Kube-proxy: ICMP reject via LBs when no endpoints #74394
Merged
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
k8s-ci-robot
added
release-note
|
[APPROVALNOTIFIER] This PR is APPROVED This pull-request has been approved by: thockin The full list of commands accepted by this bot can be found here. The pull request process is described here
Needs approval from an approver in each of these files:
Approvers can indicate their approval by writing |
k8s-ci-robot
added
approved
k8s-ci-robot
removed
the
needs-sig
thockin
force-pushed
the
proxy-reject-lb-no-endpoints
branch
2 times, most recently
from
02a0bd9 to
ef913cf
Compare
thockin
added
the
priority/important-soon
k8s-ci-robot
removed
the
needs-priority
danwinship
reviewed
Feb 22, 2019
test/e2e/network/service.go
Outdated
| jig.Scale(ns1, 0) | ||
| jig.Scale(ns2, 0) | ||
|
|
||
| //FIXME: prove ICMP-reject |
There was a problem hiding this comment.
For the internal case (#72561) we ended up just assuming that we'd get "connection refused" rather than "timed out".
You could just check the counters on the iptables rules; run iptables-save -c before and after, and verify that the packet count on the expected REJECT rule actually went up, and assume that if the right rule got hit, then the right end-user behavior happened too. (Though in fact, in this case the test would still pass even if the network plugin ate the ICMP reply...)
You could also check the global ICMP counters in netstat -s, which would let you confirm that the node had received an ICMP reject for some reason. That would have some unknowable number of false positives, but shouldn't have false negatives.
k8s-ci-robot
added
size/L
|
As this PR is virtually identical to #72879 I would appreciate if you name me as co-author and close the other one.
Testing is indeed tricky but here is how I did it manually (with some proposal to automate). Our problem was not the "connection refused" but the stale conntrack entries that were caused by not rejecting the packets. Downside is that it works only for providers that have load balancers that do not terminate the TCP connections. Prepare serviceLog on to the host or a pod with host network (in this case calico-node)Make sure there are not -> no stale SYN_SENT entries, all clean Start curl endless loopScale replicas to 0Go to host and look for stale entries:With the fix we don't see any stale This can be simplified in the following way, however that works only if the load balancer is not a terminating one:
This should be straightforward but will not work when NodePort is created for the LB (but is straightforward to automate). |
|
/milestone v1.14 |
thockin
force-pushed
the
proxy-reject-lb-no-endpoints
branch
from
26d3382 to
67cb489
Compare
|
Happy to name you as co-author (inasmuch as that's a thing). We need to have automated testing for this. I have been jumping between tasks, and have not had much time on this one. I think I finally figured why the approach I had was not working. If this run fails the way I expect, then we can proceed. |
thockin
force-pushed
the
proxy-reject-lb-no-endpoints
branch
2 times, most recently
from
e6eca1f to
2bf827b
Compare
rjaini
added a commit
to msazurestackworkloads/kubernetes
that referenced
this pull request
May 6, 2019
* Fix kubernetes#73479 AWS NLB target groups missing tags `elbv2.AddTags` doesn't seem to support assigning the same set of tags to multiple resources at once leading to the following error: Error adding tags after modifying load balancer targets: "ValidationError: Only one resource can be tagged at a time" This can happen when using AWS NLB with multiple listeners pointing to different node ports. When k8s creates a NLB it creates a target group per listener along with installing security group ingress rules allowing the traffic to reach the k8s nodes. Unfortunately if those target groups are not tagged, k8s will not manage them, thinking it is not the owner. This small changes assigns tags one resource at a time instead of batching them as before. Signed-off-by: Brice Figureau <brice@daysofwonder.com> * remove get azure accounts in the init process set timeout for get azure account operation use const for timeout value remove get azure accounts in the init process add lock for account init * add timeout in GetVolumeLimits operation add timeout for getAllStorageAccounts * add mixed protocol support for azure load balancer * record event on endpoint update failure * fix parse devicePath issue on Azure Disk * Kubernetes version v1.12.7-beta.0 openapi-spec file updates * add retry for detach azure disk add more logging info in detach disk add more logging for azure disk attach/detach * Add/Update CHANGELOG-1.12.md for v1.12.6. * Reduce cardinality of admission webhook metrics * fix negative slice index error in keymutex * Remove reflector metrics as they currently cause a memory leak * Explicitly set GVK when sending objects to webhooks * add Azure Container Registry anonymous repo support apply fix for msi and fix test failure * DaemonSet e2e: Update image and rolling upgrade test timeout Use Nginx as the DaemonSet image instead of the ServeHostname image. This was changed because the ServeHostname has a sleep after terminating which makes it incompatible with the DaemonSet Rolling Upgrade e2e test. In addition, make the DaemonSet Rolling Upgrade e2e test timeout a function of the number of nodes that make up the cluster. This is required because the more nodes there are, the longer the time it will take to