chiark / gitweb /
journal: Add sync timer to journal server
authorOleksii Shevchuk <alxchk@gmail.com>
Mon, 25 Mar 2013 16:49:03 +0000 (18:49 +0200)
committerLennart Poettering <lennart@poettering.net>
Mon, 25 Mar 2013 16:51:06 +0000 (17:51 +0100)
Add option to force journal sync with fsync. Default timeout is 5min.
Interval configured via SyncIntervalSec option at journal.conf. Synced
journal files will be marked as OFFLINE.

Manual sync can be performed via sending SIGUSR1.

man/journald.conf.xml
src/journal/journal-file.c
src/journal/journal-file.h
src/journal/journald-gperf.gperf
src/journal/journald-server.c
src/journal/journald-server.h
src/journal/journald.conf

index 0797deb..0b9de65 100644 (file)
                                 seconds. </para></listitem>
                         </varlistentry>
 
+
+                        <varlistentry>
+                                <term><varname>SyncIntervalSec=</varname></term>
+
+                                <listitem><para>The timeout before syncing journal
+                                data to disk. After syncing journal files have
+                                OFFLINE state. Default timeout is 5 minutes.
+                                </para></listitem>
+                        </varlistentry>
+
                         <varlistentry>
                                 <term><varname>ForwardToSyslog=</varname></term>
                                 <term><varname>ForwardToKMsg=</varname></term>
index 13fc8ed..5b077be 100644 (file)
 /* How many entries to keep in the entry array chain cache at max */
 #define CHAIN_CACHE_MAX 20
 
+int journal_file_set_online(JournalFile *f) {
+        assert(f);
+
+        if (!f->writable)
+                return -EPERM;
+
+        if (!(f->fd >= 0 && f->header))
+                return -EINVAL;
+
+        switch(f->header->state) {
+                case STATE_ONLINE:
+                        return 0;
+
+                case STATE_OFFLINE:
+                        f->header->state = STATE_ONLINE;
+                        fsync(f->fd);
+                        return 0;
+
+                default:
+                        return -EINVAL;
+        }
+}
+
+int journal_file_set_offline(JournalFile *f) {
+        assert(f);
+
+        if (!f->writable)
+                return -EPERM;
+
+        if (!(f->fd >= 0 && f->header))
+                return -EINVAL;
+
+        if (f->header->state != STATE_ONLINE)
+                return 0;
+
+        fsync(f->fd);
+
+        f->header->state = STATE_OFFLINE;
+
+        fsync(f->fd);
+
+        return 0;
+}
+
 void journal_file_close(JournalFile *f) {
         assert(f);
 
@@ -81,16 +125,10 @@ void journal_file_close(JournalFile *f) {
         if (f->mmap && f->fd >= 0)
                 mmap_cache_close_fd(f->mmap, f->fd);
 
-        if (f->writable && f->fd >= 0)
-                fdatasync(f->fd);
-
-        if (f->header) {
-                /* Mark the file offline. Don't override the archived state if it already is set */
-                if (f->writable && f->header->state == STATE_ONLINE)
-                        f->header->state = STATE_OFFLINE;
+        journal_file_set_offline(f);
 
+        if (f->header)
                 munmap(f->header, PAGE_ALIGN(sizeof(Header)));
-        }
 
         if (f->fd >= 0)
                 close_nointr_nofail(f->fd);
@@ -177,7 +215,7 @@ static int journal_file_refresh_header(JournalFile *f) {
 
         f->header->boot_id = boot_id;
 
-        f->header->state = STATE_ONLINE;
+        journal_file_set_online(f);
 
         /* Sync the online state to disk */
         msync(f->header, PAGE_ALIGN(sizeof(Header)), MS_SYNC);
@@ -457,6 +495,10 @@ int journal_file_append_object(JournalFile *f, int type, uint64_t size, Object *
         assert(offset);
         assert(ret);
 
+        r = journal_file_set_online(f);
+        if (r < 0)
+                return r;
+
         p = le64toh(f->header->tail_object_offset);
         if (p == 0)
                 p = le64toh(f->header->header_size);
@@ -1267,9 +1309,6 @@ int journal_file_append_entry(JournalFile *f, const dual_timestamp *ts, const st
         assert(f);
         assert(iovec || n_iovec == 0);
 
-        if (!f->writable)
-                return -EPERM;
-
         if (!ts) {
                 dual_timestamp_get(&_ts);
                 ts = &_ts;
index cdbc8e4..0eab501 100644 (file)
@@ -106,6 +106,8 @@ int journal_file_open(
                 JournalFile *template,
                 JournalFile **ret);
 
