Previous topic

DNSSEC in the PowerDNS Recursor

Next topic

PowerDNS Recursor New Style (YAML) Settings

This Page

PowerDNS Recursor Settings

Each setting can appear on the command line, prefixed by --, or in the configuration file. The command line overrides the configuration file.

Note

Starting with version 5.0.0, Recursor supports a new YAML syntax for configuration files. A configuration using the old style syntax can be converted to a YAML configuration using the instructions in Conversion of old-style settings to YAML format. In a future release support for the “old-style” settings decribed here will be dropped. See PowerDNS Recursor New Style (YAML) Settings for details.

Note

Settings marked as Boolean can either be set to an empty value, which means on, or to no or off which means off. Anything else means on.

For example:

  • serve-rfc1918 on its own means: do serve those zones.
  • serve-rfc1918 = off or serve-rfc1918 = no means: do not serve those zones.
  • Anything else means: do serve those zones.

You can use += syntax to set some variables incrementally, but this requires you to have at least one non-incremental setting for the variable to act as base setting. This is mostly useful for include-dir directive. An example:

forward-zones = foo.example.com=192.168.100.1;
forward-zones += bar.example.com=[1234::abcde]:5353;

When a list of Netmasks is mentioned, a list of subnets can be specified. A subnet that is not followed by / will be interpreted as a /32 or /128 subnet (a single address), depending on address family. For most settings, it is possible to exclude ranges by prefixing an item with the negation character !. For example:

allow-from = 2001:DB8::/32, 128.66.0.0/16, !128.66.1.2

In this case the address 128.66.1.2 is excluded from the addresses allowed access.

The Settings

aggressive-nsec-cache-size

New in version 4.5.0.

The number of records to cache in the aggressive cache. If set to a value greater than 0, the recursor will cache NSEC and NSEC3 records to generate negative answers, as defined in RFC 8198. To use this, DNSSEC processing or validation must be enabled by setting dnssec to process, log-fail or validate.

aggressive-cache-min-nsec3-hit-ratio

New in version 4.9.0.

The limit for which to put NSEC3 records into the aggressive cache. A value of n means that an NSEC3 record is only put into the aggressive cache if the estimated probability of a random name hitting the NSEC3 record is higher than 1/n. A higher n will cause more records to be put into the aggressive cache, e.g. a value of 4000 will cause records to be put in the aggressive cache even if the estimated probability of hitting them is twice as low as would be the case for n=2000. A value of 0 means no NSEC3 records will be put into the aggressive cache.

For large zones the effectiveness of the NSEC3 cache is reduced since each NSEC3 record only covers a randomly distributed subset of all possible names. This setting avoids doing unnecessary work for such large zones.

allow-from

  • Comma separated list of IP addresses or subnets, negation supported
  • Default: 127.0.0.0/8, 10.0.0.0/8, 100.64.0.0/10, 169.254.0.0/16, 192.168.0.0/16, 172.16.0.0/12, ::1/128, fc00::/7, fe80::/10
  • YAML setting: incoming.allow_from

Netmasks (both IPv4 and IPv6) that are allowed to use the server. The default allows access only from RFC 1918 private IP addresses. An empty value means no checking is done, all clients are allowed. Due to the aggressive nature of the internet these days, it is highly recommended to not open up the recursor for the entire internet. Questions from IP addresses not listed here are ignored and do not get an answer.

When the Proxy Protocol is enabled (see proxy-protocol-from), the recursor will check the address of the client IP advertised in the Proxy Protocol header instead of the one of the proxy.

Note that specifying an IP address without a netmask uses an implicit netmask of /32 or /128.

allow-from-file

Like allow-from, except reading from file. Overrides the allow-from setting. To use this feature, supply one netmask per line, with optional comments preceded by a ‘#’.

allow-notify-for

New in version 4.6.0.

Domain names specified in this list are used to permit incoming NOTIFY operations to wipe any cache entries that match the domain name. If this list is empty, all NOTIFY operations will be ignored.

allow-notify-for-file

New in version 4.6.0.

Like allow-notify-for, except reading from file. To use this feature, supply one domain name per line, with optional comments preceded by a ‘#’.

NOTIFY-allowed zones can also be specified using forward-zones-file.

allow-notify-from

New in version 4.6.0.

Netmasks (both IPv4 and IPv6) that are allowed to issue NOTIFY operations to the server. NOTIFY operations from IP addresses not listed here are ignored and do not get an answer.

When the Proxy Protocol is enabled (see proxy-protocol-from), the recursor will check the address of the client IP advertised in the Proxy Protocol header instead of the one of the proxy.

Note that specifying an IP address without a netmask uses an implicit netmask of /32 or /128.

NOTIFY operations received from a client listed in one of these netmasks will be accepted and used to wipe any cache entries whose zones match the zone specified in the NOTIFY operation, but only if that zone (or one of its parents) is included in allow-notify-for, allow-notify-for-file, or forward-zones-file with a ‘^’ prefix.

allow-notify-from-file

New in version 4.6.0.

Like allow-notify-from, except reading from file. To use this feature, supply one netmask per line, with optional comments preceded by a ‘#’.

allow-no-rd

New in version 5.0.0.

Allow no recursion desired (RD=0) queries to query cache contents. If not set (the default), these queries are answered with rcode Refused.

any-to-tcp

Answer questions for the ANY type on UDP with a truncated packet that refers the remote server to TCP. Useful for mitigating ANY reflection attacks.

allow-trust-anchor-query

New in version 4.3.0.

Allow trustanchor.server CH TXT and negativetrustanchor.server CH TXT queries to view the configured DNSSEC (negative) trust anchors.

api-config-dir

New in version 4.0.0.

Directory where the REST API stores its configuration and zones. For configuration updates to work, include-dir should have the same value when using old-style settings. When using YAML settings recursor.include_dir and webservice.api_dir must have a different value.

api-key

New in version 4.0.0.

Changed in version 4.6.0: This setting now accepts a hashed and salted version.

Static pre-shared authentication key for access to the REST API. Since 4.6.0 the key can be hashed and salted using rec_control hash-password instead of being stored in the configuration in plaintext, but the plaintext version is still supported.

auth-zones

  • Comma separated list of ‘zonename=filename’ pairs
  • Default: (empty)
  • YAML setting: recursor.auth_zones

Zones read from these files (in BIND format) are served authoritatively (but without the AA bit set in responses). DNSSEC is not supported. Example:

auth-zones=example.org=/var/zones/example.org, powerdns.com=/var/zones/powerdns.com

carbon-interval

If sending carbon updates, this is the interval between them in seconds. See Metrics and Statistics.

carbon-namespace

New in version 4.2.0.

  • String
  • Default: pdns
  • YAML setting: carbon.ns

Change the namespace or first string of the metric key. The default is pdns.

carbon-ourname

If sending carbon updates, if set, this will override our hostname. Be careful not to include any dots in this setting, unless you know what you are doing. See Sending metrics to Graphite/Metronome over Carbon.

carbon-instance

New in version 4.2.0.

Change the instance or third string of the metric key. The default is recursor.

carbon-server

  • Comma separated list or IPs of IP:port combinations
  • Default: (empty)
  • YAML setting: carbon.server

If set to an IP or IPv6 address, will send all available metrics to this server via the carbon protocol, which is used by graphite and metronome. Moreover you can specify more than one server using a comma delimited list, ex: carbon-server=10.10.10.10,10.10.10.20. You may specify an alternate port by appending :port, for example: 127.0.0.1:2004. See Metrics and Statistics.

chroot

If set, chroot to this directory for more security. This is not recommended; instead, we recommend containing PowerDNS using operating system features. We ship systemd unit files with our packages to make this easy.

Make sure that /dev/log is available from within the chroot. Logging will silently fail over time otherwise (on logrotate).

When using chroot, all other paths (except for config-dir) set in the configuration are relative to the new root.

When running on a system where systemd manages services, chroot does not work out of the box, as PowerDNS cannot use the NOTIFY_SOCKET. Either do not chroot on these systems or set the ‘Type’ of this service to ‘simple’ instead of ‘notify’ (refer to the systemd documentation on how to modify unit-files).

client-tcp-timeout

Time to wait for data from TCP clients.

config-dir

Location of configuration directory (where recursor.conf or recursor.yml is stored). Usually /etc/powerdns, but this depends on SYSCONFDIR during compile-time. Use default or set on command line.

config-name

When running multiple recursors on the same server, read settings from recursor-name.conf, this will also rename the binary image.

cpu-map

Set CPU affinity for threads, asking the scheduler to run those threads on a single CPU, or a set of CPUs. This parameter accepts a space separated list of thread-id=cpu-id, or thread-id=cpu-id-1,cpu-id-2,…,cpu-id-N. For example, to make the worker thread 0 run on CPU id 0 and the worker thread 1 on CPUs 1 and 2:

recursor:
  cpu_map: 0=0 1=1,2

The thread handling the control channel, the webserver and other internal stuff has been assigned id 0, the distributor threads if any are assigned id 1 and counting, and the worker threads follow behind. The number of distributor threads is determined by distributor-threads, the number of worker threads is determined by the threads setting.

