Updated backoff to spec.

5 files changed, 58 insertions, 34 deletions

diff --git a/src/core/ext/client_channel/subchannel.c b/src/core/ext/client_channel/subchannel.c
index a148b2a0e1..3e8c93ee45 100644
--- a/src/core/ext/client_channel/subchannel.c
+++ b/src/core/ext/client_channel/subchannel.c
@@ -334,16 +334,18 @@ grpc_subchannel *grpc_subchannel_create(grpc_exec_ctx *exec_ctx,
   int initial_backoff_ms =
       GRPC_SUBCHANNEL_INITIAL_CONNECT_BACKOFF_SECONDS * 1000;
   int max_backoff_ms = GRPC_SUBCHANNEL_RECONNECT_MAX_BACKOFF_SECONDS * 1000;
+  int min_backoff_ms = GRPC_SUBCHANNEL_MIN_CONNECT_TIMEOUT_SECONDS * 1000;
   bool fixed_reconnect_backoff = false;
   if (c->args) {
     for (size_t i = 0; i < c->args->num_args; i++) {
       if (0 == strcmp(c->args->args[i].key,
-                      "grpc.testing.fixed_reconnect_backoff")) {
+                      "grpc.testing.fixed_reconnect_backoff_ms")) {
         GPR_ASSERT(c->args->args[i].type == GRPC_ARG_INTEGER);
         fixed_reconnect_backoff = true;
-        initial_backoff_ms = max_backoff_ms = grpc_channel_arg_get_integer(
-            &c->args->args[i],
-            (grpc_integer_options){initial_backoff_ms, 100, INT_MAX});
+        initial_backoff_ms = min_backoff_ms = max_backoff_ms =
+            grpc_channel_arg_get_integer(
+                &c->args->args[i],
+                (grpc_integer_options){initial_backoff_ms, 100, INT_MAX});
       } else if (0 == strcmp(c->args->args[i].key,
                              GRPC_ARG_MAX_RECONNECT_BACKOFF_MS)) {
         fixed_reconnect_backoff = false;
@@ -360,11 +362,11 @@ grpc_subchannel *grpc_subchannel_create(grpc_exec_ctx *exec_ctx,
     }
   }
   gpr_backoff_init(
-      &c->backoff_state,
+      &c->backoff_state, initial_backoff_ms,
       fixed_reconnect_backoff ? 1.0
                               : GRPC_SUBCHANNEL_RECONNECT_BACKOFF_MULTIPLIER,
       fixed_reconnect_backoff ? 0.0 : GRPC_SUBCHANNEL_RECONNECT_JITTER,
-      initial_backoff_ms, max_backoff_ms);
+      min_backoff_ms, max_backoff_ms);
   gpr_mu_init(&c->mu);
 
   return grpc_subchannel_index_register(exec_ctx, key, c);
diff --git a/src/core/ext/lb_policy/grpclb/grpclb.c b/src/core/ext/lb_policy/grpclb/grpclb.c
index 4262d2b9a4..df0db61c22 100644
--- a/src/core/ext/lb_policy/grpclb/grpclb.c
+++ b/src/core/ext/lb_policy/grpclb/grpclb.c
@@ -123,10 +123,11 @@
 #include "src/core/lib/surface/channel.h"
 #include "src/core/lib/transport/static_metadata.h"
 
-#define BACKOFF_MULTIPLIER 1.6
-#define BACKOFF_JITTER 0.2
-#define BACKOFF_MIN_SECONDS 10
-#define BACKOFF_MAX_SECONDS 60
+#define GRPC_GRPCLB_MIN_CONNECT_TIMEOUT_SECONDS 20
+#define GRPC_GRPCLB_INITIAL_CONNECT_BACKOFF_SECONDS 1
+#define GRPC_GRPCLB_RECONNECT_BACKOFF_MULTIPLIER 1.6
+#define GRPC_GRPCLB_RECONNECT_MAX_BACKOFF_SECONDS 120
+#define GRPC_GRPCLB_RECONNECT_JITTER 0.2
 
 int grpc_lb_glb_trace = 0;
 