+int journal_file_set_offline(JournalFile *f);
+int journal_file_set_online(JournalFile *f);
 void journal_file_close(JournalFile *j);
 
 int journal_file_open_reliably(
index 1baef14..57b45f9 100644 (file)
@@ -18,6 +18,7 @@ struct ConfigPerfItem;
 Journal.Storage,            config_parse_storage,   0, offsetof(Server, storage)
 Journal.Compress,           config_parse_bool,      0, offsetof(Server, compress)
 Journal.Seal,               config_parse_bool,      0, offsetof(Server, seal)
+Journal.SyncIntervalSec,    config_parse_usec,      0, offsetof(Server, sync_interval_usec)
 Journal.RateLimitInterval,  config_parse_usec,      0, offsetof(Server, rate_limit_interval)
 Journal.RateLimitBurst,     config_parse_unsigned,  0, offsetof(Server, rate_limit_burst)
 Journal.SystemMaxUse,       config_parse_bytes_off, 0, offsetof(Server, system_metrics.max_use)
index 855430a..a9d7aa1 100644 (file)
@@ -24,6 +24,7 @@
 #include <linux/sockios.h>
 #include <sys/statvfs.h>
 #include <sys/mman.h>
+#include <sys/timerfd.h>
 
 #include <libudev.h>
 #include <systemd/sd-journal.h>
@@ -67,6 +68,7 @@
 
 #define USER_JOURNALS_MAX 1024
 
+#define DEFAULT_SYNC_INTERVAL_USEC (5*USEC_PER_MINUTE)
 #define DEFAULT_RATE_LIMIT_INTERVAL (10*USEC_PER_SEC)
 #define DEFAULT_RATE_LIMIT_BURST 200
 
@@ -344,6 +346,33 @@ void server_rotate(Server *s) {
         }
 }
 
+void server_sync(Server *s) {
+        JournalFile *f;
+        void *k;
+        Iterator i;
+        int r;
+
+        static const struct itimerspec sync_timer_disable = {};
+
+        if (s->system_journal) {
+                r = journal_file_set_offline(s->system_journal);
+                if (r < 0)
+                        log_error("Failed to sync system journal: %s", strerror(-r));
+        }
+
+        HASHMAP_FOREACH_KEY(f, k, s->user_journals, i) {
+                r = journal_file_set_offline(f);
+                if (r < 0)
+                        log_error("Failed to sync user journal: %s", strerror(-r));
+        }
+
+        r = timerfd_settime(s->sync_timer_fd, 0, &sync_timer_disable, NULL);
+        if (r < 0)
+                log_error("Failed to disable max timer: %m");
+
+        s->sync_scheduled = false;
+}
+
 void server_vacuum(Server *s) {
         char *p;
         char ids[33];
@@ -475,8 +504,10 @@ static void write_to_journal(Server *s, uid_t uid, struct iovec *iovec, unsigned
         }
 
         r = journal_file_append_entry(f, NULL, iovec, n, &s->seqnum, NULL, NULL);
-        if (r >= 0)
+        if (r >= 0) {
+                server_schedule_sync(s);
                 return;
+        }
 
         if (vacuumed || !shall_try_append_again(f, r)) {
                 log_error("Failed to write entry, ignoring: %s", strerror(-r));
@@ -990,11 +1021,10 @@ int process_event(Server *s, struct epoll_event *ev) {
                         return -errno;
                 }
 
-                log_info("Received SIG%s", signal_to_string(sfsi.ssi_signo));
-
                 if (sfsi.ssi_signo == SIGUSR1) {
                         touch("/run/systemd/journal/flushed");
                         server_flush_to_var(s);
+                        server_sync(s);
                         return 1;
                 }
 
@@ -1004,8 +1034,23 @@ int process_event(Server *s, struct epoll_event *ev) {
                         return 1;
                 }
 
+                log_info("Received SIG%s", signal_to_string(sfsi.ssi_signo));
+
                 return 0;
 
+        } else if (ev->data.fd == s->sync_timer_fd) {
+                int r;
+                uint64_t t;
+
+                log_debug("Got sync request from epoll.");
+
+                r = read(ev->data.fd, (void *)&t, sizeof(t));
+                if (r < 0)
+                        return 0;
+
+                server_sync(s);
+                return 1;
+
         } else if (ev->data.fd == s->dev_kmsg_fd) {
                 int r;
 