complete a rolling upgrade. Signed-off-by: Alexander Brand <alexbrand09@gmail.com> * Revert kubelet to default to ttl cache secret/configmap behavior * cri_stats_provider: overload nil as 0 for exited containers stats Always report 0 cpu/memory usage for exited containers to make metrics-server work as expect. Signed-off-by: Lu Fengqi <lufq.fnst@cn.fujitsu.com> * flush iptable chains first and then remove them while cleaning up ipvs mode. flushing iptable chains first and then remove the chains. this avoids trying to remove chains that are still referenced by rules in other chains. fixes kubernetes#70615 * Checks whether we have cached runtime state before starting a container that requests any device plugin resource. If not, re-issue Allocate grpc calls. This allows us to handle the edge case that a pod got assigned to a node even before it populates its extended resource capacity. * Fix panic in kubectl cp command * Augmenting API call retry in nodeinfomanager * Bump debian-iptables to v11.0.1. Rebase docker image on debian-base:0.4.1 * Adding a check to make sure UseInstanceMetadata flag is true to get data from metadata. * GetMountRefs fixed to handle corrupted mounts by treating it like an unmounted volume * Update Cluster Autoscaler version to 1.12.3 * add module 'nf_conntrack' in ipvs prerequisite check * Allow disable outbound snat when Azure standard load balancer is used * Ensure Azure load balancer cleaned up on 404 or 403 * fix smb unmount issue on Windows fix log warning use IsCorruptedMnt in GetMountRefs on Windows use errorno in IsCorruptedMnt check fix comments: add more error code add more error no checking change year fix comments fix bazel error fix bazel fix bazel fix bazel revert bazel change * kubelet: updated logic of verifying a static critical pod - check if a pod is static by its static pod info - meanwhile, check if a pod is critical by its corresponding mirror pod info * Allow session affinity a period of time to setup for new services. This is to deal with the flaky session affinity test. * Restore username and password kubectl flags * build/gci: bump CNI version to 0.7.5 * fix race condition issue for smb mount on windows change var name * allows configuring NPD release and flags on GCI and add cluster e2e test * allows configuring NPD image version in node e2e test and fix the test * bump repd min size in e2es * Kubernetes version v1.12.8-beta.0 openapi-spec file updates * Add/Update CHANGELOG-1.12.md for v1.12.7. * stop vsphere cloud provider from spamming logs with `failed to patch IP` Fixes: kubernetes#75236 * Do not delete existing VS and RS when starting * Fix updating 'currentMetrics' field for HPA with 'AverageValue' target * Populate ClientCA in delegating auth setup kubernetes#67768 accidentally removed population of the the ClientCA in the delegating auth setup code. This restores it. * Update gcp images with security patches [stackdriver addon] Bump prometheus-to-sd to v0.5.0 to pick up security fixes. [fluentd-gcp addon] Bump fluentd-gcp-scaler to v0.5.1 to pick up security fixes. [fluentd-gcp addon] Bump event-exporter to v0.2.4 to pick up security fixes. [fluentd-gcp addon] Bump prometheus-to-sd to v0.5.0 to pick up security fixes. [metatada-proxy addon] Bump prometheus-to-sd v0.5.0 to pick up security fixes. * Fix AWS driver fails to provision specified fsType * Updated regional PD minimum size; changed regional PD failover test to use StorageClassTest to generate PVC template * Bump debian-iptables to v11.0.2 * Avoid panic in cronjob sorting This change handles the case where the ith cronjob may have its start time set to nil. Previously, the Less method could cause a panic in case the ith cronjob had its start time set to nil, but the jth cronjob did not. It would panic when calling Before on a nil StartTime. * Add volume mode downgrade test: should not mount/map in <1.13 * disable HTTP2 ingress test * ensuring that logic is checking for differences in listener * Use Node-Problem-Detector v0.6.3 on GCI * Delete only unscheduled pods if node doesn't exist anymore. * proxy: Take into account exclude CIDRs while deleting legacy real servers * Increase default maximumLoadBalancerRuleCount to 250 * kube-proxy: rename internal field for clarity * kube-proxy: rename vars for clarity, fix err str * kube-proxy: rename field for congruence * kube-proxy: reject 0 endpoints on forward Previously we only REJECTed on OUTPUT which works for packets from the node but not for packets from pods on the node. * kube-proxy: remove old cleanup rules * Kube-proxy: REJECT LB IPs with no endpoints We REJECT every other case. Close this FIXME. To get this to work in all cases, we have to process service in filter.INPUT, since LB IPS might be manged as local addresses. * Retool HTTP and UDP e2e utils This is a prefactoring for followup changes that need to use very similar but subtly different test. Now it is more generic, though it pushes a little logic up the stack. That makes sense to me. * Fix small race in e2e Occasionally we get spurious errors about "no route to host" when we race with kube-proxy. This should reduce that. It's mostly just log noise. * Fix Azure SLB support for multiple backend pools Azure VM and vmssVM support multiple backend pools for the same SLB, but not for different LBs. * Revert "Merge pull request kubernetes#76529 from spencerhance/automated-cherry-pick-of-#72534-kubernetes#74394-upstream-release-1.12" This reverts commit 535e3ad, reversing changes made to 336d787.