@@ -1107,9 +1108,12 @@ static void lb_call_init_locked(glb_lb_policy *glb_policy) {
   grpc_closure_init(&glb_policy->lb_on_response_received,
                     lb_on_response_received, glb_policy);
 
-  gpr_backoff_init(&glb_policy->lb_call_backoff_state, BACKOFF_MULTIPLIER,
-                   BACKOFF_JITTER, BACKOFF_MIN_SECONDS * 1000,
-                   BACKOFF_MAX_SECONDS * 1000);
+  gpr_backoff_init(&glb_policy->lb_call_backoff_state,
+                   GRPC_GRPCLB_INITIAL_CONNECT_BACKOFF_SECONDS,
+                   GRPC_GRPCLB_RECONNECT_BACKOFF_MULTIPLIER,
+                   GRPC_GRPCLB_RECONNECT_JITTER,
+                   GRPC_GRPCLB_MIN_CONNECT_TIMEOUT_SECONDS * 1000,
+                   GRPC_GRPCLB_RECONNECT_MAX_BACKOFF_SECONDS * 1000);
 }
 
 static void lb_call_destroy_locked(glb_lb_policy *glb_policy) {
diff --git a/src/core/ext/resolver/dns/native/dns_resolver.c b/src/core/ext/resolver/dns/native/dns_resolver.c
index 665439f360..15476f5792 100644
--- a/src/core/ext/resolver/dns/native/dns_resolver.c
+++ b/src/core/ext/resolver/dns/native/dns_resolver.c
@@ -46,10 +46,11 @@
 #include "src/core/lib/support/backoff.h"
 #include "src/core/lib/support/string.h"
 
-#define BACKOFF_MULTIPLIER 1.6
-#define BACKOFF_JITTER 0.2
-#define BACKOFF_MIN_SECONDS 1
-#define BACKOFF_MAX_SECONDS 120
+#define GRPC_DNS_MIN_CONNECT_TIMEOUT_SECONDS 1
+#define GRPC_DNS_INITIAL_CONNECT_BACKOFF_SECONDS 1
+#define GRPC_DNS_RECONNECT_BACKOFF_MULTIPLIER 1.6
+#define GRPC_DNS_RECONNECT_MAX_BACKOFF_SECONDS 120
+#define GRPC_DNS_RECONNECT_JITTER 0.2
 
 typedef struct {
   /** base class: must be first */
@@ -269,8 +270,11 @@ static grpc_resolver *dns_create(grpc_resolver_args *args,
   server_name_arg.value.string = (char *)path;
   r->channel_args =
       grpc_channel_args_copy_and_add(args->args, &server_name_arg, 1);
-  gpr_backoff_init(&r->backoff_state, BACKOFF_MULTIPLIER, BACKOFF_JITTER,
-                   BACKOFF_MIN_SECONDS * 1000, BACKOFF_MAX_SECONDS * 1000);
+  gpr_backoff_init(&r->backoff_state, GRPC_DNS_INITIAL_CONNECT_BACKOFF_SECONDS,
+                   GRPC_DNS_RECONNECT_BACKOFF_MULTIPLIER,
+                   GRPC_DNS_RECONNECT_JITTER,
+                   GRPC_DNS_MIN_CONNECT_TIMEOUT_SECONDS * 1000,
+                   GRPC_DNS_RECONNECT_MAX_BACKOFF_SECONDS * 1000);
   return &r->base;
 }
 
diff --git a/src/core/lib/support/backoff.c b/src/core/lib/support/backoff.c
index e89ef47220..0612472712 100644
--- a/src/core/lib/support/backoff.c
+++ b/src/core/lib/support/backoff.c
@@ -35,8 +35,10 @@
 
