Newer
Older
*************
Configuration
*************
====================
The following example presents a simple configuration file
which can be used as a base for your Knot DNS setup::
# Example of a very simple Knot DNS configuration.
server:
listen: 0.0.0.0@53
listen: ::@53
zone:
- domain: example.com
storage: /var/lib/knot/zones/
file: example.com.zone
Now let's walk through this configuration step by step:
- The :ref:`server_listen` statement in the :ref:`server section<Server section>`
defines where the server will listen for incoming connections.
We have defined the server to listen on all available IPv4 and IPv6 addresses,
all on port 53.
- The :ref:`zone section<Zone section>` defines the zones that the server will
serve. In this case, we defined one zone named *example.com* which is stored
in the zone file :file:`/var/lib/knot/zones/example.com.zone`.
- The :ref:`log section<Logging section>` defines the log facilities for
the server. In this example, we told Knot DNS to send its log messages with
the severity ``info`` or more serious to the syslog (or systemd journal).
For detailed description of all configuration items see
:ref:`Configuration Reference`.
Zone templates
==============
A zone template allows a single zone configuration to be shared among several
zones. There is no inheritance between templates; they are exclusive. The
``default`` template identifier is reserved for the default template::
template:
- id: default
storage: /var/lib/knot/master
semantic-checks: on
- id: signed
storage: /var/lib/knot/signed
- id: slave
storage: /var/lib/knot/slave
zone:
- domain: example1.com # Uses default template
- domain: example2.com # Uses default template
semantic-checks: off # Override default settings
- domain: example.cz
template: signed
- domain: example1.eu
template: slave
master: master1
- domain: example2.eu
template: slave
master: master2
.. NOTE::
Each template option can be explicitly overridden in zone-specific configuration.
Access control list (ACL)
=========================
Some types of incoming DNS requests must be authorized before they can be
processed by the server. A zone can have configured :ref:`zone_acl` which is
a sequence of :ref:`rules <ACL section>` describing what requests are authorized.
By default if :ref:`automatic ACL <server_automatic-acl>` is not enabled, all requests,
which require authorization, are denied.
Every ACL rule can allow or deny one or more request types based on the
source IP address, network subnet, or address range and/or if the request is
secured by a given TSIG key. See :doc:`keymgr -t<man_keymgr>` on how
to generate a TSIG key.
If there are multiple ACL rules assigned to a zone, they are applied in the
specified order of the :ref:`zone_acl` configuration. The first rule that matches
the given request is applied and the remaining rules are ignored. Some examples::
- id: address_rule
address: [2001:db8::1, 192.168.2.0/24]
action: transfer
address: 192.168.2.100
action: transfer
- domain: acl1.example.com
acl: [deny_rule, address_rule] # Allow some addresses with an exception
- id: key1 # The real TSIG key name
algorithm: hmac-sha256
secret: 4Tc0K1QkcMCs7cOW2LuSWnxQY0qysdvsZlSb4yTN9pA=
acl:
- id: deny_all
address: 192.168.3.0/24
deny: on # No action specified and deny on implies denial of all actions
key: key1 # Access based just on TSIG key
acl: [deny_all, key_rule] # Allow with the TSIG except for the subnet
In the case of dynamic DNS updates, some additional conditions may be specified
for more granular filtering. See more in the section :ref:`Restricting dynamic updates`.
.. NOTE::
If more conditions (address ranges and/or a key)
are given in a single ACL rule, all of them have to be satisfied for the rule to match.
.. TIP::
In order to restrict regular DNS queries, use module :ref:`queryacl<mod-queryacl>`.
