
 use base qw(Cassandane::Cyrus::TestCase);
@@ -86,6 +88,11 @@ sub run_cyr_info
     my @res = readline(RESULTS);
     close RESULTS;
 
+    if ($args[0] eq 'proc') {
+        # if we see our fakesaslauthd, no we didn't
+        @res = grep { $_ !~ m/\bfakesaslauthd\b/ } @res;
+    }
+
     return @res;
 }
 
@@ -211,4 +218,182 @@ sub test_info_lint_partitions
     );
 }
 
+sub test_proc_services
+{
+    my ($self) = @_;
+
+    # no clients => no service daemons => no processes
+    my @output = $self->run_cyr_info('proc');
+    $self->assert_num_equals(0, scalar @output);
+
+    # master spawns service processes when clients connect to them
+    my $imap_svc = $self->{instance}->get_service('imap');
+    my @clients;
+    foreach (1..5) {
+        # five concurrent connections for a single user is normal,
+        # e.g. thunderbird does this
+        my $store = $imap_svc->create_store(username => 'cassandane');
+        my $imaptalk = $store->get_client();
+        push @clients, $imaptalk if $imaptalk;
+    }
+
+    # better have got some clients from that!
+    $self->assert_num_gte(1, scalar @clients);
+
+    # five clients => five service daemons => five processes
+    @output = $self->run_cyr_info('proc');
+    $self->assert_num_equals(scalar @clients, scalar @output);
+
+    # log clients out one at a time, expect proc count to decrease
+    while (scalar @clients) {
+        my $old = shift @clients;
+        $old->logout();
+
+        @output = $self->run_cyr_info('proc');
+        $self->assert_num_equals(scalar @clients, scalar @output);
+    }
+}
+
+sub test_proc_starts
+    :NoStartInstances
+{
+    my ($self) = @_;
+
+    # we used to recommend starting idled from START, and it will
+    # still work like that, so using it here saves me mocking something
+    $self->{instance}->add_start(name => 'idled',
+                                 argv => [ 'idled' ]);
+    $self->{instance}->start();
+
+    # entries listed in START run to completion before master fully
+    # starts up.  if they fork themselves and hang around (like idled
+    # does) then that's their business, but master can't and doesn't
+    # track them
+    my @output = $self->run_cyr_info('proc');
+
+    $self->assert_num_equals(0, scalar @output);
+}
+
+sub test_proc_periodic_events_slow
+    :NoStartInstances
+{
+    my ($self) = @_;
+
+    my $sleeper_time = 10; # seconds
+
+    # periodic events first fire immediately at startup, and then every
+    # 'period' minutes thereafter. the fastest we can schedule them is
+    # every 1 minute, so this test must run for at least several real
+    # minutes
+    $self->{instance}->add_event(
+        name => 'sleeper',
+        argv => [ realpath('utils/sleeper'), $sleeper_time ],
+        period => 1,
+    );
+    $self->{instance}->start();
+
+    sleep 2; # offset our checks a little to avoid races
+
+    # observe for three cycles
+    my $observations = 3;
+    while ($observations > 0) {
+        # event should have fired and be running
+        my @output = $self->run_cyr_info('proc');
+        $self->assert_num_equals(1, scalar @output);
+
+        # wait for it to finish and check again
+        sleep $sleeper_time;
+        @output = $self->run_cyr_info('proc');
+        $self->assert_num_equals(0, scalar @output);
+
+        # skip final wait if we're done
+        $observations--;
+        last if $observations == 0;
+
+        # wait until next period
+        sleep 60 - $sleeper_time;
+    }
+}
+
+sub test_proc_scheduled_events
+    :NoStartInstances
+{
+    my ($self) = @_;
+
+    my $sleeper_time = 10;
+
+    # schedule an event to fire at the next minute boundary that is at
+    # least ten seconds away
+    my $at = time + 70;
+    $at -= ($at % 60);
+    my $at_hm = time2str('%H%M', $at);
+    xlog $self, "scheduling event to run at $at_hm ($at)";
+    $self->{instance}->add_event(
+        name => 'sleeper',
+        argv => [ realpath('utils/sleeper'), $sleeper_time ],
+        at => $at_hm,
+    );
+    $self->{instance}->start();
+
+    # event process should not be running at startup
+    my @output = $self->run_cyr_info('proc');
+    $self->assert_num_equals(0, scalar @output);
+
+    # should be running at the scheduled time (with a little slop)
+    sleep 2 + $at - time;
+    @output = $self->run_cyr_info('proc');
+    $self->assert_num_equals(1, scalar @output);
+
+    # should not be running after we expect it to have finished
+    sleep $sleeper_time;
+    @output = $self->run_cyr_info('proc');
+    $self->assert_num_equals(0, scalar @output);
+}
+
+sub test_proc_daemons
+    :NoStartInstances
+{
+    my ($self) = @_;
+
+    my $sleeper_time = 10; # seconds
+    my $daemons = 3;
+
+    for my $i (1 .. $daemons) {
+        # you wouldn't usually run a daemon that exits and needs to be
+        # restarted every ten seconds, but it's useful for testing
+        # that cyr_info proc notices the pid changing
+        $self->{instance}->add_daemon(
+            name => "sleeper$i",
+            argv => [ realpath('utils/sleeper'), $sleeper_time ],
+        );
+    }
+    $self->{instance}->start();
+
+    sleep 2; # offset our checks a little to avoid races
+
+    my $observations = 3;
+    my %lastpid = map {; "sleeper$_" => 0 } (1 .. $daemons);
+    while ($observations > 0) {
+        my @output = $self->run_cyr_info('proc');
+
+        # always exactly one process per daemon
+        $self->assert_num_equals($daemons, scalar @output);
+
+        # expect a new pid for each daemon each time
+        foreach my $line (@output) {
+            my ($pid, $servicename, $host, $user, $mailbox, $cmd)
+                = split /\s/, $line, 6;
+            $self->assert_num_not_equals($lastpid{$servicename}, $pid);
+            $lastpid{$servicename} = $pid;
+        }
+
+        # skip final wait if we're done
+        $observations--;
+        last if $observations == 0;
+
+        # wait for next restart
+        sleep $sleeper_time;
+    }
+}
+
 1;

cassandane/Cassandane/Cyrus/Master.pm

@@ -845,7 +845,7 @@ sub test_maxforkrate
                               }, $self->lemming_census());
 }
 
-sub XXXtest_periodic_event
+sub test_periodic_event_slow
 {
     my ($self) = @_;
 
@@ -860,8 +860,8 @@ sub XXXtest_periodic_event
 
     xlog $self, "periodic events run immediately";
 
-    xlog $self, "waiting 5 mins for events to fire";
-    sleep(5*60);
+    xlog $self, "waiting 5 mins for events to fire, plus some slop";
+    sleep(5*60 + 5);
 
     $self->assert_deep_equals({
                                 B => { live => 0, dead => 6 },

cassandane/Cassandane/Cyrus/TestCase.pm

@@ -53,6 +53,7 @@ use base qw(Cassandane::Unit::TestCase);
 use Cassandane::Util::Log;
 use Cassandane::Util::Words;
 use Cassandane::Generator;
+use Cassandane::GenericListener;
 use Cassandane::MessageStoreFactory;
 use Cassandane::Instance;
 use Cassandane::PortManager;
@@ -1026,7 +1027,7 @@ sub post_tear_down
     }
 
     die "Found some stray processes"
-        if (Cassandane::GenericDaemon::kill_processes_on_ports(
+        if (Cassandane::GenericListener::kill_processes_on_ports(
                     Cassandane::PortManager::free_all()));
 }