This parameter is only available if the OS provides the pthread_setaffinity_np() function.

Note that depending on the configuration the Recursor can start more threads. Typically these threads will sleep most of the time. These threads cannot be specified in this setting as their thread-ids are left unspecified.

daemon

Changed in version 4.0.0: Default is now no, was yes before.

Operate in the background.

dont-throttle-names

New in version 4.2.0.

When an authoritative server does not answer a query or sends a reply the recursor does not like, it is throttled. Any servers’ name suffix-matching the supplied names will never be throttled.

Warning

Most servers on the internet do not respond for a good reason (overloaded or unreachable), dont-throttle-names could make this load on the upstream server even higher, resulting in further service degradation.

dont-throttle-netmasks

New in version 4.2.0.

When an authoritative server does not answer a query or sends a reply the recursor does not like, it is throttled. Any servers matching the supplied netmasks will never be throttled.

This can come in handy on lossy networks when forwarding, where the same server is configured multiple times (e.g. with forward-zones-recurse=example.com=192.0.2.1;192.0.2.1). By default, the PowerDNS Recursor would throttle the ‘first’ server on a timeout and hence not retry the ‘second’ one. In this case, dont-throttle-netmasks could be set to 192.0.2.1.

Warning

Most servers on the internet do not respond for a good reason (overloaded or unreachable), dont-throttle-netmasks could make this load on the upstream server even higher, resulting in further service degradation.

disable-packetcache

Turn off the packet cache. Useful when running with Lua scripts that modify answers in such a way they cannot be cached, though individual answer caching can be controlled from Lua as well.

disable-syslog

Do not log to syslog, only to stderr. Use this setting when running inside a supervisor that handles logging (like systemd). Note: do not use this setting in combination with daemon as all logging will disappear.

distribution-load-factor

New in version 4.1.12.

If pdns-distributes-queries is set and this setting is set to another value than 0, the distributor thread will use a bounded load-balancing algorithm while distributing queries to worker threads, making sure that no thread is assigned more queries than distribution-load-factor times the average number of queries currently processed by all the workers. For example, with a value of 1.25, no server should get more than 125 % of the average load. This helps making sure that all the workers have roughly the same share of queries, even if the incoming traffic is very skewed, with a larger number of requests asking for the same qname.

distribution-pipe-buffer-size

New in version 4.2.0.

Size in bytes of the internal buffer of the pipe used by the distributor to pass incoming queries to a worker thread. Requires support for F_SETPIPE_SZ which is present in Linux since 2.6.35. The actual size might be rounded up to a multiple of a page size. 0 means that the OS default size is used. A large buffer might allow the recursor to deal with very short-lived load spikes during which a worker thread gets overloaded, but it will be at the cost of an increased latency.

distributor-threads

New in version 4.2.0.

If pdns-distributes-queries is set, spawn this number of distributor threads on startup. Distributor threads handle incoming queries and distribute them to other threads based on a hash of the query.

dot-to-auth-names

New in version 4.6.0.

Force DoT to the listed authoritative nameservers. For this to work, DoT support has to be compiled in. Currently, the certificate is not checked for validity in any way.

dot-to-port-853

New in version 4.6.0.

Enable DoT to forwarders that specify port 853.

dns64-prefix

New in version 4.4.0.

Enable DNS64 (RFC 6147) support using the supplied /96 IPv6 prefix. This will generate ‘fake’ AAAA records for names with only A records, as well as ‘fake’ PTR records to make sure that reverse lookup of DNS64-generated IPv6 addresses generate the right name. See DNS64 support for more flexible but slower alternatives using Lua.

dnssec

New in version 4.0.0.

Changed in version 4.5.0: The default changed from process-no-validate to process

One of off, process-no-validate, process, log-fail, validate

Set the mode for DNSSEC processing, as detailed in DNSSEC in the PowerDNS Recursor.

off
No DNSSEC processing whatsoever. Ignore DO-bits in queries, don’t request any DNSSEC information from authoritative servers. This behaviour is similar to PowerDNS Recursor pre-4.0.
process-no-validate
Respond with DNSSEC records to clients that ask for it, set the DO bit on all outgoing queries. Don’t do any validation.
process
Respond with DNSSEC records to clients that ask for it, set the DO bit on all outgoing queries. Do validation for clients that request it (by means of the AD- bit or DO-bit in the query).
log-fail
Similar behaviour to process, but validate RRSIGs on responses and log bogus responses.
validate
Full blown DNSSEC validation. Send SERVFAIL to clients on bogus responses.

dnssec-disabled-algorithms

New in version 4.9.0.

A list of DNSSEC algorithm numbers that should be considered disabled. These algorithms will not be used to validate DNSSEC signatures. Zones (only) signed with these algorithms will be considered Insecure.

If this setting is empty (the default), Recursor will determine which algorithms to disable automatically. This is done for specific algorithms only, currently algorithms 5 (RSASHA1) and 7 (RSASHA1NSEC3SHA1).

This is important on systems that have a default strict crypto policy, like RHEL9 derived systems. On such systems not disabling some algorithms (or changing the security policy) will make affected zones to be considered Bogus as using these algorithms fails.

dnssec-log-bogus

Log every DNSSEC validation failure. Note: This is not logged per-query but every time records are validated as Bogus.

dont-query

  • Comma separated list of IP addresses or subnets, negation supported
  • Default: 127.0.0.0/8, 10.0.0.0/8, 100.64.0.0/10, 169.254.0.0/16, 192.168.0.0/16, 172.16.0.0/12, ::1/128, fc00::/7, fe80::/10, 0.0.0.0/8, 192.0.0.0/24, 192.0.2.0/24, 198.51.100.0/24, 203.0.113.0/24, 240.0.0.0/4, ::/96, ::ffff:0:0/96, 100::/64, 2001:db8::/32
  • YAML setting: outgoing.dont_query

The DNS is a public database, but sometimes contains delegations to private IP addresses, like for example 127.0.0.1. This can have odd effects, depending on your network, and may even be a security risk. Therefore, the PowerDNS Recursor by default does not query private space IP addresses. This setting can be used to expand or reduce the limitations.

Queries for names in forward zones and to addresses as configured in any of the settings forward-zones, forward-zones-file or forward-zones-recurse are performed regardless of these limitations.

ecs-add-for

New in version 4.2.0.

  • Comma separated list of IP addresses or subnets, negation supported
  • Default: 0.0.0.0/0, ::/0, !127.0.0.0/8, !10.0.0.0/8, !100.64.0.0/10, !169.254.0.0/16, !192.168.0.0/16, !172.16.0.0/12, !::1/128, !fc00::/7, !fe80::/10
  • YAML setting: ecs.add_for

List of requestor netmasks for which the requestor IP Address should be used as the EDNS Client Subnet for outgoing queries. Outgoing queries for requestors that do not match this list will use the ecs-scope-zero-address instead. Valid incoming ECS values from use-incoming-edns-subnet are not replaced.

Regardless of the value of this setting, ECS values are only sent for outgoing queries matching the conditions in the edns-subnet-allow-list setting. This setting only controls the actual value being sent.

This defaults to not using the requestor address inside RFC1918 and similar ‘private’ IP address spaces.

ecs-ipv4-bits

New in version 4.1.0.

Number of bits of client IPv4 address to pass when sending EDNS Client Subnet address information.

ecs-ipv4-cache-bits

New in version 4.1.12.

Maximum number of bits of client IPv4 address used by the authoritative server (as indicated by the EDNS Client Subnet scope in the answer) for an answer to be inserted into the record cache. This condition applies in conjunction with ecs-cache-limit-ttl. That is, only if both the limits apply, the record will not be cached. This decision can be overridden by ecs-ipv4-never-cache and ecs-ipv6-never-cache.

ecs-ipv6-bits

New in version 4.1.0.

Number of bits of client IPv6 address to pass when sending EDNS Client Subnet address information.

ecs-ipv6-cache-bits

New in version 4.1.12.

