bugfix:incomplete arp cause busy loop (high cpu util) #7421
Conversation
Incomplete ARP entry (nil mac address)can cause the busy loop
marvin-tigera
added
release-note-required
|
|
|
@mgleung Can you have a look this, many thanks:) |
|
cc @StevenTigera |
Co-authored-by: Shaun Crampton <shaun@tigera.io>
|
@fasaxc thanks for reviewing :) |
|
/sem-approve |
fasaxc
added
cherry-pick-candidate
docs-not-required
marvin-tigera
added
docs-pr-required
|
@detailyang Looks like the linter failed, please can you run |
@fasaxc Can you restart the CI again. It's failed at See https://tigera.semaphoreci.com/jobs/20627b73-30d4-4312-85da-0010d847a343/plain_logs.txt Generating manifest from charts/values/./canal-etcd.yaml
walk.go:74: found symbolic link in path: /home/semaphore/calico/charts/calico/crds resolves to /home/semaphore/calico/libcalico-go/config/crd. Contents of linked file included and used
make[2]: Leaving directory '/home/semaphore/calico'
make[1]: Leaving directory '/home/semaphore/calico'
make check-dirty
make[1]: Entering directory '/home/semaphore/calico'
The following files are dirty
felix/routetable/route_table.go | 4 �[32m++�[m�[31m--�[m
1 file changed, 2 insertions(+), 2 deletions(-)
make[1]: *** [lib.Makefile:1197: check-dirty] Error 1 |
|
/sem-approve |
|
@detailyang did you commit/push the result of |
fixed. |
|
/sem-approve |
felix/routetable/route_table.go
Outdated
| @@ -1035,6 +1035,10 @@ func (r *RouteTable) syncL2RoutesForLink(ifaceName string) error { | |||
| var updatesFailed bool | |||
|
|
|||
| for _, existing := range existingNeigh { | |||
| if existing.HardwareAddr == nil { | |||
| log.WithField("entry", entry).Debug("Ignoring existing ARP entry with no hardware addr") | |||
There was a problem hiding this comment.
CI now failing with entry undefined
Signed-off-by: detailyang <detailyang@gmail.com>
…atch-1 Signed-off-by: detailyang <detailyang@gmail.com>
|
@fasaxc |
|
would you backport lower version ? |
|
/sem-approve |
Description
Hello.
we have identified an issue with incomplete ARP entries that may cause a busy loop for calico-node in production.
We are unsure why every node is generating incomplete ARP entries, as shown in the attached image
These entries are causing the calico-node to experience high CPU utilization and become stuck in the syncL2RoutesForLink process. Specifically, the incomplete ARP entry, which holds a nil MAC address, is causing the kernel function rtnl_fdb_del to fail when attempting to remove the FDB entry
https://elixir.bootlin.com/linux/v3.10.108/source/net/core/rtnetlink.c#L2223
As a workaround, we can choose to ignore the incomplete ARP entries for now and continue with normal operations :)
fixes #5460
Related issues/PRs
Todos
Release Note
Reminder for the reviewer
Make sure that this PR has the correct labels and milestone set.
Every PR needs one
docs-*label.docs-pr-required: This change requires a change to the documentation that has not been completed yet.docs-completed: This change has all necessary documentation completed.docs-not-required: This change has no user-facing impact and requires no docs.Every PR needs one
release-note-*label.release-note-required: This PR has user-facing changes. Most PRs should have this label.release-note-not-required: This PR has no user-facing changes.Other optional labels:
cherry-pick-candidate: This PR should be cherry-picked to an earlier release. For bug fixes only.needs-operator-pr: This PR is related to install and requires a corresponding change to the operator.