rjaini
added a commit
to msazurestackworkloads/kubernetes
that referenced
this pull request
May 31, 2019
* Fix kubernetes#73479 AWS NLB target groups missing tags `elbv2.AddTags` doesn't seem to support assigning the same set of tags to multiple resources at once leading to the following error: Error adding tags after modifying load balancer targets: "ValidationError: Only one resource can be tagged at a time" This can happen when using AWS NLB with multiple listeners pointing to different node ports. When k8s creates a NLB it creates a target group per listener along with installing security group ingress rules allowing the traffic to reach the k8s nodes. Unfortunately if those target groups are not tagged, k8s will not manage them, thinking it is not the owner. This small changes assigns tags one resource at a time instead of batching them as before. Signed-off-by: Brice Figureau <brice@daysofwonder.com> * remove get azure accounts in the init process set timeout for get azure account operation use const for timeout value remove get azure accounts in the init process add lock for account init * add timeout in GetVolumeLimits operation add timeout for getAllStorageAccounts * add mixed protocol support for azure load balancer * record event on endpoint update failure * fix parse devicePath issue on Azure Disk * Fix scanning of failed targets If a iSCSI target is down while a volume is attached, reading from /sys/class/iscsi_host/host415/device/session383/connection383:0/iscsi_connection/connection383:0/address fails with an error. Kubelet should assume that such target is not available / logged in and try to relogin. Eventually, if such error persists, it should continue mounting the volume if the other paths are healthy instead of failing whole WaitForAttach(). * Kubernetes version v1.12.7-beta.0 openapi-spec file updates * add retry for detach azure disk add more logging info in detach disk add more logging for azure disk attach/detach * Add/Update CHANGELOG-1.12.md for v1.12.6. * Reduce cardinality of admission webhook metrics * fix negative slice index error in keymutex * Remove reflector metrics as they currently cause a memory leak * Explicitly set GVK when sending objects to webhooks * add Azure Container Registry anonymous repo support apply fix for msi and fix test failure * DaemonSet e2e: Update image and rolling upgrade test timeout Use Nginx as the DaemonSet image instead of the ServeHostname image. This was changed because the ServeHostname has a sleep after terminating which makes it incompatible with the DaemonSet Rolling Upgrade e2e test. In addition, make the DaemonSet Rolling Upgrade e2e test timeout a function of the number of nodes that make up the cluster. This is required because the more nodes there are, the longer the time it will take to complete a rolling upgrade. Signed-off-by: Alexander Brand <alexbrand09@gmail.com> * Revert kubelet to default to ttl cache secret/configmap behavior * cri_stats_provider: overload nil as 0 for exited containers stats Always report 0 cpu/memory usage for exited containers to make metrics-server work as expect. Signed-off-by: Lu Fengqi <lufq.fnst@cn.fujitsu.com> * flush iptable chains first and then remove them while cleaning up ipvs mode. flushing iptable chains first and then remove the chains. this avoids trying to remove chains that are still referenced by rules in other chains. fixes kubernetes#70615 * Checks whether we have cached runtime state before starting a container that requests any device plugin resource. If not, re-issue Allocate grpc calls. This allows us to handle the edge case that a pod got assigned to a node even before it populates its extended resource capacity. * Fix panic in kubectl cp command * Augmenting API call retry in nodeinfomanager * Bump debian-iptables to v11.0.1. Rebase docker image on debian-base:0.4.1 * Adding a check to make sure UseInstanceMetadata flag is true to get data from metadata. * GetMountRefs fixed to handle corrupted mounts by treating it like an unmounted volume * Update Cluster Autoscaler version to 1.12.3 * add module 'nf_conntrack' in ipvs prerequisite check * Allow disable outbound snat when Azure standard load balancer is used * Ensure Azure load balancer cleaned up on 404 or 403 * fix smb unmount issue on Windows fix log warning use IsCorruptedMnt in GetMountRefs on Windows use errorno in IsCorruptedMnt check fix comments: add more error code add more error no checking change year fix comments fix bazel error fix bazel fix bazel fix bazel revert bazel change * kubelet: updated logic of verifying a static critical pod - check if a pod is static by its static pod info - meanwhile, check if a pod is critical by its corresponding mirror pod info * Allow session affinity a period of time to setup for new services. This is to deal with the flaky session affinity test. * Restore username and password kubectl flags * build/gci: bump CNI version to 0.7.5 * fix race condition issue for smb mount on windows change var name * allows configuring NPD release and flags on GCI and add cluster e2e test * allows configuring NPD image version in node e2e test and fix the test * bump repd min size in e2es * Kubernetes version v1.12.8-beta.0 openapi-spec file updates * Add/Update CHANGELOG-1.12.md for v1.12.7. * stop vsphere cloud provider from spamming logs with `failed to patch IP` Fixes: kubernetes#75236 * Do not delete existing VS and RS when starting * Fix updating 'currentMetrics' field for HPA with 'AverageValue' target * Populate ClientCA in delegating auth setup kubernetes#67768 accidentally removed population of the the ClientCA in the delegating auth setup code. This restores it. * Update gcp images with security patches [stackdriver addon] Bump prometheus-to-sd to v0.5.0 to pick up security fixes. [fluentd-gcp addon] Bump fluentd-gcp-scaler to v0.5.1 to pick up security fixes. [fluentd-gcp addon] Bump event-exporter to v0.2.4 to pick up security fixes. [fluentd-gcp addon] Bump prometheus-to-sd to v0.5.0 to pick up security fixes. [metatada-proxy addon] Bump prometheus-to-sd v0.5.0 to pick up security fixes. * Fix AWS driver fails to provision specified fsType * Updated regional PD minimum size; changed regional PD failover test to use StorageClassTest to generate PVC template * Bump debian-iptables to v11.0.2 * Avoid panic in cronjob sorting This change handles the case where the ith cronjob may have its start time set to nil. Previously, the Less method could cause a panic in case the ith cronjob had its start time set to nil, but the jth cronjob did not. It would panic when calling Before on a nil StartTime. * Add volume mode downgrade test: should not mount/map in <1.13 * disable HTTP2 ingress test * ensuring that logic is checking for differences in listener * Use Node-Problem-Detector v0.6.3 on GCI * Delete only unscheduled pods if node doesn't exist anymore. * proxy: Take into account exclude CIDRs while deleting legacy real servers * Increase default maximumLoadBalancerRuleCount to 250 * kube-proxy: rename internal field for clarity * kube-proxy: rename vars for clarity, fix err str * kube-proxy: rename field for congruence * kube-proxy: reject 0 endpoints on forward Previously we only REJECTed on OUTPUT which works for packets from the node but not for packets from pods on the node. * kube-proxy: remove old cleanup rules * Kube-proxy: REJECT LB IPs with no endpoints We REJECT every other case. Close this FIXME. To get this to work in all cases, we have to process service in filter.INPUT, since LB IPS might be manged as local addresses. * Retool HTTP and UDP e2e utils This is a prefactoring for followup changes that need to use very similar but subtly different test. Now it is more generic, though it pushes a little logic up the stack. That makes sense to me. * Fix small race in e2e Occasionally we get spurious errors about "no route to host" when we race with kube-proxy. This should reduce that. It's mostly just log noise. * Fix Azure SLB support for multiple backend pools Azure VM and vmssVM support multiple backend pools for the same SLB, but not for different LBs. * Set CPU metrics for init containers under containerd Copies PR kubernetes#76503 for release-1.12. metrics-server doesn't return metrics for pods with init containers under containerd because they have incomplete CPU metrics returned by the kubelet /stats/summary API. This problem has been fixed in 1.14 (kubernetes#74336), but the cherry-picks dropped the usageNanoCores metric. This change adds the missing usageNanoCores metric for init containers in Kubernetes v1.12. Fixes kubernetes#76292 * Restore metrics-server using of IP addresses This preference list matches is used to pick prefered field from k8s node object. It was introduced in metrics-server 0.3 and changed default behaviour to use DNS instead of IP addresses. It was merged into k8s 1.12 and caused breaking change by introducing dependency on DNS configuration. * Revert "Merge pull request kubernetes#76529 from spencerhance/automated-cherry-pick-of-#72534-kubernetes#74394-upstream-release-1.12" This reverts commit 535e3ad, reversing changes made to 336d787. * Kubernetes version v1.12.9-beta.0 openapi-spec file updates * Add/Update CHANGELOG-1.12.md for v1.12.8. * Upgrade compute API to version 2019-03-01 * Replace vmss update API with instance-level update API * Cleanup codes that not required any more * Add unit tests * Update vendors * Update Cluster Autoscaler to 1.12.5 * add shareName param in azure file storage class skip create azure file if it exists remove comments * Create the "internal" firewall rule for kubemark master. This is equivalent to the "internal" firewall rule that is created for the regular masters. The main reason for doing it is to allow prometheus scraping metrics from various kubemark master