Conntrack invalid packets may cause unexpected and subtle bugs
on esblished connections, because of that we install by default an
iptables rules that drops the packets with this conntrack state.
However, there are network scenarios, specially those that use multihoming
nodes, that may have legit traffic that is detected by conntrack as
invalid, hence these iptables rules are causing problems dropping this
traffic.
An alternative to solve the spurious problems caused by the invalid
connectrack packets is to set the sysctl nf_conntrack_tcp_be_liberal
option, but this is a system wide setting and we don't want kube-proxy
to be opinionated about the whole node networking configuration.
Kube-proxy will only install the DROP rules for invalid conntrack states
if the nf_conntrack_tcp_be_liberal is not set.
Change-Id: I5eb326931ed915f5ae74d210f0a375842b6a790e
Instead of using two metrics use just one metrics with multiple labels,
since the labels can only get 2 values, 200 or 503 there is no risk of
carindality explosion and are simple to represent in graphs.
Change-Id: I0e9cbd6ec2051de44d277d673dc20f02b96aa4d1
When I first wrote TestInternalExternalMasquerade, I put "FIXME"
comments in all of the cases that seemed wrong to me, most of which
got removed as we fixed the corner cases. But there were two cases
where we decided that the implemented behavior, though odd, was
correct, and those FIXMEs never got removed.
All the code to deal with enabling/disabling the feature gate is gone,
but some of the tests were still specifying "this test case assumes
PTE is enabled".
Remove "EndpointSlice" from some unit test names, because they don't
need to clarify that they use EndpointSlices now, because all of the
tests use EndpointSlices now.
Likewise, remove TestEndpointSliceE2E entirely; it was originally an
EndpointSlice version of one of the other tests, but the other test
uses EndpointSlices now too.
TL;DR: we want to start failing the LB HC if a node is tainted with ToBeDeletedByClusterAutoscaler.
This field might need refinement, but currently is deemed our best way of understanding if
a node is about to get deleted. We want to do this only for eTP:Cluster services.
The goal is to connection draining terminating nodes
Historically, IptablesRulesTotal could have been intepreted as either
"the total number of iptables rules kube-proxy is responsible for" or
"the number of iptables rules kube-proxy rewrote on the last sync".
Post-MinimizeIPTablesRestore, these are very different things (and
IptablesRulesTotal unintentionally became the latter).
Fix IptablesRulesTotal (sync_proxy_rules_iptables_total) to be "the
total number of iptables rules kube-proxy is responsible for" and add
IptablesRulesLastSync (sync_proxy_rules_iptables_last) to be "the
number of iptables rules kube-proxy rewrote on the last sync".
This required fixing a small bug in the metric, where it had
previously been counting the "-X" lines that had been passed to
iptables-restore to delete stale chains, rather than only counting the
actual rules.
getLocalDetector() used to pass a utiliptables.Interface to
NewDetectLocalByCIDR() so that NewDetectLocalByCIDR() could verify
that the passed-in CIDR was of the same family as the iptables
interface. It would make more sense for getLocalDetector() to verify
this itself and just *not call NewDetectLocalByCIDR* if the families
don't match, and that's what the code does now. So there's no longer
any need to pass the utiliptables.Interface to the local detector.
These don't belong in pkg/proxy/util; they involve a completely
unrelated definition of proxying.
Since each is only used from one place, just inline them at the
callers.
- add new header "X-Load-Balancing-Endpoint-Weight" returned from service health. Value of the header is number of local endpoints. Header can be used in weighted load balancing. Parsing header for number of endpoints is faster than unmarshalling json from the content body.
- add missing unit test for new and old headers returned from service health
Since kube-proxy in LocalModeNodeCIDR needs to obtain the PodCIDR
assigned to the node it watches for the Node object.
However, kube-proxy startup process requires to have these watches in
different places, that opens the possibility of having a race condition
if the same node is recreated and a different PodCIDR is assigned.
Initializing the second watch with the value obtained in the first one
allows us to detect this situation.
Change-Id: I6adeedb6914ad2afd3e0694dcab619c2a66135f8
Signed-off-by: Antonio Ojea <aojea@google.com>
