Cache in Knot Resolver is stored on disk and also shared between Multiple instances so resolver doesn’t lose the cached data on restart or crash.

To improve performance even further the resolver implements so-called aggressive caching for DNSSEC-validated data (RFC 8198), which improves performance and also protects against some types of Random Subdomain Attacks.


For personal and small office use-cases cache size around 100 MB is more than enough.

For large deployments we recommend to run Knot Resolver on a dedicated machine, and to allocate 90% of machine’s free memory for resolver’s cache.


Choosing a cache size that can fit into RAM is important even if the cache is stored on disk (default). Otherwise, the extra I/O caused by disk access for missing pages can cause performance issues.

For example, imagine you have a machine with 16 GB of memory. After machine restart you use command free -m to determine amount of free memory (without swap):

$ free -m
              total        used        free
Mem:          15907         979       14928

Now you can configure cache size to be 90% of the free memory 14 928 MB, i.e. 13 453 MB:

-- 90 % of free memory after machine restart
cache.size = 13453 * MB

It is also possible to set the cache size based on the file system size. This is useful if you use a dedicated partition for cache (e.g. non-persistent tmpfs). It is recommended to leave some free space for special files, such as locks.:

cache.size = cache.fssize() - 10*MB


The Garbage Collector can be used to periodically trim the cache. It is enabled and configured by default when running kresd with systemd integration.



Using tmpfs for cache improves performance and reduces disk I/O.

By default the cache is saved on a persistent storage device so the content of the cache is persisted during system reboot. This usually leads to smaller latency after restart etc., however in certain situations a non-persistent cache storage might be preferred, e.g.:

  • Resolver handles high volume of queries and I/O performance to disk is too low.

  • Threat model includes attacker getting access to disk content in power-off state.

  • Disk has limited number of writes (e.g. flash memory in routers).

If non-persistent cache is desired configure cache directory to be on tmpfs filesystem, a temporary in-memory file storage. The cache content will be saved in memory, and thus have faster access and will be lost on power-off or reboot.


In most of the Unix-like systems /tmp and /var/run are commonly mounted as tmpfs. While it is technically possible to move the cache to an existing tmpfs filesystem, it is not recommended, since the path to cache is configured in multiple places.

Mounting the cache directory as tmpfs is the recommended approach. Make sure to use appropriate size= option and don’t forget to adjust the size in the config file as well.

# /etc/fstab
tmpfs        /var/cache/knot-resolver        tmpfs   rw,size=2G,uid=knot-resolver,gid=knot-resolver,nosuid,nodev,noexec,mode=0700 0 0
-- /etc/knot-resolver/kresd.conf
cache.size = cache.fssize() - 10*MB

Configuration reference[, config_uri])

max_size (number) – Maximum cache size in bytes.


true if cache was opened

Open cache with a size limit. The cache will be reopened if already open. Note that the max_size cannot be lowered, only increased due to how cache is implemented.


Use kB, MB, GB constants as a multiplier, e.g. 100*MB.

The URI lmdb://path allows you to change the cache directory.

Example: * MB, 'lmdb:///var/cache/knot-resolver')

Set the cache maximum size in bytes. Note that this is only a hint to the backend, which may or may not respect it. See

cache.size = 100 * MB -- equivalent to ` * MB)`

Get the maximum size in bytes.


Set the cache storage backend configuration, see cache.backends() for more information. If the new storage configuration is invalid, it is not set. = 'lmdb://.'

Get the storage backend configuration.


map of backends


For now there is only one backend implementation, even though the APIs are ready for different (synchronous) backends.

The cache supports runtime-changeable backends, using the optional RFC 3986 URI, where the scheme represents backend protocol and the rest of the URI backend-specific configuration. By default, it is a lmdb backend in working directory, i.e. lmdb://.

Example output:

[lmdb://] => true

Number of entries in the cache. Meaning of the number is an implementation detail and is subject of change.


true if cache was closed

Close the cache.


This may or may not clear the cache, depending on the cache backend.


Partition size of cache storage.