 #include <grpc/support/useful.h>
 
-void gpr_backoff_init(gpr_backoff *backoff, double multiplier, double jitter,
+void gpr_backoff_init(gpr_backoff *backoff, int64_t initial_connect_timeout,
+                      double multiplier, double jitter,
                       int64_t min_timeout_millis, int64_t max_timeout_millis) {
+  backoff->initial_connect_timeout = initial_connect_timeout;
   backoff->multiplier = multiplier;
   backoff->jitter = jitter;
   backoff->min_timeout_millis = min_timeout_millis;
@@ -45,9 +47,10 @@ void gpr_backoff_init(gpr_backoff *backoff, double multiplier, double jitter,
 }
 
 gpr_timespec gpr_backoff_begin(gpr_backoff *backoff, gpr_timespec now) {
-  backoff->current_timeout_millis = backoff->min_timeout_millis;
-  return gpr_time_add(
-      now, gpr_time_from_millis(backoff->current_timeout_millis, GPR_TIMESPAN));
+  backoff->current_timeout_millis = backoff->initial_connect_timeout;
+  const int64_t first_timeout =
+      GPR_MAX(backoff->current_timeout_millis, backoff->min_timeout_millis);
+  return gpr_time_add(now, gpr_time_from_millis(first_timeout, GPR_TIMESPAN));
 }
 
 /* Generate a random number between 0 and 1. */
@@ -57,20 +60,28 @@ static double generate_uniform_random_number(uint32_t *rng_state) {
 }
 
 gpr_timespec gpr_backoff_step(gpr_backoff *backoff, gpr_timespec now) {
-  double new_timeout_millis =
+  const double new_timeout_millis =
       backoff->multiplier * (double)backoff->current_timeout_millis;
-  double jitter_range = backoff->jitter * new_timeout_millis;
-  double jitter =
+  backoff->current_timeout_millis =
+      GPR_MIN((int64_t)new_timeout_millis, backoff->max_timeout_millis);
+
+  const double jitter_range_width = backoff->jitter * new_timeout_millis;
+  const double jitter =
       (2 * generate_uniform_random_number(&backoff->rng_state) - 1) *
-      jitter_range;
+      jitter_range_width;
+
   backoff->current_timeout_millis =
-      GPR_CLAMP((int64_t)(new_timeout_millis + jitter),
-                backoff->min_timeout_millis, backoff->max_timeout_millis);
-  return gpr_time_add(
+      (int64_t)((double)(backoff->current_timeout_millis) + jitter);
+
+  const gpr_timespec current_deadline = gpr_time_add(
       now, gpr_time_from_millis(backoff->current_timeout_millis, GPR_TIMESPAN));
+
+  const gpr_timespec min_deadline = gpr_time_add(
+      now, gpr_time_from_millis(backoff->min_timeout_millis, GPR_TIMESPAN));
+
+  return gpr_time_max(current_deadline, min_deadline);
 }
 
 void gpr_backoff_reset(gpr_backoff *backoff) {
-  // forces step() to return a timeout of min_timeout_millis
-  backoff->current_timeout_millis = 0;
+  backoff->current_timeout_millis = backoff->initial_connect_timeout;
 }
diff --git a/src/core/lib/support/backoff.h b/src/core/lib/support/backoff.h
index 6d40c15546..5e9b740824 100644
--- a/src/core/lib/support/backoff.h
+++ b/src/core/lib/support/backoff.h
@@ -37,7 +37,9 @@
 #include <grpc/support/time.h>
 
 typedef struct {
-  /// const: multiplier between retry attempts
+  /// const:  how long to wait after the first failure before retrying
+  int64_t initial_connect_timeout;
+  /// const: factor with which to multiply backoff after a failed retry
   double multiplier;
   /// const: amount to randomize backoffs
   double jitter;
@@ -54,7 +56,8 @@ typedef struct {
 } gpr_backoff;
 
 /// Initialize backoff machinery - does not need to be destroyed
-void gpr_backoff_init(gpr_backoff *backoff, double multiplier, double jitter,
+void gpr_backoff_init(gpr_backoff *backoff, int64_t initial_connect_timeout,
+                      double multiplier, double jitter,
                       int64_t min_timeout_millis, int64_t max_timeout_millis);
 
 /// Begin retry loop: returns a timespec for the NEXT retry