Secondary (slave) zone
======================
Knot DNS doesn't strictly differ between primary (formerly known as master)
and secondary (formerly known as slave) zones. The only requirement for a secondary
zone is to have a :ref:`zone_master` statement set. For effective zone synchronization,
incoming zone change notifications (NOTIFY), which require authorization, can be
enabled using :ref:`automatic ACL <server_automatic-acl>` or :ref:`explicit ACL <zone_acl>`
configuration. Optional transaction authentication (TSIG) is supported for both
zone transfers and zone notifications::
server:
automatic-acl: on # Enabled automatic ACL
key:
- id: xfr_notify_key # Common TSIG key for XFR an NOTIFY
algorithm: hmac-sha256
secret: VFRejzw8h4M7mb0xZKRFiZAfhhd1eDGybjqHr2FV3vc=
remote:
- id: primary
address: [2001:DB8:1::1, 192.168.1.1] # Primary server IP addresses
# via: [2001:DB8:2::1, 10.0.0.1] # Local source addresses (optional)
key: xfr_notify_key # TSIG key (optional)
zone:
- domain: example.com
master: primary # Primary remote(s)
An example of explicit ACL with different TSIG keys for zone transfers
and notifications::
- id: notify_key # TSIG key for NOTIFY
algorithm: hmac-sha256
secret: uBbhV4aeSS4fPd+wF2ZIn5pxOMF35xEtdq2ibi2hHEQ=
- id: xfr_key # TSIG key for XFR
algorithm: hmac-sha256
secret: VFRejzw8h4M7mb0xZKRFiZAfhhd1eDGybjqHr2FV3vc=
- id: primary
address: [2001:DB8:1::1, 192.168.1.1] # Primary server IP addresses
# via: [2001:DB8:2::1, 10.0.0.1] # Local source addresses if needed
key: xfr_key # Optional TSIG key
- id: notify_from_primary # ACL rule for NOTIFY from primary
address: [2001:DB8:1::1, 192.168.1.1] # Primary addresses (optional)
key: notify_key # TSIG key (optional)
zone:
- domain: example.com
master: primary # Primary remote(s)
acl: notify_from_primary # Explicit ACL(s)
Note that the :ref:`zone_master` option accepts a list of remotes, which are
queried for a zone refresh sequentially in the specified order. When the server
receives a zone change notification from a listed remote, only that remote is
used for a subsequent zone transfer.
.. NOTE::
When transferring a lot of zones, the server may easily get into a state
where all available ports are in the TIME_WAIT state, thus transfers
cease until the operating system closes the ports for good. There are
several ways to work around this:
* Allow reusing of ports in TIME_WAIT (sysctl -w net.ipv4.tcp_tw_reuse=1)
* Shorten TIME_WAIT timeout (tcp_fin_timeout)
* Increase available local port count
Primary (master) zone
=====================
A zone is considered primary if it doesn't have :ref:`zone_master` set. As
outgoing zone transfers (XFR) require authorization, it must be enabled
using :ref:`automatic ACL <server_automatic-acl>` or :ref:`explicit ACL <zone_acl>`
configuration. Outgoing zone change notifications (NOTIFY) to remotes can be
set by configuring :ref:`zone_notify`. Transaction authentication
(TSIG) is supported for both zone transfers and zone notifications::
server:
automatic-acl: on # Enabled automatic ACL
- id: xfr_notify_key # Common TSIG key for XFR an NOTIFY
algorithm: hmac-sha256
secret: VFRejzw8h4M7mb0xZKRFiZAfhhd1eDGybjqHr2FV3vc=
remote:
- id: secondary
address: [2001:DB8:1::1, 192.168.1.1] # Secondary server IP addresses
# via: [2001:DB8:2::1, 10.0.0.1] # Local source addresses (optional)
key: xfr_notify_key # TSIG key (optional)
acl:
- id: local_xfr # Allow XFR to localhost without TSIG
address: [::1, 127.0.0.1]
action: transfer
zone:
- domain: example.com
notify: secondary # Secondary remote(s)
acl: local_xfr # Explicit ACL for local XFR
Note that the :ref:`zone_notify` option accepts a list of remotes, which are
all notified sequentially in the specified order.
A secondary zone may serve as a primary zone for a different set of remotes
at the same time.
Dynamic updates
===============
Dynamic updates for the zone are allowed via proper ACL rule with the
``update`` action. If the zone is configured as a secondary and a DNS update
message is accepted, the server forwards the message to its primary master.