components, e.g. kubelet. Ref. kubernetes/perf-tests#503 * refactor detach azure disk retry operation * move disk lock process to azure cloud provider fix comments fix import keymux check error add unit test for attach/detach disk funcs fix bazel issue rebase * fix disk list corruption issue * Fix verify godeps failure for 1.12 github.com/evanphx/json-patch added a new tag at the same sha this morning: https://github.com/evanphx/json-patch/releases/tag/v4.2.0 This confused godeps. This PR updates our file to match godeps expectation. Fixes issue 77238 * Upgrade Stackdriver Logging Agent addon image from 1.6.0 to 1.6.8. * Test kubectl cp escape * Properly handle links in tar * use k8s.gcr.io/pause instead of kubernetes/pause * Pick up security patches for fluentd-gcp-scaler by upgrading to version 0.5.2 * Error when etcd3 watch finds delete event with nil prevKV * Make CreatePrivilegedPSPBinding reentrant Make CreatePrivilegedPSPBinding reentrant so tests using it (e.g. DNS) can be executed more than once against a cluster. Without this change, such tests will fail because the PSP already exists, short circuiting test setup. * check if Memory is not nil for container stats * In GuaranteedUpdate, retry on any error if we are working with stale data * BoundServiceAccountTokenVolume: fix InClusterConfig * fix CVE-2019-11244: `kubectl --http-cache=<world-accessible dir>` creates world-writeable cached schema files * Terminate watchers when watch cache is destroyed * honor overridden tokenfile, add InClusterConfig override tests * fix incorrect prometheus metrics
rjaini
added a commit
to msazurestackworkloads/kubernetes
that referenced
this pull request
Jul 11, 2019
* Fix bug with volume getting marked as not in-use with pending op Add test for verifying volume detach * Fix flake with e2e test that checks detach while mount in progress A volume can show up as in-use even before it gets attached to the node. * Fix kubernetes#73479 AWS NLB target groups missing tags `elbv2.AddTags` doesn't seem to support assigning the same set of tags to multiple resources at once leading to the following error: Error adding tags after modifying load balancer targets: "ValidationError: Only one resource can be tagged at a time" This can happen when using AWS NLB with multiple listeners pointing to different node ports. When k8s creates a NLB it creates a target group per listener along with installing security group ingress rules allowing the traffic to reach the k8s nodes. Unfortunately if those target groups are not tagged, k8s will not manage them, thinking it is not the owner. This small changes assigns tags one resource at a time instead of batching them as before. Signed-off-by: Brice Figureau <brice@daysofwonder.com> * remove get azure accounts in the init process set timeout for get azure account operation use const for timeout value remove get azure accounts in the init process add lock for account init * add timeout in GetVolumeLimits operation add timeout for getAllStorageAccounts * add mixed protocol support for azure load balancer * record event on endpoint update failure * fix parse devicePath issue on Azure Disk * Fix scanning of failed targets If a iSCSI target is down while a volume is attached, reading from /sys/class/iscsi_host/host415/device/session383/connection383:0/iscsi_connection/connection383:0/address fails with an error. Kubelet should assume that such target is not available / logged in and try to relogin. Eventually, if such error persists, it should continue mounting the volume if the other paths are healthy instead of failing whole WaitForAttach(). * Kubernetes version v1.12.7-beta.0 openapi-spec file updates * add retry for detach azure disk add more logging info in detach disk add more logging for azure disk attach/detach * Add/Update CHANGELOG-1.12.md for v1.12.6. * Reduce cardinality of admission webhook metrics * fix negative slice index error in keymutex * Remove reflector metrics as they currently cause a memory leak * Explicitly set GVK when sending objects to webhooks * add Azure Container Registry anonymous repo support apply fix for msi and fix test failure * DaemonSet e2e: Update image and rolling upgrade test timeout Use Nginx as the DaemonSet image instead of the ServeHostname image. This was changed because the ServeHostname has a sleep after terminating which makes it incompatible with the DaemonSet Rolling Upgrade e2e test. In addition, make the DaemonSet Rolling Upgrade e2e test timeout a function of the number of nodes that make up the cluster. This is required because the more nodes there are, the longer the time it will take to complete a rolling upgrade. Signed-off-by: Alexander Brand <alexbrand09@gmail.com> * Revert kubelet to default to ttl cache secret/configmap behavior * cri_stats_provider: overload nil as 0 for exited containers stats Always report 0 cpu/memory usage for exited containers to make