File 3162-kernel-erts-Add-text-about-the-new-resolver-config-o.patch of Package erlang
From 199c4652ed5842f718123c96078eee620652bfe2 Mon Sep 17 00:00:00 2001
From: Micael Karlberg <bmk@erlang.org>
Date: Mon, 26 Oct 2020 17:17:06 +0100
Subject: [PATCH 2/6] [kernel|erts] Add text about the new resolver config
option
Add text about the new resolver config option and its meaning.
OTP-16956
---
erts/doc/src/inet_cfg.xml | 13 +++++++-
lib/kernel/doc/src/inet_res.xml | 56 +++++++++++++++++++--------------
2 files changed, 45 insertions(+), 24 deletions(-)
diff --git a/erts/doc/src/inet_cfg.xml b/erts/doc/src/inet_cfg.xml
index 0e2437e33c..7934a6df86 100644
--- a/erts/doc/src/inet_cfg.xml
+++ b/erts/doc/src/inet_cfg.xml
@@ -4,7 +4,7 @@
<chapter>
<header>
<copyright>
- <year>2004</year><year>2016</year>
+ <year>2004</year><year>2020</year>
<holder>Ericsson AB. All Rights Reserved.</holder>
</copyright>
<legalnotice>
@@ -270,6 +270,17 @@
<seealso marker="kernel:inet_res"><c>inet_res(3)</c></seealso>
will try before giving up. Defaults to 3.</p>
</item>
+
+ <tag><c><![CDATA[{servfail_retry_timeout, Time}.]]></c></tag>
+ <item>
+ <p><c><![CDATA[Time = non_neg_integer()]]></c></p>
+ <p>After all name servers have been tried, there is a timeout
+ before the name servers are tried again. This is to prevent
+ the server from answering the query with what's in the servfail cache,
+ <seealso marker="kernel:inet_res#servfail_retry_timeout"><c>inet_res(3)</c></seealso>.
+ Defaults to 1500 milli seconds .</p>
+ </item>
+
<tag><c><![CDATA[{inet6, Bool}.]]></c></tag>
<item>
<p><c><![CDATA[Bool = true | false]]></c></p>
diff --git a/lib/kernel/doc/src/inet_res.xml b/lib/kernel/doc/src/inet_res.xml
index c30038f4cd..f50915a2e2 100644
--- a/lib/kernel/doc/src/inet_res.xml
+++ b/lib/kernel/doc/src/inet_res.xml
@@ -4,7 +4,7 @@
<erlref>
<header>
<copyright>
- <year>2009</year><year>2018</year>
+ <year>2009</year><year>2020</year>
<holder>Ericsson AB. All Rights Reserved.</holder>
</copyright>
<legalnotice>
@@ -48,31 +48,41 @@
<section>
<title>Name Resolving</title>
<p>UDP queries are used unless resolver option
- <c>usevc</c> is <c>true</c>, which forces TCP queries.
- If the query is too large for UDP, TCP is used instead.
- For regular DNS queries, 512 bytes is the size limit.</p>
+ <c>usevc</c> is <c>true</c>, which forces TCP queries.
+ If the query is too large for UDP, TCP is used instead.
+ For regular DNS queries, 512 bytes is the size limit.</p>
+
<p>When EDNS is enabled (resolver option
- <c>edns</c> is set to the EDNS version (that is, <c>0</c>
- instead of <c>false</c>), resolver option
- <c>udp_payload_size</c> sets the limit. If a name server
- replies with the TC bit set (truncation), indicating that
- the answer is incomplete, the query is retried
- to that name server using TCP. Resolver option
- <c>udp_payload_size</c> also sets the advertised
- size for the maximum allowed reply size, if EDNS is
- enabled, otherwise the name server uses the limit
- 512 bytes. If the reply is larger, it gets truncated,
- forcing a TCP requery.</p>
+ <c>edns</c> is set to the EDNS version (that is, <c>0</c>
+ instead of <c>false</c>), resolver option
+ <c>udp_payload_size</c> sets the limit. If a name server
+ replies with the TC bit set (truncation), indicating that
+ the answer is incomplete, the query is retried
+ to that name server using TCP. Resolver option
+ <c>udp_payload_size</c> also sets the advertised
+ size for the maximum allowed reply size, if EDNS is
+ enabled, otherwise the name server uses the limit
+ 512 bytes. If the reply is larger, it gets truncated,
+ forcing a TCP requery.</p>
+
<p>For UDP queries, resolver options <c>timeout</c>
- and <c>retry</c> control retransmission.
- Each name server in the <c>nameservers</c> list is
- tried with a time-out of <c>timeout</c>/<c>retry</c>.
- Then all name servers are tried again, doubling the
- time-out, for a total of <c>retry</c> times.</p>
+ and <c>retry</c> control retransmission.
+ Each name server in the <c>nameservers</c> list is
+ tried with a time-out of <c>timeout</c>/<c>retry</c>.
+ Then all name servers are tried again, doubling the
+ time-out, for a total of <c>retry</c> times.</p>
+
+ <marker id="servfail_retry_timeout"/>
+ <p>But before all name servers are tried again, there is a
+ (user configurable) timeout, <c>servfail_retry_timeout</c>.
+ The point of this is to prevent the new query to be handled
+ by to the servfail cache (a client that is to eager will
+ actually only get what is in the servfail cache). </p>
+
<p>For queries not using the <c>search</c> list,
- if the query to all <c>nameservers</c> results in
- <c>{error,nxdomain}</c> or an empty answer, the same
- query is tried for <c>alt_nameservers</c>.</p>
+ if the query to all <c>nameservers</c> results in
+ <c>{error,nxdomain}</c> or an empty answer, the same
+ query is tried for <c>alt_nameservers</c>.</p>
</section>
<section>
--
2.26.2
