[DownstreamMerge] Merge 2021-10-13 #796

trozet · 2021-10-13T21:45:44Z

Passes unit tests 😥

Add quotes around nexthop and dst-ip fields

The previous gomega matching function `ContainsElement` doesn't assert that objects which are deleted, are also not retrieved in the final result. Ex: item1, item2 exists, item1 is deleted. With `ContainsElement(item2)` we cannot fully assert that item1 is gone. With `ConsistOf(item1)` we can. Signed-off-by: Alexander Constantinescu <aconstan@redhat.com>

Succesor for PR 2331 [fix reserve joinSwitch LRP IPs]

Function was accidentally using the non-hashed names. Signed-off-by: Tim Rozet <trozet@redhat.com>

[libovsdb] Update test matcher for libovsdb

Fixes ensure address set

Signed-off-by: Alexander Constantinescu <aconstan@redhat.com>

This new version includes support for multiple DGPs per router, which is needed for an upcoming commit. Signed-off-by: Han Zhou <hzhou@ovn.org>

This patch sets gateway-chassis to the node itself for each logical router port that connects node level switch to the cluster router. This makes use of the OVN optimization [0] that avoids flood-filling unrelated datapaths by ovn-controller on each node, which perfectly fits the ovn-kubernetes deployments where each logical switch is bound to a chassis, significantly improving scalability of ovn-controller (10 times faster for recompute and 70-80% memory savings for ovn-controller + OVS combined. See more details in [0]). Note: this optimization is disabled for gateway_mode = local for now, which requires NAT on the cluster router, which is not supported when multiple DGPs are used on the same router. The support may be added in the future. [0] ovn-org/ovn@22298fd Signed-off-by: Han Zhou <hzhou@ovn.org>

Signed-off-by: Tim Rozet <trozet@redhat.com>

Update installation documentation and add a walkthrough for an installation with kubeadm. Signed-off-by: Andreas Karis <ak.karis@gmail.com>

Signed-off-by: Alexander Constantinescu <aconstan@redhat.com>

[libovsdb] Modify matchers to take into account optional field values

Dockerfile.fedora: Update ovn build to ovn-21.06.0-15.fc33.

Use DGP to connect logical switches to the cluster router.

Adding a small but important fix to the CI documentation to export missing variables. Signed-off-by: Andreas Karis <ak.karis@gmail.com>

[libovsdb] Bump to main: optional field value fix for in-mem DB

Fixes backwards values for election timer

Currently we call addPodExternalGW from both ensurePod and addLogicalPort. If ensurePod is called with addPort=false from the UpdatePodHandler, it will keep trying to call addPodExternalGW to add routes and policies repeatedly for the same pod. During exgw pod creation, we see same routes and policies getting added 3 times in a row. Note that ovn-org/ovn-kubernetes#2337 fixed this by adding a check into addGWRoutesForNamespace to return if routes already exist for the pod, but this comes later in the code flow. Its better to not call addPodExternalGW at all unless needed. This would save time and help with pod latency issues specially at scale. Signed-off-by: Surya Seetharaman <suryaseetharaman.9@gmail.com>

Fix ensurePod to call addPodExternalGW only for annotation updates

…low-up Documentation: Improve CI documentation

Documentation: Update installation documentation

factory: split watcher creation and start

Geneve packets coming from external should skip conntrack and go directly to host. Due to a typo, they were sent out back to external for IPv6. Fix this. Signed-off-by: Andreas Karis <ak.karis@gmail.com>

No reason to have an internal cache of ips anymore now that we have libovsdb. Signed-off-by: Tim Rozet <trozet@redhat.com>

Shared GW: Fix wrong action for IPv6 incoming Geneve packets

Get rid of address sets lock and ips cache

When populating the cache, if for some reason the node's GW LRP already has that IP, don't fail. E0826 13:45:04.449574 1 namespace.go:587] Failed to get join switch port IP address for node ip-10-0-152-74.us-west-2.compute.internal: provided IP is already allocated E0826 13:45:04.620345 1 master.go:1260] Failed to get join switch port IP address for node ip-10-0-152-74.us-west-2.compute.internal: provided IP is already allocated Signed-off-by: Dan Williams <dcbw@redhat.com>

When a node gets deleted, the logical switch and gateway router get removed first. However we do not update the LBCache and this causes dependency failures during deletion leaving behind wrong ip->vip pairing in the lbs. When the pods on this node go away, ensureLBs gets triggered and it tries to update endpoints across load balancers. This set load_balancer command is batched with ls-lb-del & ls-lr-del commands to remove association of the lbs from the removed switches and routers. These commands fail together because the switches and routers are not present in ovn. This PR updates the LBcache as soon as we remove the routers and switches. Signed-off-by: Surya Seetharaman <suryaseetharaman.9@gmail.com>

openshift-ci · 2021-10-28T14:46:49Z

[APPROVALNOTIFIER] This PR is APPROVED

This pull-request has been approved by: alexanderConstantinescu, trozet

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:

~~OWNERS~~ [trozet]

Approvers can indicate their approval by writing /approve in a comment
Approvers can cancel approval by writing /approve cancel in a comment

astoycos · 2021-10-28T14:57:12Z

I ran this against upstream's CI here -> https://github.com/astoycos/ovn-kubernetes/runs/4035376913?check_suite_focus=true Signal is looking alright so far (I would expect the upgrade jobs to fail)

openshift-bot · 2021-10-28T14:57:14Z

/retest-required