metrics-server work as expect. Signed-off-by: Lu Fengqi <lufq.fnst@cn.fujitsu.com> * flush iptable chains first and then remove them while cleaning up ipvs mode. flushing iptable chains first and then remove the chains. this avoids trying to remove chains that are still referenced by rules in other chains. fixes kubernetes#70615 * Checks whether we have cached runtime state before starting a container that requests any device plugin resource. If not, re-issue Allocate grpc calls. This allows us to handle the edge case that a pod got assigned to a node even before it populates its extended resource capacity. * Fix panic in kubectl cp command * Augmenting API call retry in nodeinfomanager * Bump debian-iptables to v11.0.1. Rebase docker image on debian-base:0.4.1 * Adding a check to make sure UseInstanceMetadata flag is true to get data from metadata. * GetMountRefs fixed to handle corrupted mounts by treating it like an unmounted volume * Update Cluster Autoscaler version to 1.12.3 * add module 'nf_conntrack' in ipvs prerequisite check * Allow disable outbound snat when Azure standard load balancer is used * Ensure Azure load balancer cleaned up on 404 or 403 * fix smb unmount issue on Windows fix log warning use IsCorruptedMnt in GetMountRefs on Windows use errorno in IsCorruptedMnt check fix comments: add more error code add more error no checking change year fix comments fix bazel error fix bazel fix bazel fix bazel revert bazel change * kubelet: updated logic of verifying a static critical pod - check if a pod is static by its static pod info - meanwhile, check if a pod is critical by its corresponding mirror pod info * Allow session affinity a period of time to setup for new services. This is to deal with the flaky session affinity test. * Restore username and password kubectl flags * build/gci: bump CNI version to 0.7.5 * fix race condition issue for smb mount on windows change var name * allows configuring NPD release and flags on GCI and add cluster e2e test * allows configuring NPD image version in node e2e test and fix the test * bump repd min size in e2es * Kubernetes version v1.12.8-beta.0 openapi-spec file updates * Add/Update CHANGELOG-1.12.md for v1.12.7. * stop vsphere cloud provider from spamming logs with `failed to patch IP` Fixes: kubernetes#75236 * Do not delete existing VS and RS when starting * Fix updating 'currentMetrics' field for HPA with 'AverageValue' target * Populate ClientCA in delegating auth setup kubernetes#67768 accidentally removed population of the the ClientCA in the delegating auth setup code. This restores it. * Update gcp images with security patches [stackdriver addon] Bump prometheus-to-sd to v0.5.0 to pick up security fixes. [fluentd-gcp addon] Bump fluentd-gcp-scaler to v0.5.1 to pick up security fixes. [fluentd-gcp addon] Bump event-exporter to v0.2.4 to pick up security fixes. [fluentd-gcp addon] Bump prometheus-to-sd to v0.5.0 to pick up security fixes. [metatada-proxy addon] Bump prometheus-to-sd v0.5.0 to pick up security fixes. * Fix AWS driver fails to provision specified fsType * Updated regional PD minimum size; changed regional PD failover test to use StorageClassTest to generate PVC template * Bump debian-iptables to v11.0.2 * Avoid panic in cronjob sorting This change handles the case where the ith cronjob may have its start time set to nil. Previously, the Less method could cause a panic in case the ith cronjob had its start time set to nil, but the jth cronjob did not. It would panic when calling Before on a nil StartTime. * Add volume mode downgrade test: should not mount/map in <1.13 * disable HTTP2 ingress test * ensuring that logic is checking for differences in listener * Use Node-Problem-Detector v0.6.3 on GCI * Delete only unscheduled pods if node doesn't exist anymore. * proxy: Take into account exclude CIDRs while deleting legacy real servers * Increase default maximumLoadBalancerRuleCount to 250 * kube-proxy: rename internal field for clarity * kube-proxy: rename vars for clarity, fix err str * kube-proxy: rename field for congruence * kube-proxy: reject 0 endpoints on forward Previously we only REJECTed on OUTPUT which works for packets from the node but not for packets from pods on the node. * kube-proxy: remove old cleanup rules * Kube-proxy: REJECT LB IPs with no endpoints We REJECT every other case. Close this FIXME. To get this to work in all cases, we have to process service in filter.INPUT, since LB IPS might be manged as local addresses. * Retool HTTP and UDP e2e utils This is a prefactoring for followup changes that need to use very similar but subtly different test. Now it is more generic, though it pushes a little logic up the stack. That makes sense to me. * Fix small race in e2e Occasionally we get spurious errors about "no route to host" when we race with kube-proxy. This should reduce that. It's mostly just log noise. * Fix Azure SLB support for