Merge 82ed78e into 9a59486

docs/gdnsd-plugin-multifo.podin

@@ -56,10 +56,10 @@ RR-sets.
 
 =head1 TOP-LEVEL PLUGIN CONFIG
 
-At the top level of the plugin's configuration stanza, two special
-parameters C<up_thresh> and C<service_types> are supported. These set
-default per-resource options of the same name for any resources which do
-not define them explicitly.
+At the top level of the plugin's configuration stanza, three special
+parameters C<up_thresh>, C<service_types>, and C<ignore_health> are supported.
+These set default per-resource options of the same name for any resources which
+do not define them explicitly.
 
 The rest of the hash entries at the top level are the names of the
 resources you define.  Each resource gets a configuration hash of its own
@@ -73,8 +73,9 @@ specify a set of C<label =E<gt> address> pairs which are all the same
 family (IPv4 or IPv6), or you can use the sub-stanzas C<addrs_v4> and/or
 C<addrs_v6> to specify one or both families in the same resource.
 
-The C<up_thresh> and C<service_types> parameters are inherited through
-every level, and can be overridden at any level (even per-address-family):
+The C<up_thresh>, C<service_types>, and C<ignore_health> parameters are
+inherited through every level, and can be overridden at any level (even
+per-address-family):
 
 =over 4
 
@@ -91,16 +92,24 @@ provided, all will be monitored for each address, and the net monitored
 state will be the minimum (worst) of the set.  See L<gdnsd.config(8)> for
 more details on service_types.
 
+=item B<ignore_health>
+
+Boolean, default false.  If set to true, the health of individual addresses
+will not affect whether multifo adds them to the set of output addresses, but
+it will still be checked and used for the C<up_thresh> calculation which is
+consumed by meta-plugins like geoip and metafo, which might use that
+information to fail over to a completely different datacenter as a result.
+
 =back
 
 =head1 SHORTCUT CONFIG
 
-If you have no parameters (service_types, up_thresh) to configure in a
-given stanza (single-family direct resource config, or addrs_v[46]), and do
-not care about the descriptive per-address labels used in monitoring, you
-can replace the hash with an array of addresses.  The labels will be
-generated for you as a series of integers starting with C<1>.  For example,
-the following are equivalent:
+If you have no parameters (service_types, up_thresh, ignore_health) to
+configure in a given stanza (single-family direct resource config, or
+addrs_v[46]), and do not care about the descriptive per-address labels used in
+monitoring, you can replace the hash with an array of addresses.  The labels
+will be generated for you as a series of integers starting with C<1>.  For
+example, the following are equivalent:
 
    res1 => { addrs_v4 => [ 192.0.2.1, 192.0.2.2 ] }
    res1 => { addrs_v4 => { 1 => 192.0.2.1, 2 => 192.0.2.2 } }
@@ -116,7 +125,12 @@ determine the set of response addresses for that address family: 1) Add all
 non-DOWN addresses to the result set.  2) If the set of non-DOWN addresses
 fail the up_thresh check, add *all* addresses to the result set as a
 fallback.  3) If any address is in the DOWN state, cut the
-zonefile-specified TTL in half
+zonefile-specified TTL in half.
+
+If C<ignore_health> is true, all addresses are added to the result set
+regardless of health, but the up_thresh and TTL effects still happen, and the
+final resource-level state still reflects the overall state as it would without
+C<ignore_health>.
 
 This process is repeated independently for each of the IPv4 and IPv6
 address subsets, in the case that a resource has both address families

plugins/multifo.c

@@ -46,6 +46,7 @@ typedef struct {
     unsigned num_svcs;
     unsigned count;
     unsigned up_thresh;
+    bool ignore_health;
 } addrset_t;
 
 typedef struct {
@@ -91,6 +92,7 @@ static vscf_data_t* addrs_hash_from_array(vscf_data_t* ary, const char* resname,
 
     vscf_hash_inherit(parent, newhash, "up_thresh", false);
     vscf_hash_inherit(parent, newhash, "service_types", false);
+    vscf_hash_inherit(parent, newhash, "ignore_health", false);
     return newhash;
 }
 