The primary master's response is then forwarded back to the originator.
However, if the zone is configured as a primary, the update is accepted and
processed::
acl:
- id: update_acl
address: 192.168.3.0/24
action: update
zone:
- domain: example.com
acl: update_acl
.. _Restricting dynamic updates:
Restricting dynamic updates
---------------------------
There are several additional ACL options for dynamic DNS updates which affect
the request classification based on the update contents.
Updates can be restricted to specific resource record types::
- id: type_rule
action: update
update-type: [A, AAAA, MX] # Updated records must match one of the specified types
Another possibility is restriction on the owner name of updated records. The option
:ref:`acl_update-owner` is used to select the source of domain
names which are used for the comparison. And the option :ref:`acl_update-owner-match`
specifies the required relation between the record owner and the reference domain
names. Example::
acl:
- id: owner_rule1
action: update
update-owner: name # Updated record owners are restricted by the next conditions
update-owner-match: equal # The record owner must exactly match one name from the next list
update-owner-name: [foo, bar.] # Reference domain names
.. NOTE::
If the specified owner name is non-FQDN (e.g. ``foo``), it's considered relatively
to the effective zone name. So it can apply to more zones
(e.g. ``foo.example.com.`` or ``foo.example.net.``). Alternatively, if the
name is FQDN (e.g. ``bar.``), the rule only applies to this name.
If the reference domain name is the zone name, the following variant can be used::
acl:
- id: owner_rule2
action: update
update-owner: zone # The reference name is the zone name
update-owner-match: sub # Any record owner matches except for the zone name itself
template:
- id: default
acl: owner_rule2
zone:
- domain: example.com.
- domain: example.net.
The last variant is for the cases where the reference domain name is a TSIG key name,
which must be used for the transaction security::
- id: example.com # Key names are always considered FQDN
...
- id: steve.example.net
...
- id: jane.example.net
...
- id: owner_rule3_com
action: update
update-owner: key # The reference name is the TSIG key name
update-owner-match: sub # The record owner must be a subdomain of the key name
key: [example.com] # One common key for updating all non-apex records
- id: owner_rule3_net
action: update
update-owner: key # The reference name is the TSIG key name
update-owner-match: equal # The record owner must exactly match the used key name
key: [steve.example.net, jane.example.net] # Keys for updating specific zone nodes
zone:
- domain: example.com.
acl: owner_rule3_com
acl: owner_rule3_net
Automatic DNSSEC signing
========================
Knot DNS supports automatic DNSSEC signing of zones. The signing
can operate in two modes:
1. :ref:`Automatic key management <dnssec-automatic-zsk-management>`.
In this mode, the server maintains signing keys. New keys are generated
according to assigned policy and are rolled automatically in a safe manner.
No zone operator intervention is necessary.
2. :ref:`Manual key management <dnssec-manual-key-management>`.
In this mode, the server maintains zone signatures only. The signatures
are kept up-to-date and signing keys are rolled according to timing
parameters assigned to the keys. The keys must be generated and timing
parameters must be assigned by the zone operator.
The DNSSEC signing process maintains some metadata which is stored in the
:abbr:`KASP (Key And Signature Policy)` database. This database is backed
by LMDB.
Make sure to set the KASP database permissions correctly. For manual key
management, the database must be *readable* by the server process. For
automatic key management, it must be *writeable*. If no HSM is used,
the database also contains private key material – don't set the permissions
For automatic ZSK management a signing :ref:`policy<Policy section>` has to
be configured and assigned to the zone. The policy specifies how the zone
is signed (i.e. signing algorithm, key size, key lifetime, signature lifetime,
etc.). If no policy is specified or the ``default`` one is assigned, the
default signing parameters are used.
A minimal zone configuration may look as follows::
zone:
- domain: myzone.test
dnssec-signing: on
With a custom signing policy, the policy section will be added::
- id: custom_policy
signing-threads: 4
algorithm: ECDSAP256SHA256
zsk-lifetime: 60d
zone:
- domain: myzone.test
After configuring the server, reload the changes:
.. code-block:: console
The server will generate initial signing keys and sign the zone properly. Check
the server logs to see whether everything went well.
