This commit changes how Kilo allows nodes and peers behind NAT to roam.
Rather that ignore changes to endpoints when comparing WireGuard
configurations, Kilo now incorporates changes to endpoints for peers
behind NAT into its configuration first and later compares the
configurations.
Signed-off-by: Lucas Servén Marín <lserven@gmail.com>
This commit documents the use of the persistent-keepalive annotation and
corrects the implementation of keepalives.
Signed-off-by: Lucas Servén Marín <lserven@gmail.com>
Kilo had a routine that synchronized the endpoints of peers back into
the API to ensure that endpoints updated by WireGuard for a roaming peer
would always positively compare with the endpoints in the API. This is
no longer needed as Kilo will now simply ignore changes to endpoints for
peers with a non-zero persistent keepalive.
Signed-off-by: Lucas Servén Marín <lserven@gmail.com>
This commit enables Kilo to ignore changes to the endpoints of peers
that sit behind a NAT gateway. We use the heuristic of a non-zero
persistent keepalive to decide whether the endpoint field should be
ignored. This will allow NATed peers to roam and for every node in the
cluster to have a different value for a peer's endpoint, as is natural
when a peer's connections are NATed.
Signed-off-by: Lucas Servén Marín <lserven@gmail.com>
This commit fixes the issue encountered in #36, where the CNI config is
touched even though CNI management is disabled.
Fixes: #36
Signed-off-by: Lucas Servén Marín <lserven@gmail.com>
This commit allows DNS names to be used when specifying the endpoint
for a node in the WireGuard mesh. This is useful in many scenarios, in
particular when operating an IoT device whose public IP is dynamic. This
change allows the administrator to use a dynamic DNS name in the node's
endpoint.
One of the side-effects of this change is that the WireGuard port can
now be specified individually for each node in the mesh, if the
administrator wishes to do so.
*Note*: this commit introduces a breaking change; the
`force-external-ip` node annotation has been removed; its functionality
has been ported over to the `force-endpoint` annotation. This annotation
is documented in the annotations.md file. The expected content of this
annotation is no longer a CIDR but rather a host:port. The host can be
either a DNS name or an IP.
Signed-off-by: Lucas Servén Marín <lserven@gmail.com>
This commit entirely replaces NAT in Kilo with a few iproute2 rules.
Previously, Kilo would source-NAT the majority of packets in order to
avoid problems with strict source checks in cloud providers causing
packets to be considered martians. This source-NAT-ing made it
difficult to correctly apply Kuberenetes NetworkPolicies based on source
IPs.
This rewrite instead relies on a handful of iproute2 rules to ensure
that packets get encapsulated in certain scenarios based on the source
network and/or source interface.
This has the benefit of avoiding extra iptables bloat as well as
enabling better compatibility with NetworkPolicies.
Signed-off-by: Lucas Servén Marín <lserven@gmail.com>
This commit cleans up the iptables package to allow other packages to
create rules.
This commit also removes all NAT from Kilo.
Signed-off-by: Lucas Servén Marín <lserven@gmail.com>
This commit fixes the ip allocator `newAllocator` to produce IP
addresses with the original network mask. This is makes more sense. The
original functionality can be reproduced by wrapping the produced IP
address with the `oneAddressCIDR` helper.
Signed-off-by: Lucas Servén Marín <lserven@gmail.com>
This commit updates the well-known label to determine the region of the
node to topology.kubernetes.io/region, which is the new standard as
defined by the Kubernetes documentation, now that
failure-domain.beta.kubernetes.io/region has been deprecated.
This commit takes a big step towards ensuring that iptables rules are
always kept in the correct order. Specifically, when re-setting a a
ruleset, any time a rule is missing, that rule and all following rules
are re-added to ensure that from that index onwards all rules are in the
right order. Similarly, when reconciling an existing ruleset against the
backend, if a rule is missing, that rule an all following rules are
re-added.
This change does not guarantee that the order of rules in the backend
is correct. Unless an actor is modifying the order of rules in iptables,
all rules created by Kilo should now be kept in the correct order.
Fixes: #19
This commit makes it possible to specify the Kilo interface name. If the
specified interface exists, it will be used; if it does not exist, Kilo
will create it. If the interface already existed, then it will not be
deleted on shutdown; otherwise Kilo will destroy the interface.
Fixes: https://github.com/squat/kilo/issues/8
Addresses: 1/2 of https://github.com/squat/kilo/issues/17
If the hostname fails to resolve, this should not be considered a
blocking error. Most likely, it means that the hostname is simply not
resolvable, which should not be a requirement to run Kilo. In this case,
simply try to find a valid IP from other sources.
This commit adds basic support to run in compatibility mode with
Flannel. This allows clusters running Flannel as their principal
networking solution to leverage some advances Kilo features. In certain
Flannel setups, the clusters can even leverage muti-cloud. For this, the
cluster needs to either run in a full mesh, or Flannel needs to use the
API server's external IP address.
Add an exception to the route generation rules for when the external IP
of a node equals the internal IP. In this case, we cannot route traffic
through a tunnel.
This commit ensures that the WireGuard private key is re-used between
container restarts. The result of this is that external peers can keep
using their configuration and don't need to be re-configured just
because the Kilo container restarted.
We need to defensively deduplicate peer allowed IPs.
If two peers claim the same IP, the WireGuard configuration
could flap, causing the interface to churn.
This commit adds several output options to the `showconf` command of the
`kgctl` binary:
* `--as-peer`: this can be used to generate a peer configuration, which
can be used to configure the selected resource as a peer of another
WireGuard interface
* `--output`: this can be used to select the desired output format of
the peer resource, available options are: WireGuard, YAML, and JSON.
When interfaces on the host churn, the kernel will remove routes
associated with those interfaces. This could cause the Kilo route
controller to become out of sync with the routes that really exist. This
commit fixes this behavior.
This commit enables Kilo to work as an independent networking provider.
This is done by leveraging CNI. Kilo brings the necessary CNI plugins to
operate and takes care of all networking.
Add-on compatibility for Calico, Flannel, etc, will be re-introduced
shortly.
This commit exposes a new Prometheus to track the number of
reconciliation attempts. This is important, as without this, the number
of errors it not too helpful. A more valuable statistic is the
proportion of reconciliations that result in an error.
This commit introduces liveness checks to Kilo. This allows the Kilo
daemons to take nodes with inactive or dead Kilo deamons out of the
topology until they are alive again.