@@ -177,6 +179,14 @@ static void config_addrs(const char* resname, const char* stanza, addrset_t* ase
             log_fatal("plugin_multifo: resource %s (%s): 'up_thresh' must be a floating point value in the range (0.0 - 1.0]", resname, stanza);
     }
 
+    aset->ignore_health = false;
+    vscf_data_t* ignore_health_cfg = vscf_hash_get_data_byconstkey(cfg, "ignore_health", true);
+    if(ignore_health_cfg) {
+        num_addrs--;
+        if(!vscf_is_simple(ignore_health_cfg) || !vscf_simple_get_as_bool(ignore_health_cfg, &aset->ignore_health))
+            log_fatal("plugin_multifo: resource %s (%s): 'ignore_health' must have a boolean value", resname, stanza);
+    }
+
     if(!num_addrs)
         log_fatal("plugin_multifo: resource '%s' (%s): must define one or more 'desc => IP' mappings, either directly or inside a subhash named 'addrs'", resname, stanza);
 
@@ -219,6 +229,7 @@ static void config_auto(res_t* res, const char* stanza, vscf_data_t* auto_cfg) {
     // mark parameters
     vscf_hash_get_data_byconstkey(auto_cfg, "up_thresh", true);
     vscf_hash_get_data_byconstkey(auto_cfg, "service_types", true);
+    vscf_hash_get_data_byconstkey(auto_cfg, "ignore_health", true);
 
     // clone down to just address-label keys
     vscf_data_t* auto_cfg_noparams = vscf_clone(auto_cfg, true);
@@ -265,6 +276,7 @@ static bool config_res(const char* resname, unsigned resname_len V_UNUSED, vscf_
         // inherit params downhill if applicable
         vscf_hash_bequeath_all(opts, "up_thresh", true, false);
         vscf_hash_bequeath_all(opts, "service_types", true, false);
+        vscf_hash_bequeath_all(opts, "ignore_health", true, false);
 
         addrs_v4_cfg = vscf_hash_get_data_byconstkey(opts, "addrs_v4", true);
         addrs_v6_cfg = vscf_hash_get_data_byconstkey(opts, "addrs_v6", true);
@@ -307,6 +319,8 @@ void plugin_multifo_load_config(vscf_data_t* config, const unsigned num_threads
         num_resources--;
     if(vscf_hash_bequeath_all(config, "service_types", true, false))
         num_resources--;
+    if(vscf_hash_bequeath_all(config, "ignore_health", true, false))
+        num_resources--;
 
     resources = xcalloc(num_resources, sizeof(res_t));
     unsigned residx = 0;
@@ -333,26 +347,31 @@ static gdnsd_sttl_t resolve(const gdnsd_sttl_t* sttl_tbl, const addrset_t* aset,
     dmn_assert(aset->count);
 
     gdnsd_sttl_t rv = GDNSD_STTL_TTL_MAX;
-    unsigned added = 0;
+    unsigned notdown = 0;
     for(unsigned i = 0; i < aset->count; i++) {
         const addrstate_t* as = &aset->as[i];
         const gdnsd_sttl_t as_sttl = gdnsd_sttl_min(sttl_tbl, as->indices, aset->num_svcs);
         rv = gdnsd_sttl_min2(rv, as_sttl);
         if(!(as_sttl & GDNSD_STTL_DOWN)) {
             gdnsd_result_add_anysin(result, &as->addr);
-            added++;
+            notdown++;
+        }
+        else if(aset->ignore_health) {
+            gdnsd_result_add_anysin(result, &as->addr);
         }
     }
 
