A tag for anything related to Kubernetes. For the discussion see T147187: Create a tag for #kubernetes.
See also:
A tag for anything related to Kubernetes. For the discussion see T147187: Create a tag for #kubernetes.
See also:
Change #1153978 merged by jenkins-bot:
[operations/deployment-charts@master] cfssl-issuer: Allow to provide a custom CA certificate store
Change #1154266 merged by JMeybohm:
[operations/software/cfssl-issuer@main] Make simple-cfssl usable for local WMF PKI deployments
Change #1155212 merged by jenkins-bot:
[operations/deployment-charts@master] Add a script to visualize the dependencies of admin_ng environments
I 've gone ahead and switch all of aux-k8s to MTU 1460. This time around, I went for a more hands off approach, namely:
Change #1155543 merged by Alexandros Kosiaris:
[operations/puppet@production] aux-k8s: Switch MTU to 1460
Change #1155543 had a related patch set uploaded (by Alexandros Kosiaris; author: Alexandros Kosiaris):
[operations/puppet@production] aux-k8s: Switch MTU to 1460
In T352956#10900449, @akosiaris wrote:In T352956#10900403, @cmooney wrote:@akosiaris a quick question about this:
meaning that ICMP traffic to e.g. coredns gets dropped
In terms of pmtud that means that if coredns sends large UDP packets - which get dropped elsewhere - it won't get the ICMP "packet too big" messages back. But that is not really a worry. The CoreDNS PODs have a lower MTU than pretty much everything on the network, they are not going to send packets that are too large for anything else.
Agreed on this.
Are the physical K8s hosts blocked from sending ICMP? So for instance if a 1500-byte UDP packet was sent to a pod IP - and couldn't get there because we have reduced the MTU on the veth interface connecting the POD - can the host send an ICMP back to the client?
No, they are not blocked. Indeed the host could send that ICMP back instead. I didn't see that happening in my tests, however I also didn't specifically try this out, we can test that.
@akosiaris thanks for confirming. So overall my thinking is:
In T352956#10900403, @cmooney wrote:@akosiaris a quick question about this:
meaning that ICMP traffic to e.g. coredns gets dropped
In terms of pmtud that means that if coredns sends large UDP packets - which get dropped elsewhere - it won't get the ICMP "packet too big" messages back. But that is not really a worry. The CoreDNS PODs have a lower MTU than pretty much everything on the network, they are not going to send packets that are too large for anything else.
@akosiaris a quick question about this:
Dependencies are fixed now for wikikube clusters. Other cluster maintainers might want to check/update releases which are not part what gets deployed to all clusters.
I've added the script to the root of the repo since we already have a bunch of scripts there.
Change #1155212 had a related patch set uploaded (by JMeybohm; author: JMeybohm):
[operations/deployment-charts@master] Add a script to visualize the dependencies of admin_ng environments
Change #1153975 merged by jenkins-bot:
[operations/deployment-charts@master] Use Wikimedia DNS IPs as mock
Add gitlab-ci pipeline to build a cfssl container image
Add sre/cfssl to the trusted runners with dockerfile support
Change #1154293 had a related patch set uploaded (by JMeybohm; author: JMeybohm):
[operations/deployment-charts@master] kind.sh can bootstrap a wikikube like cluster with kind
Change #1153982 merged by jenkins-bot:
[operations/deployment-charts@master] admin_ng: Fix dependencies/needs of helmfiles
Change #1153979 merged by jenkins-bot:
[operations/deployment-charts@master] admin_ng: Split envoyfilters installation into a separate release
Change #1154266 had a related patch set uploaded (by JMeybohm; author: JMeybohm):
[operations/software/cfssl-issuer@main] Make simple-cfssl usable for local WMF PKI deployments
We dismantled the CI access to k8s staging in T288629: Automated validation of mediawiki-multiversion images, so access from Gitlab runners to the ci namespace are no longer possible (the namespace does not exist anymore). If this is actually used/required, we need to find a different way. Otherwise we should kubestagemaster from the allowed_services
In T389080#10887016, @JMeybohm wrote:I made a small script to visualize dependencies while fixing them, not sure where to put it so:
Change #1153982 had a related patch set uploaded (by JMeybohm; author: JMeybohm):
[operations/deployment-charts@master] admin_ng: Fix dependencies/needs of helmfiles
Change #1153979 had a related patch set uploaded (by JMeybohm; author: JMeybohm):
[operations/deployment-charts@master] admin_ng: Split envoyfilters installation into a separate release
Change #1153979 had a related patch set uploaded (by JMeybohm; author: JMeybohm):
[operations/deployment-charts@master] admin_ng: Split envoyfilters installation into a separate release
Change #1153978 had a related patch set uploaded (by JMeybohm; author: JMeybohm):
[operations/deployment-charts@master] cfssl-issuer: Allow to provide a custom CA certificate store
Change #1153977 had a related patch set uploaded (by JMeybohm; author: JMeybohm):
[operations/deployment-charts@master] coredns: Run coredns on an unprivileged port (5353) instead of 53
Change #1153976 had a related patch set uploaded (by JMeybohm; author: JMeybohm):
[operations/deployment-charts@master] calico: Add support to manage CNI installation by daemonset
Change #1153975 had a related patch set uploaded (by JMeybohm; author: JMeybohm):
[operations/deployment-charts@master] Use Wikimedia DNS IPs as mock
I made a small script to visualize dependencies while fixing them, not sure where to put it so:
Change #1114000 merged by jenkins-bot:
[operations/cookbooks@master] k8s.pool-depool-node: Add support to downtime/remove downtime
Change #1127950 merged by jenkins-bot:
[operations/deployment-charts@master] aux-k8s-services/*: use the correct helm version in each cluster
Recycled all the pods in ml-serve-eqiad to be sure, no PSS violation registered. Migration completed!
Change #1152194 merged by Elukey:
[operations/puppet@production] kubernetes: disable PSP for ml-serve-eqiad
Change #1152194 had a related patch set uploaded (by Elukey; author: Elukey):
[operations/puppet@production] kubernetes: disable PSP for ml-serve-eqiad
Change #1152190 merged by Elukey:
[operations/deployment-charts@master] admin_ng: disable PSP and enable PSS for ml-serve-eqiad
Change #1152190 had a related patch set uploaded (by Elukey; author: Elukey):
[operations/deployment-charts@master] admin_ng: disable PSP and enable PSS for ml-serve-eqiad
@klausman since today it was very quiet for ML, I took the opportunity to apply all the changes stated in T369493#10792884 (including recycling all the isvc pods in ml-serve-eqiad).
Change #1151604 merged by Elukey:
[operations/deployment-charts@master] admin_ng: set secure-pod-defaults to "enabled" for knative clusters
Change #1151600 merged by Elukey:
[operations/deployment-charts@master] kserve-inference: set seccomp defaults in the chart