Maximum number of bits of client IPv6 address used by the authoritative server (as indicated by the EDNS Client Subnet scope in the answer) for an answer to be inserted into the record cache. This condition applies in conjunction with ecs-cache-limit-ttl. That is, only if both the limits apply, the record will not be cached. This decision can be overridden by ecs-ipv4-never-cache and ecs-ipv6-never-cache.

ecs-ipv4-never-cache

New in version 4.5.0.

When set, never cache replies carrying EDNS IPv4 Client Subnet scope in the record cache. In this case the decision made by `ecs-ipv4-cache-bits and ecs-cache-limit-ttl is no longer relevant.

ecs-ipv6-never-cache

New in version 4.5.0.

When set, never cache replies carrying EDNS IPv6 Client Subnet scope in the record cache. In this case the decision made by `ecs-ipv6-cache-bits and ecs-cache-limit-ttl is no longer relevant.

ecs-minimum-ttl-override

Changed in version 4.5.0: Old versions used default 0.

This setting artificially raises the TTLs of records in the ANSWER section of ECS-specific answers to be at least this long. Setting this to a value greater than 1 technically is an RFC violation, but might improve performance a lot. Using a value of 0 impacts performance of TTL 0 records greatly, since it forces the recursor to contact authoritative servers every time a client requests them. Can be set at runtime using rec_control set-ecs-minimum-ttl 3600.

ecs-cache-limit-ttl

New in version 4.1.12.

The minimum TTL for an ECS-specific answer to be inserted into the record cache. This condition applies in conjunction with ecs-ipv4-cache-bits or ecs-ipv6-cache-bits. That is, only if both the limits apply, the record will not be cached. This decision can be overridden by ecs-ipv4-never-cache and ecs-ipv6-never-cache.

ecs-scope-zero-address

New in version 4.1.0.

The IP address sent via EDNS Client Subnet to authoritative servers listed in edns-subnet-allow-list when use-incoming-edns-subnet is set and the query has an ECS source prefix-length set to 0. The default is to look for the first usable (not an any one) address in query-local-address (starting with IPv4). If no suitable address is found, the recursor fallbacks to sending 127.0.0.1.

edns-outgoing-bufsize

Changed in version 4.2.0: Before 4.2.0, the default was 1680

Note

Why 1232?

1232 is the largest number of payload bytes that can fit in the smallest IPv6 packet. IPv6 has a minimum MTU of 1280 bytes (RFC 8200, section 5), minus 40 bytes for the IPv6 header, minus 8 bytes for the UDP header gives 1232, the maximum payload size for the DNS response.

This is the value set for the EDNS0 buffer size in outgoing packets. Lower this if you experience timeouts.

edns-padding-from

New in version 4.5.0.

Changed in version 5.0.5: YAML settings only: previously this was defined as a string instead of a sequence

List of netmasks (proxy IP in case of proxy-protocol presence, client IP otherwise) for which EDNS padding will be enabled in responses, provided that edns-padding-mode applies.

edns-padding-mode

New in version 4.5.0.

One of always, padded-queries-only. Whether to add EDNS padding to all responses (always) or only to responses for queries containing the EDNS padding option (padded-queries-only, the default). In both modes, padding will only be added to responses for queries coming from edns-padding-from sources.

edns-padding-out

New in version 4.8.0.

Whether to add EDNS padding to outgoing DoT queries.

edns-padding-tag

New in version 4.5.0.

The packetcache tag to use for padded responses, to prevent a client not allowed by the :ref::setting-edns-padding-from list to be served a cached answer generated for an allowed one. This effectively divides the packet cache in two when edns-padding-from is used. Note that this will not override a tag set from one of the Lua hooks.

edns-subnet-whitelist

Deprecated since version 4.5.0: Use edns-subnet-allow-list.

  • String
  • Default: (empty)
  • YAML setting does not exist

edns-subnet-allow-list

New in version 4.5.0.

List of netmasks and domains that EDNS Client Subnet should be enabled for in outgoing queries.

For example, an EDNS Client Subnet option containing the address of the initial requestor (but see ecs-add-for) will be added to an outgoing query sent to server 192.0.2.1 for domain X if 192.0.2.1 matches one of the supplied netmasks, or if X matches one of the supplied domains. The initial requestor address will be truncated to 24 bits for IPv4 (see ecs-ipv4-bits) and to 56 bits for IPv6 (see ecs-ipv6-bits), as recommended in the privacy section of RFC 7871.

Note that this setting describes the destination of outgoing queries, not the sources of incoming queries, nor the subnets described in the EDNS Client Subnet option.

By default, this option is empty, meaning no EDNS Client Subnet information is sent.

entropy-source

Changed in version 4.9.0: This setting is no longer used.

  • String
  • Default: /dev/urandom
  • YAML setting does not exist

PowerDNS can read entropy from a (hardware) source. This is used for generating random numbers which are very hard to predict. Generally on UNIX platforms, this source will be /dev/urandom, which will always supply random numbers, even if entropy is lacking. Change to /dev/random if PowerDNS should block waiting for enough entropy to arrive.

etc-hosts-file

The path to the /etc/hosts file, or equivalent. This file can be used to serve data authoritatively using export-etc-hosts.

event-trace-enabled

New in version 4.6.0.

Enable the recording and logging of ref:event traces. This is an experimental feature and subject to change. Possible values are 0: (disabled), 1 (add information to protobuf logging messages) and 2 (write to log) and 3 (both).

export-etc-hosts

If set, this flag will export the host names and IP addresses mentioned in /etc/hosts.

export-etc-hosts-search-suffix

If set, all hostnames in the export-etc-hosts file are loaded in canonical form, based on this suffix, unless the name contains a ‘.’, in which case the name is unchanged. So an entry called ‘pc’ with export-etc-hosts-search-suffix='home.com' will lead to the generation of ‘pc.home.com’ within the recursor. An entry called ‘server1.home’ will be stored as ‘server1.home’, regardless of this setting.

extended-resolution-errors

New in version 4.5.0.

Changed in version 5.0.0: Default changed to enabled, previously it was disabled.

If set, the recursor will add an EDNS Extended Error (RFC 8914) to responses when resolution failed, like DNSSEC validation errors, explaining the reason it failed. This setting is not needed to allow setting custom error codes from Lua or from a RPZ hit.

forward-zones

Queries for zones listed here will be forwarded to the IP address listed. i.e.

forward-zones=example.org=203.0.113.210, powerdns.com=2001:DB8::BEEF:5

Multiple IP addresses can be specified and port numbers other than 53 can be configured:

forward-zones=example.org=203.0.113.210:5300;127.0.0.1, powerdns.com=127.0.0.1;198.51.100.10:530;[2001:DB8::1:3]:5300

Forwarded queries have the recursion desired (RD) bit set to 0, meaning that this setting is intended to forward queries to authoritative servers. If an NS record set for a subzone of the forwarded zone is learned, that record set will be used to determine addresses for name servers of the subzone. This allows e.g. a forward to a local authoritative server holding a copy of the root zone, delegations received from that server will work.

IMPORTANT: When using DNSSEC validation (which is default), forwards to non-delegated (e.g. internal) zones that have a DNSSEC signed parent zone will validate as Bogus. To prevent this, add a Negative Trust Anchor (NTA) for this zone in the lua-config-file with addNTA('your.zone', 'A comment'). If this forwarded zone is signed, instead of adding NTA, add the DS record to the lua-config-file. See the DNSSEC in the PowerDNS Recursor information.

forward-zones-file

Changed in version 4.0.0: (Old style settings only) Comments are allowed, everything behind # is ignored.

Changed in version 4.6.0: (Old style settings only) Zones prefixed with a ^ are added to the allow-notify-for list. Both prefix characters can be used if desired, in any order.

Same as forward-zones, parsed from a file. Only 1 zone is allowed per line, specified as follows:

example.org=203.0.113.210, 192.0.2.4:5300

Zones prefixed with a + are treated as with forward-zones-recurse. Default behaviour without + is as with forward-zones.

The DNSSEC notes from forward-zones apply here as well.

forward-zones-recurse

Like regular forward-zones, but forwarded queries have the recursion desired (RD) bit set to 1, meaning that this setting is intended to forward queries to other recursive servers. In contrast to regular forwarding, the rule that delegations of the forwarded subzones are respected is not active. This is because we rely on the forwarder to resolve the query fully.

See forward-zones for additional options (such as supplying multiple recursive servers) and an important note about DNSSEC.

gettag-needs-edns-options

New in version 4.1.0.