     // if up_thresh was not met, signal upstream failure through rv and add all addresses
-    if(added < aset->up_thresh) {
+    if(notdown < aset->up_thresh) {
         rv |= GDNSD_STTL_DOWN;
-        if(isv6)
-            gdnsd_result_wipe_v6(result);
-        else
-            gdnsd_result_wipe_v4(result);
-        for(unsigned i = 0; i < aset->count; i++)
-            gdnsd_result_add_anysin(result, &aset->as[i].addr);
+        if(!aset->ignore_health) {
+            if(isv6)
+                gdnsd_result_wipe_v6(result);
+            else
+                gdnsd_result_wipe_v4(result);
+            for(unsigned i = 0; i < aset->count; i++)
+                gdnsd_result_add_anysin(result, &aset->as[i].addr);
+        }
     }
     // else force non-down response in retval, even if "rv" currently has the down flag from
     //   the min/min2 operations on the individual addrs

t/011upthresh/037up_thresh.t

@@ -2,7 +2,7 @@
 
 use _GDT ();
 use File::Temp qw/tmpnam/;
-use Test::More tests => 9;
+use Test::More tests => 14;
 
 # We use dns_port_2 as a custom http listener
 #  for something to monitor
@@ -72,6 +72,57 @@ _GDT->test_dns(
     ],
 );
 
+###### mmih -> metafo+multifo using "ignore_health"
+
+# All are up by default, so we get all 3x DCA IPs
+_GDT->test_dns(
+    qname => 'mmih.example.com', qtype => 'A',
+    answer => [
+        'mmih.example.com 86400 A 192.0.2.70',
+        'mmih.example.com 86400 A 192.0.2.71',
+        'mmih.example.com 86400 A 192.0.2.72',
+    ],
+);
+
+_GDT->write_statefile('admin_state', qq{
+    192.0.2.70/up => DOWN/42
+});
+_GDT->test_log_output([
+    q{admin_state: state of '192.0.2.70/up' forced to DOWN/42, real state is UP/MAX},
+]);
+
+# Marking one down doesn't fail the default 0.5 up_thresh check, but would
+# normally remove the failing IP from results here without "ignore_health".
+# Note the TTLs are still affected (this can be controlled by clamping the
+# minimum dynamic TTL at the zonefile level, if desired)
+_GDT->test_dns(
+    qname => 'mmih.example.com', qtype => 'A',
+    answer => [
+        'mmih.example.com 43200 A 192.0.2.70',
+        'mmih.example.com 43200 A 192.0.2.71',
+        'mmih.example.com 43200 A 192.0.2.72',
+    ],
+);
+
+_GDT->write_statefile('admin_state', qq{
+    192.0.2.70/up => DOWN/42
+    192.0.2.71/up => DOWN/42
+});
+_GDT->test_log_output([
+    q{admin_state: state of '192.0.2.71/up' forced to DOWN/42, real state is UP/MAX},
+]);
+
+# Now we've marked 2/3 down, which will fail the default 0.5 up_thresh, causing
+# failover to the DCB datacenter.
+_GDT->test_dns(
+    qname => 'mmih.example.com', qtype => 'A',
+    answer => [
+        'mmih.example.com 43200 A 192.0.2.80',
+        'mmih.example.com 43200 A 192.0.2.81',
+        'mmih.example.com 43200 A 192.0.2.82',
+    ],
+);
+
 _GDT->test_kill_daemon($pid);
 _GDT->test_kill_daemon($http_pid);
 

t/011upthresh/etc/config.tmpl

@@ -55,4 +55,17 @@ plugins => {
       four = [ 127.0.0.1, 10 ]
     }
   }
+  # metafo with multifo underneath using "ignore_health"
+  metafo => {
+    service_types => up
+    resources => {
+      meta_multi_ignore_health => {
+        datacenters => [ DCA, DCB ]
+        dcmap => {
+          DCA => { lb01 => 192.0.2.70, lb02 => 192.0.2.71, lb03 => 192.0.2.72, ignore_health => true }
+          DCB => { lb01 => 192.0.2.80, lb02 => 192.0.2.81, lb03 => 192.0.2.82 }
+        }
+      }
+    }
+  }
 }

t/011upthresh/etc/zones/example.com

@@ -14,3 +14,4 @@ m3dn  DYNA multifo!multi_3dead_normal
 m3dl  DYNA multifo!multi_3dead_lowthresh
 wlow  DYNA weighted!w_low
 wnorm DYNA weighted!w_norm
+mmih  DYNA metafo!meta_multi_ignore_health