.. _dnssec-automatic-ksk-management:
Automatic KSK management
------------------------
For automatic KSK management, first configure ZSK management like above, and use
additional options in :ref:`policy section <Policy section>`, mostly specifying
desired (finite) lifetime for KSK: ::
remote:
- id: parent_zone_server
- id: parent_zone_sbm
parent: [parent_zone_server]
- id: custom_policy
signing-threads: 4
algorithm: ECDSAP256SHA256
zsk-lifetime: 60d
ksk-submission: parent_zone_sbm
zone:
- domain: myzone.test
dnssec-signing: on
After the initially-generated KSK reaches its lifetime, new KSK is published and after
convenience delay the submission is started. The server publishes CDS and CDNSKEY records
and the user shall propagate them to the parent. The server periodically checks for
DS at the parent zone and when positive, finishes the rollover.
.. _dnssec-manual-key-management:
Manual key management
---------------------
For automatic DNSSEC signing with manual key management, a signing policy
with manual key management flag has to be set::
zone:
- domain: myzone.test
dnssec-signing: on
To generate signing keys, use the :doc:`keymgr<man_keymgr>` utility.
For example, we can use Single-Type Signing:
.. code-block:: console
$ keymgr myzone.test. generate algorithm=ECDSAP256SHA256 ksk=yes zsk=yes
And reload the server. The zone will be signed.
To perform a manual rollover of a key, the timing parameters of the key need
to be set. Let's roll the key. Generate a new key, but do not activate
it yet:
.. code-block:: console
$ keymgr myzone.test. generate algorithm=ECDSAP256SHA256 ksk=yes zsk=yes active=+1d
Take the key ID (or key tag) of the old key and disable it the same time
the new key gets activated:
.. code-block:: console
$ keymgr myzone.test. set <old_key_id> retire=+2d remove=+3d
Reload the server again. The new key will be published (i.e. the DNSKEY record
will be added into the zone). Remember to update the DS record in the
parent zone to include a reference to the new key. This must happen within one
day (in this case) including a delay required to propagate the new DS to
caches.
.. WARNING::
If you ever decide to switch from manual key management to automatic key management,
note that the automatic key management uses
:ref:`policy_zsk-lifetime` and :ref:`policy_ksk-lifetime` policy configuration
options to schedule key rollovers and it internally uses timestamps of keys differently
than in the manual case. As a consequence it might break if the ``retire`` or ``remove`` timestamps
are set for the manually generated keys currently in use. Make sure to set these timestamps
to zero using :doc:`keymgr<man_keymgr>`:
.. code-block:: console
$ keymgr myzone.test. set <key_id> retire=0 remove=0
and configure your policy suitably according to :ref:`dnssec-automatic-zsk-management`
and :ref:`dnssec-automatic-ksk-management`.
Zone signing
------------
The signing process consists of the following steps:
#. Processing KASP database events. (e.g. performing a step of a rollover).
#. Updating the DNSKEY records. The whole DNSKEY set in zone apex is replaced
by the keys from the KASP database. Note that keys added into the zone file
manually will be removed. To add an extra DNSKEY record into the set, the
key must be imported into the KASP database (possibly deactivated).
#. Fixing the NSEC or NSEC3 chain.
#. Removing expired signatures, invalid signatures, signatures expiring
in a short time, and signatures issued by an unknown key.
#. Creating missing signatures. Unless the Single-Type Signing Scheme
is used, DNSKEY records in a zone apex are signed by KSK keys and
all other records are signed by ZSK keys.
The signing is initiated on the following occasions:
- Start of the server
- Zone reload
- Reaching the signature refresh period
- Key set changed due to rollover event
- Forced zone re-sign via server control interface
On a forced zone re-sign, all signatures in the zone are dropped and recreated.