@@ -1285,16 +1330,68 @@ static int server_parse_config_file(Server *s) {
         return r;
 }
 
+static int server_open_sync_timer(Server *s) {
+        int r;
+        struct epoll_event ev;
+
+        assert(s);
+
+        s->sync_timer_fd = timerfd_create(CLOCK_MONOTONIC, TFD_CLOEXEC);
+        if (s->sync_timer_fd < 0)
+                return -errno;
+
+        zero(ev);
+        ev.events = EPOLLIN;
+        ev.data.fd = s->sync_timer_fd;
+
+        r = epoll_ctl(s->epoll_fd, EPOLL_CTL_ADD, s->sync_timer_fd, &ev);
+        if (r < 0) {
+                log_error("Failed to add idle timer fd to epoll object: %m");
+                return -errno;
+        }
+
+        return 0;
+}
+
+int server_schedule_sync(Server *s) {
+        int r;
+
+        struct itimerspec sync_timer_enable;
+
+        assert(s);
+
+        if (s->sync_scheduled)
+                return 0;
+
+        if (s->sync_interval_usec) {
+                zero(sync_timer_enable);
+                sync_timer_enable.it_value.tv_sec = s->sync_interval_usec / USEC_PER_SEC;
+                sync_timer_enable.it_value.tv_nsec = s->sync_interval_usec % MSEC_PER_SEC;
+
+                r = timerfd_settime(s->sync_timer_fd, 0, &sync_timer_enable, NULL);
+                if (r < 0)
+                        return -errno;
+        }
+
+        s->sync_scheduled = true;
+
+        return 0;
+}
+
 int server_init(Server *s) {
         int n, r, fd;
 
         assert(s);
 
         zero(*s);
-        s->syslog_fd = s->native_fd = s->stdout_fd = s->signal_fd = s->epoll_fd = s->dev_kmsg_fd = -1;
+        s->sync_timer_fd = s->syslog_fd = s->native_fd = s->stdout_fd =
+            s->signal_fd = s->epoll_fd = s->dev_kmsg_fd = -1;
         s->compress = true;
         s->seal = true;
 
+        s->sync_interval_usec = DEFAULT_SYNC_INTERVAL_USEC;
+        s->sync_scheduled = false;
+
         s->rate_limit_interval = DEFAULT_RATE_LIMIT_INTERVAL;
         s->rate_limit_burst = DEFAULT_RATE_LIMIT_BURST;
 
@@ -1394,6 +1491,10 @@ int server_init(Server *s) {
         if (r < 0)
                 return r;
 
+        r = server_open_sync_timer(s);
+        if (r < 0)
+                return r;
+
         r = open_signalfd(s);
         if (r < 0)
                 return r;
@@ -1466,6 +1567,9 @@ void server_done(Server *s) {
         if (s->dev_kmsg_fd >= 0)
                 close_nointr_nofail(s->dev_kmsg_fd);
 
+        if (s->sync_timer_fd >= 0)
+                close_nointr_nofail(s->sync_timer_fd);
+
         if (s->rate_limit)
                 journal_rate_limit_free(s->rate_limit);
 
index 9f50a29..21edd6b 100644 (file)
@@ -71,6 +71,7 @@ typedef struct Server {
         size_t buffer_size;
 
         JournalRateLimit *rate_limit;
+        usec_t sync_interval_usec;
         usec_t rate_limit_interval;
         unsigned rate_limit_burst;
 
@@ -119,6 +120,9 @@ typedef struct Server {
         uint64_t *kernel_seqnum;
 
         struct udev *udev;
+
+        int sync_timer_fd;
+        bool sync_scheduled;
 } Server;
 
 #define N_IOVEC_META_FIELDS 17
@@ -145,8 +149,10 @@ void server_fix_perms(Server *s, JournalFile *f, uid_t uid);
 bool shall_try_append_again(JournalFile *f, int r);
 int server_init(Server *s);
 void server_done(Server *s);
+void server_sync(Server *s);
 void server_vacuum(Server *s);
 void server_rotate(Server *s);
+int server_schedule_sync(Server *s);
 int server_flush_to_var(Server *s);
 int process_event(Server *s, struct epoll_event *ev);
 void server_maybe_append_tags(Server *s);
index 948318b..5410477 100644 (file)
@@ -12,6 +12,7 @@
 #Compress=yes
 #Seal=yes
 #SplitMode=login
+#SyncIntervalSec=5m
 #RateLimitInterval=10s
 #RateLimitBurst=200
 #SystemMaxUse=