[kernel] Add new option 'servfail_retry_timeout'

As of version NNN of the DNS server, there is a servfail cache
which has the servfail-ttl default value of 1 sec.
This means if the agent is "to eager" after a servfail, it will
just get the cache'd value (that is, no dns lookup is attempted).
To avoid the situation, a new config option has been added;
servfail_retry_timeout (with default 1500 = 1.5 sec), which
defines the timeout after each failed query, before attempting
a new query.

OTP-16956

1 files changed, 11 insertions, 1 deletions

diff --git a/lib/kernel/src/inet_res.erl b/lib/kernel/src/inet_res.erl
index 7886ef83ac..620dfd0052 100644
--- a/lib/kernel/src/inet_res.erl
+++ b/lib/kernel/src/inet_res.erl
@@ -259,7 +259,7 @@ do_nslookup(Name, Class, Type, Opts, Timeout) ->
 %%
 -record(options, { % These must be sorted!
 	  alt_nameservers,edns,inet6,nameservers,recurse,
-	  retry,timeout,udp_payload_size,usevc,
+	  retry,servfail_retry_timeout,timeout,udp_payload_size,usevc,
 	  verbose}). % this is a local option, not in inet_db
 %%
 %% Opts when is_list(Opts) -> #options{}
@@ -772,8 +772,18 @@ query_retries(_Q, _NSs, _Timer, Retry, Retry, S, Reason) ->
 query_retries(_Q, [], _Timer, _Retry, _I, S, Reason) ->
     query_retries_error(S, Reason);
 query_retries(Q, NSs, Timer, Retry, I, S_0, Reason) ->
+    servfail_retry_wait(Q, I),
     query_nss(Q, NSs, Timer, Retry, I, S_0, Reason, NSs).
 
+servfail_retry_wait(_Q, 0) ->
+    ok;
+servfail_retry_wait(#q{options = #options{servfail_retry_timeout = T}}, _)
+  when (T > 0) ->
+    receive after T -> ok end;
+servfail_retry_wait(_, _) ->
+    ok.
+
+
 %% Loop for all name servers, for each:
 %%     If EDNS is enabled, try that first,
 %%     and for selected failures fall back to plain DNS.