Support for multiple replicas with auto-tls #198
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
Uses a leader-elector sidecar to ensure only one injector replica generates the CA and cert+key. The other replicas pick up the cert+key from a k8s secret for use in their TLS listeners. The leader-elector sidecars coordinate using the annotations of a k8s Endpoint object, which is why those extra permissions were added to the role in the deployment yaml. Build a dev image: `make image VERSION=dev` Deploy: `kubectl apply -k deploy/ -n vault`
Added command-line and env option to control leader-elector usage. Using Secrets informer for followers to ensure retrieving timely cert updates without overloading the k8s api.
Added ttl and health checks to leader-elector. Added AGENT_INJECT_USE_LEADER_ELECTOR env option to deployment.
Slightly refactored the leader package so it's more testable
Co-authored-by: Tom Proctor <tomhjp@users.noreply.github.com>
…r' into VAULT-84/leader-elector-container
Using the community gcr registry (k8s.gcr.io), election arg to match the endpoint deploy yaml, and updating the secret name to match the code.
To ensure the behavior is unchanged for users not using the leader elector logic.
Passing the CA around in the Secret seemed to prevent followers from recreating a CA if they're promoted to leader. Since the followers don't need the CA, removed it from the Secret.
| // For followers, run different function here that reads bundle from Secret, | ||
| // and returns that in the result. That will flow through the existing | ||
| // notify channel structure, testing if it's the same cert as last, etc. | ||
| if !leaderCheck { |
There was a problem hiding this comment.
Can we get a log saying whether this pod is the leader or follower? Also wondering if we should apply a label so we can query for the leader (not blocking, future enhancement probably).
Currently I can't tell which is which:
[~] kubectl logs vault-injector-57485f744-24jwd -c sidecar-injector
Using leader elector logic
2020-12-16T15:22:43.623Z [INFO] handler: Starting handler..
Listening on ":8080"...
Updated certificate bundle received. Updating certs...
[~] kubectl logs vault-injector-57485f744-lts8w -c sidecar-injector
Using leader elector logic
2020-12-16T15:22:44.109Z [INFO] handler: Starting handler..
Listening on ":8080"...
Updated certificate bundle received. Updating certs...
There was a problem hiding this comment.
Yep, good point, added more logging in ac4b4f4. I set the leader and follower log lines to debug since they are logged pretty frequently. Looks something like this:
❯ kubectl logs -n vault vault-agent-injector-c6d5c7689-6q4w7 -c sidecar-injector
Using leader elector logic
2020-12-17T02:36:22.710Z [INFO] handler: Starting handler..
Listening on ":8080"...
2020-12-17T02:36:22.719Z [INFO] handler.auto-tls: Currently the leader
2020-12-17T02:36:22.726Z [INFO] handler.auto-tls: Generated CA
2020-12-17T02:36:22.734Z [INFO] handler.certwatcher: Updated certificate bundle received. Updating certs...
2020-12-17T02:36:27.711Z [INFO] handler.auto-tls: Currently the leader
❯ kubectl logs -n vault vault-agent-injector-c6d5c7689-lmrjb -c sidecar-injector
Using leader elector logic
2020-12-17T02:36:26.307Z [INFO] handler: Starting handler..
Listening on ":8080"...
2020-12-17T02:36:26.314Z [DEBUG] handler.auto-tls: Currently a follower
2020-12-17T02:36:26.314Z [INFO] handler.certwatcher: Updated certificate bundle received. Updating certs...
2020-12-17T02:36:31.308Z [DEBUG] handler.auto-tls: Currently a follower
...
jasonodonnell
approved these changes
Dec 16, 2020
Added a named logger to GenSource (auto-tls), mostly debug-level for telling which is the leader and which is the follower.
While the leader is waiting for the current certificate to expire, a leadership change could occur, and then the former leader's certificate would be out of sync with the new leader. Added a goroutine that signals a channel on a leadership change.
NLRemco
pushed a commit
to NLRemco/vault-k8s
that referenced
this pull request
Feb 22, 2022
Uses a leader-elector sidecar to ensure only one injector replica generates the CA and cert+key. The other replicas pick up the cert+key from a k8s secret for use in their TLS listeners. The leader-elector sidecars coordinate using the annotations of a k8s Endpoint object, which is why those extra permissions were added to the role in the deployment yaml. Added command-line and env option to control leader-elector usage. Using Secrets informer for followers to ensure retrieving timely cert updates without overloading the k8s api. Co-authored-by: Tom Proctor <tomhjp@users.noreply.github.com>
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.
Adds a
use-leader-electoroption to vault-k8s to enable use of a leader-elector sidecar, which lets multiple replicas of the injector determine a "leader" which is then in charge of generating the CA, Cert, and Key, distributing them to the followers, and patching the webhook caBundle. Defaults tofalse.The leader-elector sidecar in use is described in detail here: https://kubernetes.io/blog/2016/01/simple-leader-election-with-kubernetes/
The leader-elector containers determine a leader by modifying an Endpoint object, and they expose this information to the injector pod with a local http service that returns the current leader's hostname.
The followers read the cert and key needed for the webhook listener from a k8s Secret, which is updated by the leader when a certificate is near expiration.