Return table with low-level statistics for internal cache operation and storage. This counts each access to cache and does not directly map to individual DNS queries or resource records. For query-level statistics see stats module.


> cache.stats()
[clear] => 0
[close] => 0
[commit] => 117
[count] => 2
[count_entries] => 6187
[match] => 21
[match_miss] => 2
[open] => 0
[read] => 4313
[read_leq] => 9
[read_leq_miss] => 4
[read_miss] => 1143
[remove] => 17
[remove_miss] => 0
[usage_percent] => 15.625
[write] => 189

Cache operation read_leq (read less or equal, i.e. range search) was requested 9 times, and 4 out of 9 operations were finished with cache miss. Cache contains 6187 internal entries which occupy 15.625 % cache size.


ttl (number) – maximum TTL in seconds (default: 1 day)


current maximum TTL

Get or set upper TTL bound applied to all received records.


The ttl value must be in range (min_ttl, 2147483647).

-- Get maximum TTL
-- Set maximum TTL

ttl (number) – minimum TTL in seconds (default: 5 seconds)


current minimum TTL

Get or set lower TTL bound applied to all received records. Forcing TTL higher than specified violates DNS standards, so use higher values with care. TTL still won’t be extended beyond expiration of the corresponding DNSSEC signature.


The ttl value must be in range <0, max_ttl).

-- Get minimum TTL
-- Set minimum TTL

timeout (number) – NS retry interval in milliseconds (default: KR_NS_TIMEOUT_RETRY_INTERVAL)


current timeout

Get or set time interval for which a nameserver address will be ignored after determining that it doesn’t return (useful) answers. The intention is to avoid waiting if there’s little hope; instead, kresd can immediately SERVFAIL or immediately use stale records (with serve_stale module).


This settings applies only to the current kresd process.


This function is not implemented at this moment. We plan to re-introduce it soon, probably with a slightly different API.

cache.clear([name][, exact_name][, rr_type][, chunk_size][, callback][, prev_state])

Purge cache records matching specified criteria. There are two specifics:

  • To reliably remove negative cache entries you need to clear subtree with the whole zone. E.g. to clear negative cache entries for (formerly non-existing) record A you need to flush whole subtree starting at zone apex, e.g. [1].

  • This operation is asynchronous and might not be yet finished when call to cache.clear() function returns. Return value indicates if clearing continues asynchronously or not.

  • name (string) – subtree to purge; if the name isn’t provided, whole cache is purged (and any other parameters are disregarded).

  • exact_name (bool) – if set to true, only records with the same name are removed; default: false.

  • rr_type (kres.type) – you may additionally specify the type to remove, but that is only supported with exact_name == true; default: nil.

  • chunk_size (integer) – the number of records to remove in one round; default: 100. The purpose is not to block the resolver for long. The default callback repeats the command after one millisecond until all matching data are cleared.

  • callback (function) – a custom code to handle result of the underlying C call. Its parameters are copies of those passed to cache.clear() with one additional parameter rettable containing table with return value from current call. count field contains a return code from kr_cache_remove_subtree().

  • prev_state (table) – return value from previous run (can be used by callback)

Return type:



count key is always present. Other keys are optional and their presence indicate special conditions.

  • count (integer) - number of items removed from cache by this call (can be 0 if no entry matched criteria)

  • not_apex - cleared subtree is not cached as zone apex; proofs of non-existence were probably not removed

  • subtree (string) - hint where zone apex lies (this is estimation from cache content and might not be accurate)

  • chunk_limit - more than chunk_size items needs to be cleared, clearing will continue asynchronously


-- Clear whole cache
> cache.clear()
[count] => 76

-- Clear records at and below 'com.'
> cache.clear('com.')
[chunk_limit] => chunk size limit reached; the default callback will continue asynchronously
[not_apex] => to clear proofs of non-existence call cache.clear('com.')
[count] => 100
[round] => 1
[subtree] => com.
> worker.sleep(0.1)
[cache] asynchronous cache.clear('com', false) finished

-- Clear only ''
> cache.clear('', true)
[round] => 1
[count] => 1
[not_apex] => to clear proofs of non-existence call cache.clear('')
[subtree] =>