The ``knotc zone-status`` command can be used to see when the next scheduled
On-secondary (on-slave) signing
-------------------------------
It is possible to enable automatic DNSSEC zone signing even on a secondary
server. If enabled, the zone is signed after every AXFR/IXFR transfer
from primary, so that the secondary always serves a signed up-to-date version
It is strongly recommended to block any outside access to the primary
server, so that only the secondary server's signed version of the zone is served.
Enabled on-secondary signing introduces events when the secondary zone changes
while the primary zone remains unchanged, such as a key rollover or
refreshing of RRSIG records, which cause inequality of zone SOA serial
between primary and secondary. The secondary server handles this by saving the
primary's SOA serial in a special variable inside KASP DB and appropriately
modifying AXFR/IXFR queries/answers to keep the communication with
primary server consistent while applying the changes with a different serial.
.. _catalog-zones:
Catalog zones
=============
Catalog zones are a concept whereby a list of zones to be configured is maintained
as contents of a separate, special zone. This approach has the benefit of simple
propagation of a zone list to secondary servers, especially when the list is
frequently updated. Currently, catalog zones are described in this `Internet Draft
David Vasek
committed
<https://tools.ietf.org/html/draft-ietf-dnsop-dns-catalog-zones>`_.
Terminology first. *Catalog zone* is a meta-zone which shall not be a part
of the DNS tree, but it contains information about the set of member zones and
is transferable to secondary servers using common AXFR/IXFR techniques.
A *catalog-member zone* (or just *member zone*) is a zone based on
information from the catalog zone and not from configuration file/database.
*Member properties* are some additional information related to each member zone,
A catalog zone is handled almost in the same way as a regular zone:
It can be configured using all the standard options (but for example
DNSSEC signing is useless as the zone won't be queried by clients), including primary/secondary configuration
and ACLs. A catalog zone is indicated by setting the option
Libor Peltan
committed
:ref:`zone_catalog-role`. The difference is that standard DNS
queries to a catalog zone are answered with REFUSED as though the zone
doesn't exist, unless querying over TCP from an address with transfers enabled
by ACL. The name of the catalog zone is arbitrary. It's possible to configure
multiple catalog zones.
Don't choose a name for a catalog zone below a name of any other
existing zones configured on the server as it would effectively "shadow"
part of your DNS subtree.
Upon catalog zone (re)load or change, all the PTR records in the format
``unique-id.zones.catalog. 0 IN PTR member.com.`` (but not ``too.deep.zones.catalog.``!)
are processed and member zones created, with zone names taken from the
PTR records' RData, and zone settings taken from the configuration
templates specified by :ref:`zone_catalog-template`.
The owner names of the PTR records shall follow this scheme:
.. code-block:: console
<unique-id>.zones.<catalog-zone>.
where the mentioned labels shall match:
- *<unique-id>* — Single label that is recommended to be unique among member zones.
- ``zones`` — Required label.
- *<catalog-zone>* — Name of the catalog zone.
Additionally, records in the format
``group.unique-id.zones.catalog. 0 IN TXT "conf-template"``
are processed as a definition of the member's *group* property. The
``unique-id`` must match the one of the PTR record defining the member.
All other records and other member properties are ignored. They remain in the catalog
zone, however, and might be for example transferred to a secondary server,
which may interpret catalog zones differently. SOA still needs to be present in
the catalog zone and its serial handled appropriately. An apex NS record must be
present as for any other zone. The version record ``version 0 IN TXT "2"``
is required at the catalog zone apex.
A catalog zone may be modified using any standard means (e.g. AXFR/IXFR, DDNS,
zone file reload). In the case of incremental change, only affected
member zones are reloaded.
The catalog zone must have at least one :ref:`zone_catalog-template`
configured. The configuration for any defined member zone is taken from its
*group* property value, which should match some catalog-template name.
If the *group* property is not defined for a member, is empty, or doesn't match
any of defined catalog-template names, the first catalog-template
(in the order from configuration) is used. Nesting of catalog zones isn't
supported.
Any de-cataloged member zone is purged immediately, including its
zone file, journal, timers, and DNSSEC keys. The zone file is not
deleted if :ref:`zone_zonefile-sync` is set to *-1* for member zones.