If set, EDNS options in incoming queries are extracted and passed to the gettag() hook in the ednsoptions table.

hint-file

Changed in version 4.6.2: Introduced the value no to disable root-hints processing.

Changed in version 4.9.0: Introduced the value no-refresh to disable both root-hints processing and periodic refresh of the cached root NS records.

If set, the root-hints are read from this file. If empty, the default built-in root hints are used.

In some special cases, processing the root hints is not needed, for example when forwarding all queries to another recursor. For these special cases, it is possible to disable the processing of root hints by setting the value to no or no-refresh. See Handling of root hints for more information on root hints handling.

ignore-unknown-settings

Names of settings to be ignored while parsing configuration files, if the setting name is unknown to PowerDNS.

Useful during upgrade testing.

include-dir

Directory to scan for additional config files. All files that end with .conf are loaded in order using POSIX as locale.

latency-statistic-size

Indication of how many queries will be averaged to get the average latency reported by the ‘qa-latency’ metric.

local-address

  • Comma separated list or IPs of IP:port combinations
  • Default: 127.0.0.1
  • YAML setting: incoming.listen

Local IP addresses to which we bind. Each address specified can include a port number; if no port is included then the local-port port will be used for that address. If a port number is specified, it must be separated from the address with a ‘:’; for an IPv6 address the address must be enclosed in square brackets.

Examples:

local-address=127.0.0.1 ::1
local-address=0.0.0.0:5353
local-address=[::]:8053
local-address=127.0.0.1:53, [::1]:5353

local-port

Local port to bind to. If an address in local-address does not have an explicit port, this port is used.

log-timestamp

non-local-bind

Bind to addresses even if one or more of the local-address’s do not exist on this server. Setting this option will enable the needed socket options to allow binding to non-local addresses. This feature is intended to facilitate ip-failover setups, but it may also mask configuration issues and for this reason it is disabled by default.

loglevel

Changed in version 5.0.0: Previous version would not allow setting a level below 3 (error).

Amount of logging. The higher the number, the more lines logged. Corresponds to syslog level values (e.g. 0 = emergency, 1 = alert, 2 = critical, 3 = error, 4 = warning, 5 = notice, 6 = info, 7 = debug). Each level includes itself plus the lower levels before it. Not recommended to set this below 3. If quiet is no/false, loglevel will be minimally set to 6 (info).

log-common-errors

Some DNS errors occur rather frequently and are no cause for alarm.

log-rpz-changes

New in version 4.1.0.

Log additions and removals to RPZ zones at Info (6) level instead of Debug (7).

logging-facility

If set to a digit, logging is performed under this LOCAL facility. See Logging. Do not pass names like ‘local0’!

lowercase-outgoing

Set to true to lowercase the outgoing queries. When set to ‘no’ (the default) a query from a client using mixed case in the DNS labels (such as a user entering mixed-case names or draft-vixie-dnsext-dns0x20-00), PowerDNS preserves the case of the query. Broken authoritative servers might give a wrong or broken answer on this encoding. Setting lowercase-outgoing to ‘yes’ makes the PowerDNS Recursor lowercase all the labels in the query to the authoritative servers, but still return the proper case to the client requesting.

lua-config-file

If set, and Lua support is compiled in, this will load an additional configuration file for newer features and more complicated setups. See Advanced Configuration Using Lua for the options that can be set in this file.

lua-dns-script

Path to a lua file to manipulate the Recursor’s answers. See Scripting PowerDNS Recursor for more information.

lua-maintenance-interval

New in version 4.2.0.

The interval between calls to the Lua user defined maintenance() function in seconds. See Maintenance callback

max-busy-dot-probes

New in version 4.7.0.

Limit the maximum number of simultaneous DoT probes the Recursor will schedule. The default value 0 means no DoT probes are scheduled.

DoT probes are used to check if an authoritative server’s IP address supports DoT. If the probe determines an IP address supports DoT, the Recursor will use DoT to contact it for subsequent queries until a failure occurs. After a failure, the Recursor will stop using DoT for that specific IP address for a while. The results of probes are remembered and can be viewed by the rec_control dump-dot-probe-map command. If the maximum number of pending probes is reached, no probes will be scheduled, even if no DoT status is known for an address. If the result of a probe is not yet available, the Recursor will contact the authoritative server in the regular way, unless an authoritative server is configured to be contacted over DoT always using dot-to-auth-names. In that case no probe will be scheduled.

Note

DoT probing is an experimental feature. Please test thoroughly to determine if it is suitable in your specific production environment before enabling.

max-cache-bogus-ttl

New in version 4.2.0.

Maximum number of seconds to cache an item in the DNS cache (negative or positive) if its DNSSEC validation failed, no matter what the original TTL specified, to reduce the impact of a broken domain.

max-cache-entries

Maximum number of DNS record cache entries, shared by all threads since 4.4.0. Each entry associates a name and type with a record set. The size of the negative cache is 10% of this number.

max-cache-ttl

Changed in version 4.1.0: The minimum value of this setting is 15. i.e. setting this to lower than 15 will make this value 15.

Maximum number of seconds to cache an item in the DNS cache, no matter what the original TTL specified. This value also controls the refresh period of cached root data. See Handling of root hints for more information on this.

max-concurrent-requests-per-tcp-connection

New in version 4.3.0.

Maximum number of incoming requests handled concurrently per tcp connection. This number must be larger than 0 and smaller than 65536 and also smaller than max-mthreads.

max-chain-length

New in version 5.1.0.

The maximum number of queries that can be attached to an outgoing request chain. Attaching requests to a chain saves on outgoing queries, but the processing of a chain when the reply to the outgoing query comes in might result in a large outgoing traffic spike. Reducing the maximum chain length mitigates this. If this value is zero, no maximum is enforced, though the maximum number of mthreads (max-mthreads) also limits the chain length.

max-include-depth

New in version 4.6.0.

Maximum number of nested $INCLUDE directives while processing a zone file. Zero mean no $INCLUDE directives will be accepted.

max-generate-steps

New in version 4.3.0.

Maximum number of steps for a ‘$GENERATE’ directive when parsing a zone file. This is a protection measure to prevent consuming a lot of CPU and memory when untrusted zones are loaded. Default to 0 which means unlimited.

max-mthreads

Maximum number of simultaneous MTasker threads, per worker thread.

max-packetcache-entries

Maximum number of Packet Cache entries. Sharded and shared by all threads since 4.9.0.

max-qperq

Changed in version 5.1.0: The default used to be 60, with an extra allowance if qname minimization was enabled. Having better algorithms allows for a lower default limit.

The maximum number of outgoing queries that will be sent out during the resolution of a single client query. This is used to avoid cycles resolving names.

max-cnames-followed

New in version 5.1.0.

Maximum length of a CNAME chain. If a CNAME chain exceeds this length, a ServFail answer will be returned. Previously, this limit was fixed at 10.

max-ns-address-qperq

New in version 4.1.16.

New in version 4.2.2.

New in version 4.3.1.

The maximum number of outgoing queries with empty replies for resolving nameserver names to addresses we allow during the resolution of a single client query. If IPv6 is enabled, an A and a AAAA query for a name counts as 1. If a zone publishes more than this number of NS records, the limit is further reduced for that zone by lowering it by the number of NS records found above the max-ns-address-qperq value. The limit wil not be reduced to a number lower than 5.

max-ns-per-resolve

New in version 4.8.0.

New in version 4.7.3.

New in version 4.6.4.

New in version 4.5.11.

The maximum number of NS records that will be considered to select a nameserver to contact to resolve a name. If a zone has more than max-ns-per-resolve NS records, a random sample of this size will be used. If max-ns-per-resolve is zero, no limit applies.

max-negative-ttl

A query for which there is authoritatively no answer is cached to quickly deny a record’s existence later on, without putting a heavy load on the remote server. In practice, caches can become saturated with hundreds of thousands of hosts which are tried only once. This setting, which defaults to 3600 seconds, puts a maximum on the amount of time negative entries are cached.

max-recursion-depth

Changed in version 4.1.0: Before 4.1.0, this settings was unlimited.

Changed in version 4.9.0: Before 4.9.0 this setting’s default was 40 and the limit on CNAME chains (fixed at 16) acted as a bound on he recursion depth.

Total maximum number of internal recursion calls the server may use to answer a single query. 0 means unlimited. The value of stack-size should be increased together with this one to prevent the stack from overflowing. If qname-minimization is enabled, the fallback code in case of a failing resolve is allowed an additional max-recursion-depth/2.

max-tcp-clients

Maximum number of simultaneous incoming TCP connections allowed.

max-tcp-per-client

Maximum number of simultaneous incoming TCP connections allowed per client (remote IP address).
0 means unlimited.

max-tcp-queries-per-connection

New in version 4.1.0.