multiple backend pools Azure VM and vmssVM support multiple backend pools for the same SLB, but not for different LBs. * Set CPU metrics for init containers under containerd Copies PR kubernetes#76503 for release-1.12. metrics-server doesn't return metrics for pods with init containers under containerd because they have incomplete CPU metrics returned by the kubelet /stats/summary API. This problem has been fixed in 1.14 (kubernetes#74336), but the cherry-picks dropped the usageNanoCores metric. This change adds the missing usageNanoCores metric for init containers in Kubernetes v1.12. Fixes kubernetes#76292 * Restore metrics-server using of IP addresses This preference list matches is used to pick prefered field from k8s node object. It was introduced in metrics-server 0.3 and changed default behaviour to use DNS instead of IP addresses. It was merged into k8s 1.12 and caused breaking change by introducing dependency on DNS configuration. * Revert "Merge pull request kubernetes#76529 from spencerhance/automated-cherry-pick-of-#72534-kubernetes#74394-upstream-release-1.12" This reverts commit 535e3ad, reversing changes made to 336d787. * Kubernetes version v1.12.9-beta.0 openapi-spec file updates * Add/Update CHANGELOG-1.12.md for v1.12.8. * Upgrade compute API to version 2019-03-01 * Replace vmss update API with instance-level update API * Cleanup codes that not required any more * Add unit tests * Update vendors * Update Cluster Autoscaler to 1.12.5 * add shareName param in azure file storage class skip create azure file if it exists remove comments * Create the "internal" firewall rule for kubemark master. This is equivalent to the "internal" firewall rule that is created for the regular masters. The main reason for doing it is to allow prometheus scraping metrics from various kubemark master components, e.g. kubelet. Ref. kubernetes/perf-tests#503 * refactor detach azure disk retry operation * move disk lock process to azure cloud provider fix comments fix import keymux check error add unit test for attach/detach disk funcs fix bazel issue rebase * fix disk list corruption issue * Fix verify godeps failure for 1.12 github.com/evanphx/json-patch added a new tag at the same sha this morning: https://github.com/evanphx/json-patch/releases/tag/v4.2.0 This confused godeps. This PR updates our file to match godeps expectation. Fixes issue 77238 * Upgrade Stackdriver Logging Agent addon image from 1.6.0 to 1.6.8. * Test kubectl cp escape * Properly handle links in tar * use k8s.gcr.io/pause instead of kubernetes/pause * Pick up security patches for fluentd-gcp-scaler by upgrading to version 0.5.2 * Error when etcd3 watch finds delete event with nil prevKV * Make CreatePrivilegedPSPBinding reentrant Make CreatePrivilegedPSPBinding reentrant so tests using it (e.g. DNS) can be executed more than once against a cluster. Without this change, such tests will fail because the PSP already exists, short circuiting test setup. * check if Memory is not nil for container stats * Bump ip-masq-agent version to v2.3.0 * In GuaranteedUpdate, retry on any error if we are working with stale data * BoundServiceAccountTokenVolume: fix InClusterConfig * fix CVE-2019-11244: `kubectl --http-cache=<world-accessible dir>` creates world-writeable cached schema files * Terminate watchers when watch cache is destroyed * honor overridden tokenfile, add InClusterConfig override tests * fix incorrect prometheus metrics * Kubernetes version v1.12.10-beta.0 openapi-spec file updates * Add/Update CHANGELOG-1.12.md for v1.12.9. * fix azure retry issue when return 2XX with error fix comments * Disable graceful termination for udp * fix: update vm if detach a non-existing disk fix gofmt issue fix build error * Fix incorrect procMount defaulting * ipvs: fix string check for IPVS protocol during graceful termination Signed-off-by: Andrew Sy Kim <kiman@vmware.com> * kubeadm: apply taints on non-control-plane node join This backports a change made in 1.13 which fixes the process applying taints when joining worker nodes. * fix flexvol stuck issue due to corrupted mnt point fix comments about PathExists fix comments revert change in PathExists func * Avoid the default server mux * Default resourceGroup should be used when value of annotation azure-load-balancer-resource-group is empty string
tssurya
added a commit
to tssurya/kubernetes
that referenced
this pull request
Oct 19, 2020
In kubernetes#56164, we had split the reject rules for non-ep existing services into KUBE-EXTERNAL-SERVICES chain in order to avoid calling KUBE-SERVICES from INPUT. However in kubernetes#74394 KUBE-SERVICES was re-added into INPUT. As noted in kubernetes#56164, kernel is sensitive to the size of INPUT chain. This patch refrains from calling the KUBE-SERVICES chain from INPUT and FORWARD, instead adds the lb reject rule to the KUBE-EXTERNAL-SERVICES chain which will be called from INPUT and FORWARD.
