From: Klaus Darilion <klaus.darilion@nic.at>
Date: Tue, 31 Jul 2018 21:01:42 +0000 (+0000)
Subject: add incremental slave-check backoff also for failed AXFR due to master problems
X-Git-Tag: dnsdist-1.3.3~84^2~2
X-Git-Url: http://git.ipfire.org/cgi-bin/gitweb.cgi?a=commitdiff_plain;h=b177f6e1f32ef947a4cb011da964bce53679b989;p=thirdparty%2Fpdns.git

add incremental slave-check backoff also for failed AXFR due to master problems

If SOA-checks against a master with failing AXFR still succeed, we would
constantly try to AXFR the zone. To avoid this, we add the zone to the list
of failed slave-checks to suspend slave-checks (and subsequent AXFR) for
this zone for some time.
---

diff --git a/pdns/slavecommunicator.cc b/pdns/slavecommunicator.cc
index 9dd57548b8..ec5cfb4d02 100644
--- a/pdns/slavecommunicator.cc
+++ b/pdns/slavecommunicator.cc
@@ -615,7 +615,17 @@ void CommunicatorClass::suck(const DNSName &domain, const ComboAddress& remote)
     }
   }
   catch(ResolverException &re) {
-    g_log<<Logger::Error<<"Unable to AXFR zone '"<<domain<<"' from remote '"<<remote<<"' (resolver): "<<re.reason<<endl;
+    // The AXFR probably failed due to a problem on the master server. If SOA-checks against this master
+    // still succeed, we would constantly try to AXFR the zone. To avoid this, we add the zone to the list of
+    // failed slave-checks. This will suspend slave-checks (and subsequent AXFR) for this zone for some time.
+    uint64_t newCount = 1;
+    time_t now = time(0);
+    const auto failedEntry = d_failedSlaveRefresh.find(domain);
+    if (failedEntry != d_failedSlaveRefresh.end())
+      newCount = d_failedSlaveRefresh[domain].first + 1;
+    time_t nextCheck = now + std::min(newCount * d_tickinterval, (uint64_t)::arg().asNum("soa-retry-default"));
+    d_failedSlaveRefresh[domain] = {newCount, nextCheck};
+    g_log<<Logger::Error<<"Unable to AXFR zone '"<<domain<<"' from remote '"<<remote<<"' (resolver): "<<re.reason<<" (This was the "<<(newCount == 1 ? "first" : std::to_string(newCount) + "th")<<" time. Excluding zone from slave-checks until "<<nextCheck<<")"<<endl;
     if(di.backend && transaction) {
       g_log<<Logger::Error<<"Aborting possible open transaction for domain '"<<domain<<"' AXFR"<<endl;
       di.backend->abortTransaction();