Maximum number of DNS queries in a TCP connection. 0 means unlimited.

max-total-msec

Total maximum number of milliseconds of wallclock time the server may use to answer a single query. 0 means unlimited.

max-udp-queries-per-round

New in version 4.1.4.

Under heavy load the recursor might be busy processing incoming UDP queries for a long while before there is no more of these, and might therefore neglect scheduling new mthreads, handling responses from authoritative servers or responding to rec_control requests. This setting caps the maximum number of incoming UDP DNS queries processed in a single round of looping on recvmsg() after being woken up by the multiplexer, before returning back to normal processing and handling other events.

minimum-ttl-override

Changed in version 4.5.0: Old versions used default 0.

This setting artificially raises all TTLs to be at least this long. Setting this to a value greater than 1 technically is an RFC violation, but might improve performance a lot. Using a value of 0 impacts performance of TTL 0 records greatly, since it forces the recursor to contact authoritative servers each time a client requests them. Can be set at runtime using rec_control set-minimum-ttl 3600.

new-domain-tracking

New in version 4.2.0.

Whether to track newly observed domains, i.e. never seen before. This is a probabilistic algorithm, using a stable bloom filter to store records of previously seen domains. When enabled for the first time, all domains will appear to be newly observed, so the feature is best left enabled for e.g. a week or longer before using the results. Note that this feature is optional and must be enabled at compile-time, thus it may not be available in all pre-built packages. If protobuf is enabled and configured, then the newly observed domain status will appear as a flag in Response messages.

new-domain-log

New in version 4.2.0.

  • Boolean
  • Default: yes
  • YAML setting: nod.log

If a newly observed domain is detected, log that domain in the recursor log file. The log line looks something like:

Jul 18 11:31:25 Newly observed domain nod=sdfoijdfio.com

new-domain-lookup

New in version 4.2.0.

  • String
  • Default: (empty)
  • YAML setting: nod.lookup

If a domain is specified, then each time a newly observed domain is detected, the recursor will perform an A record lookup of ‘<newly observed domain>.<lookup domain>’. For example if ‘new-domain-lookup’ is configured as ‘nod.powerdns.com’, and a new domain ‘example.com’ is detected, then an A record lookup will be made for ‘example.com.nod.powerdns.com’. This feature gives a way to share the newly observed domain with partners, vendors or security teams. The result of the DNS lookup will be ignored by the recursor.

new-domain-db-size

New in version 4.2.0.

The default size of the stable bloom filter used to store previously observed domains is 67108864. To change the number of cells, use this setting. For each cell, the SBF uses 1 bit of memory, and one byte of disk for the persistent file. If there are already persistent files saved to disk, this setting will have no effect unless you remove the existing files.

new-domain-history-dir

New in version 4.2.0.

This setting controls which directory is used to store the on-disk cache of previously observed domains.

The default depends on LOCALSTATEDIR when building the software. Usually this comes down to /var/lib/pdns-recursor/nod or /usr/local/var/lib/pdns-recursor/nod).

The newly observed domain feature uses a stable bloom filter to store a history of previously observed domains. The data structure is synchronized to disk every 10 minutes, and is also initialized from disk on startup. This ensures that previously observed domains are preserved across recursor restarts. If you change the new-domain-db-size setting, you must remove any files from this directory.

new-domain-db-snapshot-interval

New in version 5.1.0.

Interval (in seconds) to write the NOD and UDR DB snapshots. Set to zero to disable snapshot writing.’,

new-domain-whitelist

New in version 4.2.0.

Deprecated since version 4.5.0: Use new-domain-ignore-list.

  • String
  • Default: (empty)
  • YAML setting does not exist

new-domain-ignore-list

New in version 4.5.0.

  • Comma separated list of strings
  • Default: (empty)
  • YAML setting: nod.ignore_list

This setting is a list of all domains (and implicitly all subdomains) that will never be considered a new domain. For example, if the domain ‘example.com’ is in the list, then ‘foo.bar.example.com’ will never be considered a new domain. One use-case for the ignore list is to never reveal details of internal subdomains via the new-domain-lookup feature.

ignore-list-file

New in version 5.1.0.

Path to a file with a list of domains. File should have one domain per line, with no extra characters or comments. See new-domain-ignore-list.

new-domain-pb-tag

New in version 4.2.0.

  • String
  • Default: pdns-nod
  • YAML setting: nod.pb_tag

If protobuf is configured, then this tag will be added to all protobuf response messages when a new domain is observed.

network-timeout

Number of milliseconds to wait for a remote authoritative server to respond. If the number of concurrent requests is high, the :program:Recursor uses a lower value.

non-resolving-ns-max-fails

New in version 4.5.0.

Number of failed address resolves of a nameserver name to start throttling it, 0 is disabled. Nameservers matching dont-throttle-names will not be throttled.

non-resolving-ns-throttle-time

New in version 4.5.0.

Number of seconds to throttle a nameserver with a name failing to resolve.

nothing-below-nxdomain

New in version 4.3.0.

The type of RFC 8020 handling using cached NXDOMAIN responses. This RFC specifies that NXDOMAIN means that the DNS tree under the denied name MUST be empty. When an NXDOMAIN exists in the cache for a shorter name than the qname, no lookup is done and an NXDOMAIN is sent to the client.

For instance, when foo.example.net is negatively cached, any query matching *.foo.example.net will be answered with NXDOMAIN directly without consulting authoritative servers.

no
No RFC 8020 processing is done.
dnssec
RFC 8020 processing is only done using cached NXDOMAIN records that are DNSSEC validated.
yes
RFC 8020 processing is done using any non-Bogus NXDOMAIN record available in the cache.

nsec3-max-iterations

New in version 4.1.0.

Changed in version 4.5.2: Default is now 150, was 2500 before.

Changed in version 5.0.0: Default is now 50, was 150 before.

Maximum number of iterations allowed for an NSEC3 record. If an answer containing an NSEC3 record with more iterations is received, its DNSSEC validation status is treated as Insecure.

max-rrsigs-per-record

New in version 5.0.2.

New in version 4.9.3.

New in version 4.8.6.

Maximum number of RRSIGs we are willing to cryptographically check when validating a given record. Expired or not yet incepted RRSIGs do not count toward to this limit.

max-nsec3s-per-record

New in version 5.0.2.

New in version 4.9.3.

New in version 4.8.6.

Maximum number of NSEC3s to consider when validating a given denial of existence.

max-signature-validations-per-query

New in version 5.0.2.

New in version 4.9.3.

New in version 4.8.6.

Maximum number of RRSIG signatures we are willing to validate per incoming query.

max-nsec3-hash-computations-per-query

New in version 5.0.2.

New in version 4.9.3.

New in version 4.8.6.

Maximum number of NSEC3 hashes that we are willing to compute during DNSSEC validation, per incoming query.

aggressive-cache-max-nsec3-hash-cost

New in version 5.0.2.

New in version 4.9.3.

New in version 4.8.6.

Maximum estimated NSEC3 cost for a given query to consider aggressive use of the NSEC3 cache. The cost is estimated based on a heuristic taking the zone’s NSEC3 salt and iterations parameters into account, as well at the number of labels of the requested name. For example a query for a name like a.b.c.d.e.f.example.com. in an example.com zone. secured with NSEC3 and 10 iterations (NSEC3 iterations count of 9) and an empty salt will have an estimated worst-case cost of 10 (iterations) * 6 (number of labels) = 60. The aggressive NSEC cache is an optimization to reduce the number of queries to authoritative servers, which is especially useful when a zone is under pseudo-random subdomain attack, and we want to skip it the zone parameters make it expensive.

max-ds-per-zone

New in version 5.0.2.

New in version 4.9.3.

New in version 4.8.6.

Maximum number of DS records to consider when validating records inside a zone.

max-dnskeys

New in version 5.0.2.

New in version 4.9.3.

New in version 4.8.6.

Maximum number of DNSKEYs with the same algorithm and tag to consider when validating a given record. Setting this value to 1 effectively denies DNSKEY tag collisions in a zone.

packetcache-ttl

Changed in version 4.9.0: The default was changed from 3600 (1 hour) to 86400 (24 hours).

Maximum number of seconds to cache an item in the packet cache, no matter what the original TTL specified.

packetcache-negative-ttl

New in version 4.9.0.

Maximum number of seconds to cache an NxDomain or NoData answer in the packetcache. This setting’s maximum is capped to packetcache-ttl. i.e. setting packetcache-ttl=15 and keeping packetcache-negative-ttl at the default will lower packetcache-negative-ttl to 15.

packetcache-servfail-ttl

‘versionchanged’: (‘4.0.0’, “This setting’s maximum is capped to packetcache-ttl.
i.e. setting packetcache-ttl=15 and keeping packetcache-servfail-ttl at the default will lower packetcache-servfail-ttl to 15.”)

Maximum number of seconds to cache an answer indicating a failure to resolve in the packet cache. Before version 4.6.0 only ServFail answers were considered as such. Starting with 4.6.0, all responses with a code other than NoError and NXDomain, or without records in the answer and authority sections, are considered as a failure to resolve. Since 4.9.0, negative answers are handled separately from resolving failures.

packetcache-shards

New in version 4.9.0.

Sets the number of shards in the packet cache. If you have high contention as reported by packetcache-contented/packetcache-acquired, you can try to enlarge this value or run with fewer threads.

pdns-distributes-queries

Changed in version 4.9.0: Default changed to no, previously it was yes.

If set, PowerDNS will use distinct threads to listen to client sockets and distribute that work to worker-threads using a hash of the query. This feature should maximize the cache hit ratio on versions before 4.9.0. To use more than one thread set distributor-threads in version 4.2.0 or newer. Enabling should improve performance on systems where reuseport does not have the effect of balancing the queries evenly over multiple worker threads.

protobuf-use-kernel-timestamp

New in version 4.2.0.

Whether to compute the latency of responses in protobuf messages using the timestamp set by the kernel when the query packet was received (when available), instead of computing it based on the moment we start processing the query.

proxy-protocol-from

New in version 4.4.0.

Changed in version 5.0.5: YAML settings only: previously this was defined as a string instead of a sequence

Ranges that are required to send a Proxy Protocol version 2 header in front of UDP and TCP queries, to pass the original source and destination addresses and ports to the recursor, as well as custom values. Queries that are not prefixed with such a header will not be accepted from clients in these ranges. Queries prefixed by headers from clients that are not listed in these ranges will be dropped.

Note that once a Proxy Protocol header has been received, the source address from the proxy header instead of the address of the proxy will be checked against the allow-from ACL.

The dnsdist docs have more information about the PROXY protocol.

proxy-protocol-exceptions

New in version 5.1.0.

If set, clients sending from an address in proxy-protocol-from to a address:port listed here are excluded from using the Proxy Protocol. If no port is specified, port 53 is assumed. This is typically used to provide an easy to use address and port to send debug queries to.

proxy-protocol-maximum-size

New in version 4.4.0.

The maximum size, in bytes, of a Proxy Protocol payload (header, addresses and ports, and TLV values). Queries with a larger payload will be dropped.

public-suffix-list-file

New in version 4.2.0.

Path to the Public Suffix List file, if any. If set, PowerDNS will try to load the Public Suffix List from this file instead of using the built-in list. The PSL is used to group the queries by relevant domain names when displaying the top queries.

qname-minimization

New in version 4.3.0.

Enable Query Name Minimization. This implements a relaxed form of Query Name Mimimization as described in RFC 9156.

qname-max-minimize-count

New in version 5.0.0.

Max minimize count parameter, described in RFC 9156. This is the maximum number of iterations of the Query Name Minimization Algorithm.

qname-minimize-one-label

New in version 5.0.0.

Minimize one label parameter, described in RFC 9156. The value for the number of iterations of the Query Name Minimization Algorithm that should only have one label appended. This value has precedence over qname-max-minimize-count.

query-local-address

Changed in version 4.4.0: IPv6 addresses can be set with this option as well.

  • Comma separated list of IP addresses or subnets, negation supported
  • Default: 0.0.0.0
  • YAML setting: outgoing.source_address

Note

While subnets and their negations are syntactically accepted, the handling of subnets has not been implemented yet. Only individual IP addresses can be listed.

Send out local queries from this address, or addresses. By adding multiple addresses, increased spoofing resilience is achieved. When no address of a certain address family is configured, there are no queries sent with that address family. In the default configuration this means that IPv6 is not used for outgoing queries.

quiet

Don’t log queries.

record-cache-locked-ttl-perc

New in version 4.8.0.

Replace record sets in the record cache only after this percentage of the original TTL has passed. The PowerDNS Recursor already has several mechanisms to protect against spoofing attempts. This adds an extra layer of protection—as it limits the window of time cache updates are accepted—at the cost of a less efficient record cache.

The default value of 0 means no extra locking occurs. When non-zero, record sets received (e.g. in the Additional Section) will not replace existing record sets in the record cache until the given percentage of the original TTL has expired. A value of 100 means only expired record sets will be replaced.

There are a few cases where records will be replaced anyway:

  • Record sets that are expired will always be replaced.
  • Authoritative record sets will replace unauthoritative record sets unless DNSSEC validation of the new record set failed.
  • If the new record set belongs to a DNSSEC-secure zone and successfully passed validation it will replace an existing entry.
  • Record sets produced by refresh-on-ttl-perc tasks will also replace existing record sets.

record-cache-shards

New in version 4.4.0.

Sets the number of shards in the record cache. If you have high contention as reported by record-cache-contented/record-cache-acquired, you can try to enlarge this value or run with fewer threads.

refresh-on-ttl-perc

New in version 4.5.0.

Sets the ‘refresh almost expired’ percentage of the record cache. Whenever a record is fetched from the packet or record cache and only refresh-on-ttl-perc percent or less of its original TTL is left, a task is queued to refetch the name/type combination to update the record cache. In most cases this causes future queries to always see a non-expired record cache entry. A typical value is 10. If the value is zero, this functionality is disabled.

reuseport

Changed in version 4.9.0: The default is changed to yes, previously it was no. If SO_REUSEPORT support is not available, the setting defaults to no.

If SO_REUSEPORT support is available, allows multiple threads and processes to open listening sockets for the same port.

Since 4.1.0, when pdns-distributes-queries is disabled and reuseport is enabled, every worker-thread will open a separate listening socket to let the kernel distribute the incoming queries instead of running a distributor thread (which could otherwise be a bottleneck) and avoiding thundering herd issues, thus leading to much higher performance on multi-core boxes.

rng

Changed in version 4.9.0: This setting is no longer used.

  • String
  • Default: auto
  • YAML setting does not exist
  • String
  • Default: auto
Specify which random number generator to use. Permissible choices are
  • auto - choose automatically
  • sodium - Use libsodium randombytes_uniform
  • openssl - Use libcrypto RAND_bytes
  • getrandom - Use libc getrandom, falls back to urandom if it does not really work
  • arc4random - Use BSD arc4random_uniform
  • urandom - Use /dev/urandom
  • kiss - Use simple settable deterministic RNG. FOR TESTING PURPOSES ONLY!

root-nx-trust

Changed in version 4.0.0: Default is yes now, was no before 4.0.0

If set, an NXDOMAIN from the root-servers will serve as a blanket NXDOMAIN for the entire TLD the query belonged to. The effect of this is far fewer queries to the root-servers.

save-parent-ns-set

New in version 4.7.0.

If set, a parent (non-authoritative) NS set is saved if it contains more entries than a newly encountered child (authoritative) NS set for the same domain. The saved parent NS set is tried if resolution using the child NS set fails.

security-poll-suffix

Domain name from which to query security update notifications. Setting this to an empty string disables secpoll.

serve-rfc1918

This makes the server authoritatively aware of: 10.in-addr.arpa, 168.192.in-addr.arpa, 16-31.172.in-addr.arpa, which saves load on the AS112 servers. Individual parts of these zones can still be loaded or forwarded.

serve-stale-extensions

New in version 4.8.0.

Maximum number of times an expired record’s TTL is extended by 30s when serving stale. Extension only occurs if a record cannot be refreshed. A value of 0 means the Serve Stale mechanism is not used. To allow records becoming stale to be served for an hour, use a value of 120. See Serve Stale for a description of the Serve Stale mechanism.

server-down-max-fails

If a server has not responded in any way this many times in a row, no longer send it any queries for server-down-throttle-time seconds. Afterwards, we will try a new packet, and if that also gets no response at all, we again throttle for server-down-throttle-time seconds. Even a single response packet will drop the block.

server-down-throttle-time

Throttle a server that has failed to respond server-down-max-fails times for this many seconds.

bypass-server-throttling-probability

New in version 5.0.0.

This setting determines the probability of a server marked down to be used anyway. A value of n means that the chance of a server marked down still being used after it wins speed selection is is 1/n. If this setting is zero throttled servers will never be selected to be used anyway.

server-id

The reply given by The PowerDNS recursor to a query for ‘id.server’ with its hostname, useful for in clusters. When a query contains the NSID EDNS0 Option, this value is returned in the response as the NSID value.

This setting can be used to override the answer given to these queries. Set to ‘disabled’ to disable NSID and ‘id.server’ answers.

Query example (where 192.0.2.14 is your server):

dig @192.0.2.14 CHAOS TXT id.server.
dig @192.0.2.14 example.com IN A +nsid

setgid

PowerDNS can change its user and group id after binding to its socket. Can be used for better security.

setuid

PowerDNS can change its user and group id after binding to its socket. Can be used for better security.

signature-inception-skew

New in version 4.1.5.

Changed in version 4.2.0: Default is now 60, was 0 before.

Allow the signature inception to be off by this number of seconds. Negative values are not allowed.

single-socket

Use only a single socket for outgoing queries.

snmp-agent

New in version 4.1.0.

If set to true and PowerDNS has been compiled with SNMP support, it will register as an SNMP agent to provide statistics and be able to send traps.

snmp-master-socket

New in version 4.1.0.

Deprecated since version 4.5.0: Use snmp-daemon-socket.

  • String
  • Default: (empty)
  • YAML setting does not exist

snmp-daemon-socket

New in version 4.5.0.

If not empty and snmp-agent is set to true, indicates how PowerDNS should contact the SNMP daemon to register as an SNMP agent.

socket-dir

Where to store the control socket and pidfile. The default depends on LOCALSTATEDIR or the --with-socketdir setting when building (usually /var/run or /run).

When using chroot the default becomes /. The default value is overruled by the RUNTIME_DIRECTORY environment variable when that variable has a value (e.g. under systemd).

socket-group

Group and mode of the controlsocket. Owner and group can be specified by name, mode is in octal.

socket-mode

Mode of the controlsocket. Owner and group can be specified by name, mode is in octal.

socket-owner

Owner of the controlsocket. Owner and group can be specified by name, mode is in octal.

spoof-nearmiss-max

Changed in version 4.5.0: Older versions used 20 as the default value.

If set to non-zero, PowerDNS will assume it is being spoofed after seeing this many answers with the wrong id.

stack-cache-size

New in version 4.9.0.

Maximum number of mthread stacks that can be cached for later reuse, per thread. Caching these stacks reduces the CPU load at the cost of a slightly higher memory usage, each cached stack consuming stack-size bytes of memory. It makes no sense to cache more stacks than the value of max-mthreads, since there will never be more stacks than that in use at a given time.

stack-size

Size in bytes of the stack of each mthread.

statistics-interval

New in version 4.1.0.

Interval between logging statistical summary on recursor performance. Use 0 to disable.

stats-api-blacklist

New in version 4.2.0.

Deprecated since version 4.5.0: Use stats-api-disabled-list.

  • Comma separated list of strings
  • Default:
  • YAML setting does not exist

stats-api-disabled-list

New in version 4.5.0.

  • Comma separated list of strings
  • Default: cache-bytes, packetcache-bytes, special-memory-usage, ecs-v4-response-bits-*, ecs-v6-response-bits-*
  • YAML setting: recursor.stats_api_disabled_list

A list of comma-separated statistic names, that are disabled when retrieving the complete list of statistics via the API for performance reasons. These statistics can still be retrieved individually by specifically asking for it.

stats-carbon-blacklist

New in version 4.2.0.

Deprecated since version 4.5.0: Use stats-carbon-disabled-list.

  • Comma separated list of strings
  • Default:
  • YAML setting does not exist

stats-carbon-disabled-list

New in version 4.5.0.

  • Comma separated list of strings
  • Default: cache-bytes, packetcache-bytes, special-memory-usage, ecs-v4-response-bits-*, ecs-v6-response-bits-*, cumul-answers-*, cumul-auth4answers-*, cumul-auth6answers-*
  • YAML setting: recursor.stats_carbon_disabled_list

A list of comma-separated statistic names, that are prevented from being exported via carbon for performance reasons.

stats-rec-control-blacklist

New in version 4.2.0.

Deprecated since version 4.5.0: Use stats-rec-control-disabled-list.

  • Comma separated list of strings
  • Default:
  • YAML setting does not exist

stats-rec-control-disabled-list

New in version 4.5.0.

  • Comma separated list of strings
  • Default: cache-bytes, packetcache-bytes, special-memory-usage, ecs-v4-response-bits-*, ecs-v6-response-bits-*, cumul-answers-*, cumul-auth4answers-*, cumul-auth6answers-*
  • YAML setting: recursor.stats_rec_control_disabled_list

A list of comma-separated statistic names, that are disabled when retrieving the complete list of statistics via rec_control get-all, for performance reasons. These statistics can still be retrieved individually.

stats-ringbuffer-entries

Number of entries in the remotes ringbuffer, which keeps statistics on who is querying your server. Can be read out using rec_control top-remotes.

stats-snmp-blacklist

New in version 4.2.0.

Deprecated since version 4.5.0: Use stats-snmp-disabled-list.

  • Comma separated list of strings
  • Default:
  • YAML setting does not exist

stats-snmp-disabled-list

New in version 4.5.0.

  • Comma separated list of strings
  • Default: cache-bytes, packetcache-bytes, special-memory-usage, ecs-v4-response-bits-*, ecs-v6-response-bits-*
  • YAML setting: recursor.stats_snmp_disabled_list

A list of comma-separated statistic names, that are prevented from being exported via SNMP, for performance reasons.

structured-logging

New in version 4.6.0.

Changed in version 5.1.0: Disabling structured logging is not supported

Prefer structured logging when both an old style and a structured log messages is available.

structured-logging-backend

New in version 4.8.0.

Changed in version 5.1.0: The JSON backend was added

The backend used for structured logging output. This setting must be set on the command line (--structured-logging-backend=...) to be effective. Available backends are:

  • default: use the traditional logging system to output structured logging information.
  • systemd-journal: use systemd-journal. When using this backend, provide -o verbose or simular output option to journalctl to view the full information.
  • json: JSON objects are written to the standard error stream.

See Structured Logging Dictionary for more details.

tcp-fast-open

New in version 4.1.0.

Enable TCP Fast Open support, if available, on the listening sockets. The numerical value supplied is used as the queue size, 0 meaning disabled. See TCP Fast Open Support.

tcp-fast-open-connect

New in version 4.5.0.

Enable TCP Fast Open Connect support, if available, on the outgoing connections to authoritative servers. See TCP Fast Open Support.

tcp-out-max-idle-ms

New in version 4.6.0.

Time outgoing TCP/DoT connections are left idle in milliseconds or 0 if no limit. After having been idle for this time, the connection is eligible for closing.

tcp-out-max-idle-per-auth

New in version 4.6.0.

Maximum number of idle outgoing TCP/DoT connections to a specific IP per thread, 0 means do not keep idle connections open.

tcp-out-max-queries

Maximum total number of queries per outgoing TCP/DoT connection, 0 means no limit. After this number of queries, the connection is closed and a new one will be created if needed.

tcp-out-max-idle-per-thread

New in version 4.6.0.

Maximum number of idle outgoing TCP/DoT connections per thread, 0 means do not keep idle connections open.

threads

Spawn this number of threads on startup.

tcp-threads

New in version 5.0.0.

Spawn this number of TCP processing threads on startup.

trace

One of no, yes or fail. If turned on, output impressive heaps of logging. May destroy performance under load. To log only queries resulting in a ServFail answer from the resolving process, this value can be set to fail, but note that the performance impact is still large. Also note that queries that do produce a result but with a failing DNSSEC validation are not written to the log

udp-source-port-min

New in version 4.2.0.

This option sets the low limit of UDP port number to bind on.

In combination with udp-source-port-max it configures the UDP port range to use. Port numbers are randomized within this range on initialization, and exceptions can be configured with udp-source-port-avoid

udp-source-port-max

New in version 4.2.0.

This option sets the maximum limit of UDP port number to bind on.

See udp-source-port-min.

udp-source-port-avoid

New in version 4.2.0.

A list of comma-separated UDP port numbers to avoid when binding. Ex: 5300,11211

See udp-source-port-min.

udp-truncation-threshold

Changed in version 4.2.0: Before 4.2.0, the default was 1680.

EDNS0 allows for large UDP response datagrams, which can potentially raise performance. Large responses however also have downsides in terms of reflection attacks. This setting limits the accepted size. Maximum value is 65535, but values above 4096 should probably not be attempted.

To know why 1232, see the note at edns-outgoing-bufsize.

unique-response-tracking

New in version 4.2.0.

Whether to track unique DNS responses, i.e. never seen before combinations of the triplet (query name, query type, RR[rrname, rrtype, rrdata]). This can be useful for tracking potentially suspicious domains and behaviour, e.g. DNS fast-flux. If protobuf is enabled and configured, then the Protobuf Response message will contain a flag with udr set to true for each RR that is considered unique, i.e. never seen before. This feature uses a probabilistic data structure (stable bloom filter) to track unique responses, which can have false positives as well as false negatives, thus it is a best-effort feature. Increasing the number of cells in the SBF using the unique-response-db-size setting can reduce FPs and FNs.

unique-response-log

New in version 4.2.0.

Whether to log when a unique response is detected. The log line looks something like:

Oct 24 12:11:27 Unique response observed: qname=foo.com qtype=A rrtype=AAAA rrname=foo.com rrcontent=1.2.3.4

unique-response-db-size

New in version 4.2.0.

The default size of the stable bloom filter used to store previously observed responses is 67108864. To change the number of cells, use this setting. For each cell, the SBF uses 1 bit of memory, and one byte of disk for the persistent file. If there are already persistent files saved to disk, this setting will have no effect unless you remove the existing files.

unique-response-history-dir

New in version 4.2.0.

This setting controls which directory is used to store the on-disk cache of previously observed responses.

The default depends on LOCALSTATEDIR when building the software. Usually this comes down to /var/lib/pdns-recursor/udr or /usr/local/var/lib/pdns-recursor/udr).

The newly observed domain feature uses a stable bloom filter to store a history of previously observed responses. The data structure is synchronized to disk every 10 minutes, and is also initialized from disk on startup. This ensures that previously observed responses are preserved across recursor restarts. If you change the unique-response-db-size, you must remove any files from this directory.

unique-response-pb-tag

New in version 4.2.0.

If protobuf is configured, then this tag will be added to all protobuf response messages when a unique DNS response is observed.

unique-response-ignore-list

New in version 5.1.0.

This setting is a list of all domains (and implicitly all subdomains) that will never be considered for a new unique domain request. For example, if the domain ‘example.com’ is in the list, then ‘foo.bar.example.com’ will never be considered for a new unique domain request.

unique-response-ignore-list-file

New in version 5.1.0.

Path to a file with a list of domains. File should have one domain per line, with no extra characters or comments. See unique-response-ignore-list.

use-incoming-edns-subnet

Whether to process and pass along a received EDNS Client Subnet to authoritative servers. The ECS information will only be sent for netmasks and domains listed in edns-subnet-allow-list and will be truncated if the received scope exceeds ecs-ipv4-bits for IPv4 or ecs-ipv6-bits for IPv6.

version-string

By default, PowerDNS replies to the ‘version.bind’ query with its version number. Security conscious users may wish to override the reply PowerDNS issues.

webserver

Start the webserver (for REST API).

webserver-address

IP address for the webserver to listen on.

webserver-allow-from

Changed in version 4.1.0: Default is now 127.0.0.1,::1, was 0.0.0.0/0,::/0 before.

  • Comma separated list of IP addresses or subnets, negation supported
  • Default: 127.0.0.1, ::1
  • YAML setting: webservice.allow_from

These IPs and subnets are allowed to access the webserver. Note that specifying an IP address without a netmask uses an implicit netmask of /32 or /128.

webserver-hash-plaintext-credentials

New in version 4.6.0.

Whether passwords and API keys supplied in the configuration as plaintext should be hashed during startup, to prevent the plaintext versions from staying in memory. Doing so increases significantly the cost of verifying credentials and is thus disabled by default. Note that this option only applies to credentials stored in the configuration as plaintext, but hashed credentials are supported without enabling this option.

webserver-loglevel

New in version 4.2.0.

One of none, normal, detailed. The amount of logging the webserver must do. ‘none’ means no useful webserver information will be logged. When set to ‘normal’, the webserver will log a line per request that should be familiar:

[webserver] e235780e-a5cf-415e-9326-9d33383e739e 127.0.0.1:55376 'GET /api/v1/servers/localhost/bla HTTP/1.1' 404 196

When set to ‘detailed’, all information about the request and response are logged:

[webserver] e235780e-a5cf-415e-9326-9d33383e739e Request Details:
[webserver] e235780e-a5cf-415e-9326-9d33383e739e  Headers:
[webserver] e235780e-a5cf-415e-9326-9d33383e739e   accept: text/html,application/xhtml+xml,application/xml;q=0.9,*/*;q=0.8
[webserver] e235780e-a5cf-415e-9326-9d33383e739e   accept-encoding: gzip, deflate
[webserver] e235780e-a5cf-415e-9326-9d33383e739e   accept-language: en-US,en;q=0.5
[webserver] e235780e-a5cf-415e-9326-9d33383e739e   connection: keep-alive
[webserver] e235780e-a5cf-415e-9326-9d33383e739e   dnt: 1
[webserver] e235780e-a5cf-415e-9326-9d33383e739e   host: 127.0.0.1:8081
[webserver] e235780e-a5cf-415e-9326-9d33383e739e   upgrade-insecure-requests: 1
[webserver] e235780e-a5cf-415e-9326-9d33383e739e   user-agent: Mozilla/5.0 (X11; Linux x86_64; rv:64.0) Gecko/20100101 Firefox/64.0
[webserver] e235780e-a5cf-415e-9326-9d33383e739e  No body
[webserver] e235780e-a5cf-415e-9326-9d33383e739e Response details:
[webserver] e235780e-a5cf-415e-9326-9d33383e739e  Headers:
[webserver] e235780e-a5cf-415e-9326-9d33383e739e   Connection: close
[webserver] e235780e-a5cf-415e-9326-9d33383e739e   Content-Length: 49
[webserver] e235780e-a5cf-415e-9326-9d33383e739e   Content-Type: text/html; charset=utf-8
[webserver] e235780e-a5cf-415e-9326-9d33383e739e   Server: PowerDNS/0.0.15896.0.gaba8bab3ab
[webserver] e235780e-a5cf-415e-9326-9d33383e739e  Full body:
[webserver] e235780e-a5cf-415e-9326-9d33383e739e   <!html><title>Not Found</title><h1>Not Found</h1>
[webserver] e235780e-a5cf-415e-9326-9d33383e739e 127.0.0.1:55376 'GET /api/v1/servers/localhost/bla HTTP/1.1' 404 196

The value between the hooks is a UUID that is generated for each request. This can be used to find all lines related to a single request.

Note

The webserver logs these line on the NOTICE level. The loglevel seting must be 5 or higher for these lines to end up in the log.

webserver-password

Changed in version 4.6.0: This setting now accepts a hashed and salted version.

Password required to access the webserver. Since 4.6.0 the password can be hashed and salted using rec_control hash-password instead of being present in the configuration in plaintext, but the plaintext version is still supported.

webserver-port

TCP port where the webserver should listen on.

write-pid

If a PID file should be written to socket-dir

x-dnssec-names

New in version 4.5.0.

List of names whose DNSSEC validation metrics will be counted in a separate set of metrics that start with x-dnssec-result-. The names are suffix-matched. This can be used to not count known failing (test) name validations in the ordinary DNSSEC metrics.

system-resolver-ttl

New in version 5.1.0.

Sets TTL in seconds of the system resolver feature. If not equal to zero names can be used for forwarding targets. The names will be resolved by the system resolver configured in the OS.

The TTL is used as a time to live to see if the names used in forwarding resolve to a different address than before. If the TTL is expired, a re-resolve will be done by the next iteration of the check function; if a change is detected, the recursor performs an equivalent of rec_control reload-zones.

Make sure the recursor itself is not used by the system resolver! Default is 0 (not enabled). A suggested value is 60.

system-resolver-interval

New in version 5.1.0.

Sets the check interval (in seconds) of the system resolver feature. All names known by the system resolver subsystem are periodically checked for changing values.

If the TTL of a name has expired, it is checked by re-resolving it. if a change is detected, the recursor performs an equivalent of rec_control reload-zones.

This settings sets the interval between the checks. If set to zero (the default), the value system-resolver-ttl is used.

system-resolver-self-resolve-check

New in version 5.1.0.

Warn on potential self-resolve. If this check draws the wrong conclusion, you can disable it.