Any member zone, whose PTR record's owner has been changed, is purged
immediately if and only if the *<unique-id>* has been changed.
When setting up catalog zones, it might be useful to set
:ref:`database_catalog-db` and :ref:`database_catalog-db-max-size`
to non-default values.
.. NOTE::
Whenever a catalog zone is updated, the server reloads itself with
all configured zones, including possibly existing other catalog zones.
It's similar to calling `knotc zone-reload` (for all zones).
The consequence is that new zone files might be discovered and reloaded,
even for zones that do not relate to updated catalog zone.
Catalog zones never expire automatically, regardless of what is declared
in the catalog zone SOA. However, a catalog zone can be expired manually
at any time using `knotc -f zone-purge +expire`.
Currently, expiration of a catalog zone doesn't have any effect on its
member zones. This will likely change in the future depending on the
Internet Draft.
The server does not work well if one member zone appears in two catalog zones
concurrently. The user is encouraged to avoid this situation whatsoever.
Thus, there is no way a member zone can be migrated from one catalog
to another while preserving its metadata. Following steps may be used
as a workaround:
* :ref:`Back up<Data and metadata backup>` the member zone's metadata
(on each server separately).
* Remove the member zone from the catalog it's a member of.
* Wait for the catalog zone to be propagated to all servers.
* Add the member zone to the other catalog.
* Restore the backed up metadata (on each server separately).
Catalog zones configuration examples
------------------------------------
Below are configuration snippets (e.g. `server` and `log` sections missing)
of very simple catalog zone setups, in order to illustrate the relations
between catalog-related configuration options.
First setup represents a very simple scenario where the primary is
the catalog zone generator and the secondary is the catalog zone consumer.
Primary configuration::
acl:
- id: slave_xfr
address: ...
action: transfer
template:
- id: mmemb
catalog-role: member
catalog-zone: catz.
acl: slave_xfr
zone:
- domain: catz.
catalog-role: generate
acl: slave_xfr
- domain: foo.com.
template: mmemb
- domain: bar.com.
template: mmemb
Secondary configuration::
acl:
- id: master_notify
address: ...
action: notify
template:
- id: smemb
master: master
acl: master_notify
zone:
- domain: catz.
master: master
acl: master_notify
catalog-role: interpret
catalog-template: smemb
When new zones are added (or removed) to the primary configuration with assigned
`mmemb` template, they will automatically propagate to the secondary
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
and have the `smemb` template assigned there.
Second example is with a hand-written (or script-generated) catalog zone,
while employing configuration groups::
catz. 0 SOA invalid. invalid. 1625079950 3600 600 2147483646 0
catz. 0 NS invalid.
version.catz. 0 TXT "2"
nj2xg5bnmz2w4ltd.zones.catz. 0 PTR just-fun.com.
group.nj2xg5bnmz2w4ltd.zones.catz. 0 TXT unsigned
nvxxezjnmz2w4ltd.zones.catz. 0 PTR more-fun.com.
group.nvxxezjnmz2w4ltd.zones.catz. 0 TXT unsigned
nfwxa33sorqw45bo.zones.catz. 0 PTR important.com.
group.nfwxa33sorqw45bo.zones.catz. 0 TXT signed
mjqw42zomnxw2lq0.zones.catz. 0 PTR bank.com.
group.mjqw42zomnxw2lq0.zones.catz. 0 TXT signed
And the server in this case is configured to distinguish the groups by applying
different templates::
template:
- id: unsigned
...
- id: signed
dnssec-signing: on
dnssec-policy: ...
...
zone:
- domain: catz.
file: ...
catalog-role: interpret
catalog-template: [ unsigned, signed ]
.. _query-modules:
Query modules
=============
Knot DNS supports configurable query modules that can alter the way
queries are processed. Each query requires a finite number of steps to
be resolved. We call this set of steps a *query plan*, an abstraction
that groups these steps into several stages.