tssurya
added a commit
to tssurya/kubernetes
that referenced
this pull request
Oct 26, 2020
In kubernetes#56164, we had split the reject rules for non-ep existing services into KUBE-EXTERNAL-SERVICES chain in order to avoid calling KUBE-SERVICES from INPUT. However in kubernetes#74394 KUBE-SERVICES was re-added into INPUT. As noted in kubernetes#56164, kernel is sensitive to the size of INPUT chain. This patch refrains from calling the KUBE-SERVICES chain from INPUT and FORWARD, instead adds the lb reject rule to the KUBE-EXTERNAL-SERVICES chain which will be called from INPUT and FORWARD. Conflicts: pkg/proxy/iptables/proxier.go Minor conflict due to 1f7ea16
openshift-cherrypick-robot
pushed a commit
to openshift-cherrypick-robot/kubernetes
that referenced
this pull request
Oct 27, 2020
In kubernetes#56164, we had split the reject rules for non-ep existing services into KUBE-EXTERNAL-SERVICES chain in order to avoid calling KUBE-SERVICES from INPUT. However in kubernetes#74394 KUBE-SERVICES was re-added into INPUT. As noted in kubernetes#56164, kernel is sensitive to the size of INPUT chain. This patch refrains from calling the KUBE-SERVICES chain from INPUT and FORWARD, instead adds the lb reject rule to the KUBE-EXTERNAL-SERVICES chain which will be called from INPUT and FORWARD. Conflicts: pkg/proxy/iptables/proxier.go Minor conflict due to 1f7ea16
tssurya
added a commit
to tssurya/kubernetes
that referenced
this pull request
Nov 3, 2020
In kubernetes#56164, we had split the reject rules for non-ep existing services into KUBE-EXTERNAL-SERVICES chain in order to avoid calling KUBE-SERVICES from INPUT. However in kubernetes#74394 KUBE-SERVICES was re-added into INPUT. As noted in kubernetes#56164, kernel is sensitive to the size of INPUT chain. This patch refrains from calling the KUBE-SERVICES chain from INPUT and FORWARD, instead adds the lb reject rule to the KUBE-EXTERNAL-SERVICES chain which will be called from INPUT and FORWARD. Conflicts: pkg/proxy/iptables/proxier.go Minor conflict due to 1f7ea16
tssurya
added a commit
to tssurya/kubernetes
that referenced
this pull request
Nov 9, 2020
In kubernetes#56164, we had split the reject rules for non-ep existing services into KUBE-EXTERNAL-SERVICES chain in order to avoid calling KUBE-SERVICES from INPUT. However in kubernetes#74394 KUBE-SERVICES was re-added into INPUT. As noted in kubernetes#56164, kernel is sensitive to the size of INPUT chain. This patch refrains from calling the KUBE-SERVICES chain from INPUT and FORWARD, instead adds the lb reject rule to the KUBE-EXTERNAL-SERVICES chain which will be called from INPUT and FORWARD. Conflicts: pkg/proxy/iptables/proxier.go Minor conflict due to 1f7ea16
mrmassis
pushed a commit
to mrmassis/kubernetes
that referenced
this pull request
Nov 12, 2020
In kubernetes#56164, we had split the reject rules for non-ep existing services into KUBE-EXTERNAL-SERVICES chain in order to avoid calling KUBE-SERVICES from INPUT. However in kubernetes#74394 KUBE-SERVICES was re-added into INPUT. As noted in kubernetes#56164, kernel is sensitive to the size of INPUT chain. This patch refrains from calling the KUBE-SERVICES chain from INPUT and FORWARD, instead adds the lb reject rule to the KUBE-EXTERNAL-SERVICES chain which will be called from INPUT and FORWARD.
tssurya
added a commit
to tssurya/kubernetes
that referenced
this pull request
Nov 20, 2020
In kubernetes#56164, we had split the reject rules for non-ep existing services into KUBE-EXTERNAL-SERVICES chain in order to avoid calling KUBE-SERVICES from INPUT. However in kubernetes#74394 KUBE-SERVICES was re-added into INPUT. As noted in kubernetes#56164, kernel is sensitive to the size of INPUT chain. This patch refrains from calling the KUBE-SERVICES chain from INPUT and FORWARD, instead adds the lb reject rule to the KUBE-EXTERNAL-SERVICES chain which will be called from INPUT and FORWARD. Conflicts: pkg/proxy/iptables/proxier.go Minor conflict due to 1f7ea16
tssurya
added a commit
to tssurya/kubernetes
that referenced
this pull request
Nov 20, 2020
In kubernetes#56164, we had split the reject rules for non-ep existing services into KUBE-EXTERNAL-SERVICES chain in order to avoid calling KUBE-SERVICES from INPUT. However in kubernetes#74394 KUBE-SERVICES was re-added into INPUT. As noted in kubernetes#56164, kernel is sensitive to the size of INPUT chain. This patch refrains from calling the KUBE-SERVICES chain from INPUT and FORWARD, instead adds the lb reject rule to the KUBE-EXTERNAL-SERVICES chain which will be called from INPUT and FORWARD. Conflicts: pkg/proxy/iptables/proxier.go Minor conflict due to 1f7ea16
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
/kind bug
What this PR does / why we need it:
ICMP reject services with no endpoints through LBs. We already reject all other cases.
Fixes #48719
#72879 (by @marwinski) was almost identical, but I didn't see that until I had already written this. Full credit to @marwinski .
Does this PR introduce a user-facing change?: