New function instance_failure().

activitypub.c

@@ -2261,6 +2261,9 @@ int process_input_message(snac *snac, const xs_dict *msg, const xs_dict *req)
         return -1;
     }
 
+    /* this instance is alive */
+    instance_failure(actor, 2);
+
     /* question votes may not have a type */
     if (xs_is_null(type))
         type = "Note";
@@ -3152,6 +3155,11 @@ void process_queue_item(xs_dict *q_item)
             return;
         }
 
+        if (instance_failure(inbox, 0)) {
+            srv_debug(1, xs_fmt("too many failures for instance %s", inbox));
+            return;
+        }
+
         /* deliver (if previous error status was a timeout, try now longer) */
         if (p_status == 599)
             timeout = xs_number_get(xs_dict_get_def(srv_config, "queue_timeout_2", "8"));
@@ -3163,6 +3171,9 @@ void process_queue_item(xs_dict *q_item)
 
         status = send_to_inbox_raw(keyid, seckey, inbox, msg, &payload, &p_size, timeout);
 
+        /* register or clear a value for this instance */
+        instance_failure(inbox, valid_status(status) ? 2 : 1);
+
         if (payload) {
             if (p_size > 64) {
                 /* trim the message */

data.c

@@ -112,6 +112,9 @@ int srv_open(const char *basedir, int auto_upgrade)
     xs *tmpdir = xs_fmt("%s/tmp", srv_basedir);
     mkdirx(tmpdir);
 
+    xs *faildir = xs_fmt("%s/failure", srv_basedir);
+    mkdirx(faildir);
+
 #ifdef __APPLE__
 /* Apple uses st_atimespec instead of st_atim etc */
 #define st_atim st_atimespec
@@ -3039,6 +3042,56 @@ xs_list *content_search(snac *user, const char *regex,
 }
 
 
+int instance_failure(const char *url, int op)
+/* do some checks and accounting on instance failures */
+{
+    int ret = 0;
+    xs *l = xs_split(url, "/");
+    const char *hostname = xs_list_get(l, 2);
+    double mt;
+
+    if (!xs_is_string(hostname))
+        return 0;
+
+    xs *md5 = xs_md5_hex(hostname, strlen(hostname));
+    xs *fn = xs_fmt("%s/%s", srv_basedir, md5);
+
+    switch (op) {
+    case 0: /** check **/
+        if ((mt = mtime(fn)) != 0.0) {
+            /* grace time */
+            double seconds_failing = 30 * (24 * 60 * 60);
+
+            if ((double)time(NULL) - mt > seconds_failing)
+                ret = -1;
+        }
+
+        break;
+
+    case 1: /** register a failure **/
+        if (mtime(fn) == 0.0) {
+            FILE *f;
+
+            /* only create once, as the date will be used */
+            if ((f = fopen(fn, "w")) != NULL) {
+                fprintf(f, "%s\n", hostname);
+                fclose(f);
+            }
+        }
+
+        break;
+
+    case 2: /** clear a failure **/
+        /* called whenever a message comes from this instance */
+        unlink(fn);
+
+        break;
+    }
+
+    return ret;
+}
+
+
 /** notifications **/
 
 xs_str *notify_check_time(snac *snac, int reset)

snac.h

@@ -278,6 +278,8 @@ int content_match(const char *file, const xs_dict *msg);
 xs_list *content_search(snac *user, const char *regex,
             int priv, int skip, int show, int max_secs, int *timeout);
 
+int instance_failure(const char *url, int op);
+
 void enqueue_input(snac *snac, const xs_dict *msg, const xs_dict *req, int retries);
 void enqueue_shared_input(const xs_dict *msg, const xs_dict *req, int retries);
 void enqueue_output_raw(const char *keyid, const char *seckey,