* Before-query processing
* Answer, Authority, Additional records packet sections processing
* After-query processing
For example, processing an Internet-class query needs to find an
answer. Then based on the previous state, it may also append an
authority SOA or provide additional records. Each of these actions
represents a 'processing step'. Now, if a query module is loaded for a
zone, it is provided with an implicit query plan which can be extended
by the module or even changed altogether.
A module is active if its name, which includes the ``mod-`` prefix, is assigned
to the zone/template :ref:`zone_module` option or to the ``default`` template
:ref:`template_global-module` option if activating for all queries.
If the module is configurable, a corresponding module section with
an identifier must be created and then referenced in the form of
``module_name/module_id``. See :ref:`Modules` for the list of available modules.
.. NOTE::
Query modules are processed in the order they are specified in the
zone/template configuration. In most cases, the recommended order is::
mod-synthrecord, mod-onlinesign, mod-cookies, mod-rrl, mod-dnstap, mod-stats
Performance Tuning
==================
Numbers of Workers
------------------
There are three types of workers ready for parallel execution of performance-oriented tasks:
UDP workers, TCP workers, and Background workers. The first two types handle all network requests
via the UDP and TCP protocol (respectively) and do the response jobs for common
queries. Background workers process changes to the zone.
By default, Knot determines a well-fitting number of workers based on the number of CPU cores.
The user can specify the number of workers for each type with configuration/server section:
:ref:`server_udp-workers`, :ref:`server_tcp-workers`, :ref:`server_background-workers`.
An indication of when to increase the number of workers is when the server is lagging behind
expected performance, while CPU usage remains low. This is usually due to waiting for network
or I/O response during the operation. It may be caused by Knot design not fitting the use-case well.
The user should try increasing the number of workers (of the related type) slightly above 100 and if
the performance improves, decide a further, exact setting.
Number of available file descriptors
------------------------------------
A name server configured for a large number of zones (hundreds or more) needs enough file descriptors
available for zone transfers and zone file updates, which default OS settings often don't provide.
It's necessary to check with the OS configuration and documentation and ensure the number of file
descriptors (sometimes called a number of concurrently open files) effective for the knotd process
is set suitably high. The number of concurrently open incoming TCP connections must be taken into
account too. In other words, the required setting is affected by the :ref:`server_tcp-max-clients`
setting.
Sysctl and NIC optimizations
----------------------------
There are several recommendations based on Knot developers' experience with their specific HW and SW
(mainstream Intel-based servers, Debian-based GNU/Linux distribution). They may improve or impact
performance in common use cases.
If your NIC driver allows it (see /proc/interrupts for hint), set CPU affinity (/proc/irq/$IRQ/smp_affinity)
manually so that each NIC channel is served by unique CPU core(s). You must turn off irqbalance service
in advance to avoid configuration override.
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
Configure sysctl as follows: ::
socket_bufsize=1048576
busy_latency=0
backlog=40000
optmem_max=20480
net.core.wmem_max = $socket_bufsize
net.core.wmem_default = $socket_bufsize
net.core.rmem_max = $socket_bufsize
net.core.rmem_default = $socket_bufsize
net.core.busy_read = $busy_latency
net.core.busy_poll = $busy_latency
net.core.netdev_max_backlog = $backlog
net.core.optmem_max = $optmem_max
Disable huge pages.
Configure your CPU to "performance" mode. This can be achieved depending on architecture, e.g. in BIOS,
or e.g. configuring /sys/devices/system/cpu/cpu*/cpufreq/scaling_governor to "performance".
Tune your NIC device with ethtool: ::
ethtool -A $dev autoneg off rx off tx off
ethtool -K $dev tso off gro off ufo off
ethtool -G $dev rx 4096 tx 4096
ethtool -C $dev rx-usecs 75
ethtool -C $dev tx-usecs 75
ethtool -N $dev rx-flow-hash udp4 sdfn
ethtool -N $dev rx-flow-hash udp6 sdfn
On FreeBSD you can just: ::
ifconfig ${dev} -rxcsum -txcsum -lro -tso
Knot developers are open to hear about users' further suggestions about network devices tuning/optimization.