add respawn handling
authorJohn Crispin <blogic@openwrt.org>
Tue, 27 Aug 2013 09:41:11 +0000 (11:41 +0200)
committerJohn Crispin <blogic@openwrt.org>
Tue, 3 Sep 2013 17:42:44 +0000 (19:42 +0200)
Signed-off-by: John Crispin <blogic@openwrt.org>
instance.c
instance.h
service.c

index d61bb332e2c7047b99f5d3d069576df63e224557..e263b84f90c64684463f64035474a1848b145023 100644 (file)
@@ -32,6 +32,7 @@ enum {
        INSTANCE_ATTR_NETDEV,
        INSTANCE_ATTR_FILE,
        INSTANCE_ATTR_TRIGGER,
+       INSTANCE_ATTR_RESPAWN,
        INSTANCE_ATTR_NICE,
        __INSTANCE_ATTR_MAX
 };
@@ -43,6 +44,7 @@ static const struct blobmsg_policy instance_attr[__INSTANCE_ATTR_MAX] = {
        [INSTANCE_ATTR_NETDEV] = { "netdev", BLOBMSG_TYPE_ARRAY },
        [INSTANCE_ATTR_FILE] = { "file", BLOBMSG_TYPE_ARRAY },
        [INSTANCE_ATTR_TRIGGER] = { "triggers", BLOBMSG_TYPE_ARRAY },
+       [INSTANCE_ATTR_RESPAWN] = { "respawn", BLOBMSG_TYPE_ARRAY },
        [INSTANCE_ATTR_NICE] = { "nice", BLOBMSG_TYPE_INT32 },
 };
 
@@ -102,6 +104,8 @@ instance_start(struct service_instance *in)
                return;
 
        in->restart = false;
+       in->halt = !in->respawn;
+
        if (!in->valid)
                return;
 
@@ -117,6 +121,7 @@ instance_start(struct service_instance *in)
 
        DEBUG(1, "Started instance %s::%s\n", in->srv->name, in->name);
        in->proc.pid = pid;
+       clock_gettime(CLOCK_MONOTONIC, &in->start);
        uloop_process_add(&in->proc);
 }
 
@@ -126,29 +131,58 @@ instance_timeout(struct uloop_timeout *t)
        struct service_instance *in;
 
        in = container_of(t, struct service_instance, timeout);
-       kill(in->proc.pid, SIGKILL);
-       uloop_process_delete(&in->proc);
-       in->proc.cb(&in->proc, -1);
+
+       if (!in->halt && (in->restart || in->respawn))
+               instance_start(in);
 }
 
 static void
 instance_exit(struct uloop_process *p, int ret)
 {
        struct service_instance *in;
+       struct timespec tp;
+       long runtime;
 
        in = container_of(p, struct service_instance, proc);
-       DEBUG(1, "Instance %s::%s exit with error code %d\n", in->srv->name, in->name, ret);
+
+       clock_gettime(CLOCK_MONOTONIC, &tp);
+       runtime = tp.tv_sec - in->start.tv_sec;
+
+       DEBUG(1, "Instance %s::%s exit with error code %d after %ld seconds\n", in->srv->name, in->name, ret, runtime);
        uloop_timeout_cancel(&in->timeout);
-       if (in->restart)
+       if (in->halt) {
+               /* no action */
+       } else if (in->restart) {
                instance_start(in);
+       } else if (in->respawn) {
+               if (runtime < RESPAWN_ERROR)
+                       in->respawn_count++;
+               else
+                       in->respawn_count = 0;
+               if (in->respawn_count > 5)
+                       DEBUG(1, "Instance %s::%s s in a crash loop %d crashes, %ld seconds since last crash\n",
+                                                               in->srv->name, in->name, in->respawn_count, runtime);
+               uloop_timeout_set(&in->timeout, 5000);
+       }
 }
 
 void
-instance_stop(struct service_instance *in, bool restart)
+instance_stop(struct service_instance *in)
 {
        if (!in->proc.pending)
                return;
+       in->halt = true;
+       in->restart = in->respawn = false;
+       kill(in->proc.pid, SIGTERM);
+}
 
+static void
+instance_restart(struct service_instance *in)
+{
+       if (!in->proc.pending)
+               return;
+       in->halt = false;
+       in->restart = true;
        kill(in->proc.pid, SIGTERM);
 }
 
@@ -348,9 +382,9 @@ instance_update(struct service_instance *in, struct service_instance *in_new)
                        instance_config_move(in, in_new);
                instance_start(in);
        } else {
-               in->restart = true;
-               instance_stop(in, true);
+               instance_restart(in);
                instance_config_move(in, in_new);
+               /* restart happens in the child callback handler */
        }
        return true;
 }
@@ -375,6 +409,8 @@ instance_init(struct service_instance *in, struct service *s, struct blob_attr *
        in->config = config;
        in->timeout.cb = instance_timeout;
        in->proc.cb = instance_exit;
+       in->respawn = true;
+       in->respawn_count = 0;
 
        blobmsg_list_init(&in->netdev, struct instance_netdev, node, instance_netdev_cmp);
        blobmsg_list_init(&in->file, struct instance_file, node, instance_file_cmp);
index ceae8346b888a7f8aa01c5fb784f78737be11a4c..1c8c0a081cecbab7e24d0e2060d54fd3659cffa0 100644 (file)
@@ -19,6 +19,8 @@
 #include <libubox/uloop.h>
 #include "utils.h"
 
+#define RESPAWN_ERROR  (5 * 60)
+
 struct service_instance {
        struct vlist_node node;
        struct service *srv;
@@ -26,7 +28,16 @@ struct service_instance {
 
        int8_t nice;
        bool valid;
+
+       bool halt;
        bool restart;
+       bool respawn;
+       int respawn_count;
+       struct timespec start;
+
+       int respawn_timeout;
+       int respawn_threshold;
+       int respawn_retry;
 
        struct blob_attr *config;
        struct uloop_process proc;
@@ -41,7 +52,7 @@ struct service_instance {
 };
 
 void instance_start(struct service_instance *in);
-void instance_stop(struct service_instance *in, bool restart);
+void instance_stop(struct service_instance *in);
 bool instance_update(struct service_instance *in, struct service_instance *in_new);
 void instance_init(struct service_instance *in, struct service *s, struct blob_attr *config);
 void instance_free(struct service_instance *in);
index 561c76c5de8e8a86b8bf8a2fa32e407b6e4f371c..c80402c809aa854bc314071693f5a1b1b82e2eb8 100644 (file)
--- a/service.c
+++ b/service.c
@@ -55,7 +55,7 @@ service_instance_update(struct vlist_tree *tree, struct vlist_node *node_new,
                instance_free(in_n);
        } else if (in_o) {
                DEBUG(1, "Free instance %s::%s\n", in_o->srv->name, in_o->name);
-               instance_stop(in_o, false);
+               instance_stop(in_o);
                instance_free(in_o);
        } else if (in_n) {
                DEBUG(1, "Create instance %s::%s\n", in_n->srv->name, in_n->name);