-
Notifications
You must be signed in to change notification settings - Fork 24
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
parse serviceName in VerticaReplicator #826
base: main
Are you sure you want to change the base?
Conversation
This pulls in the latest vclusterops library changes. Many of the changes that we are pulling in were changes related to improving the vcluster CLI.
This pulls in the new signature for VStartDatabase. A new parameter was returned, which we can safely ignore for the operators usage.
During a replicated upgrade, we must categorize the subclusters into one of two replica groups. Initially, all subclusters participating in the upgrade will be placed in replica group A. Subsequently, new subclusters will be generated throughout the process and assigned to the second group, named replica group B. This change manages the assignment of subclusters at the beginning of the upgrade process. To facilitate this change, new status fields will be introduced to the VerticaDB, enabling the tracking of this state across reconcile iterations.
This adds a new controller for sandbox. It is going to watch a ConfigMap that has labels on it to indicate it contains state for a sandbox. That ConfigMap also contains the vdb name whose subclusters are involved in sandbox/unsanbox. For now it simply reads the VerticaDB found in the configMap and do nothing else. Some changes are also made to the VerticaDB API to add new fields related to sanbox
This PR eliminates labels from the `VerticaReplicator` sample config. The intention is to enhance the user experience on OpenShift when generating this CR from the webUI. OpenShift utilizes this sample to prepopulate fields in the webUI interface.
This adds a few webhook rules for replicated upgrade. It's probably not a complete list, but we can always add more later when we find rules to check. This commit adds the following: - ensures the subclusters involved in the upgrade are stable. They cannot be removed and the sizes can't change. - We do allow new subclusters to be added to one of the replica groups. But they must be secondary subclusters. No new primaries. - ensures that subclusters listed in the `.status.updateState.replicaGroups` aren't repeated.
We now collect the sandbox name in podfacts. We updated the vsql query to return the extra state. We built an interface to get the information as we intend to use a REST endpoint in the future but still need support via vsql for older releases.
This adds a new leg, e2e-leg-9, to the CI. This is going to be used to test changes done in 24.3.0. One new change with this is that it will use a vertica license so that we can test scaling past 3 nodes. This will be required in order to test out the new replicated upgrade work.
#764) `vrep` reconciler enforces the minimum source and target db versions to be at least 24.3.0 --------- Co-authored-by: Roy Paulin <rnguetsopken@opentext.com> Co-authored-by: spilchen <mspilchen@opentext.com>
This PR adds below webhooks rules for sandboxes in CRD: - cannot scale (up or down) any subcluster that is in a sandbox - cannot remove a subcluster that is sandboxed. It must be unsandboxed first. - cannot have multiple sandboxes with the same name - cannot have the image of a sandbox be different than the main cluster before the sandbox has been setup - cannot be used on versions older than 24.3.0 - cannot be used before the database has been initialized - cannot have duplicate subclusters defined in a sandbox - cannot have a subcluster defined in multiple sandboxes - cannot have a non-existing subcluster defined in a sandbox We could add more rules later for sandboxes when we have needs.
we want to use restart reconciler in both the VerticaDB controller and the sandbox controller. This makes some changes to the restart reconciler in order to achieve that. The sandbox name (empty string if VerticaDB controller) is passed to the reconciler and it uses it to target only pods belonging to that sandbox( or to target only pods belonging to the main cluster if sandbox name is empty)
This change adds stubs to the replicated upgrade reconciler. It attempts to have the functions present that are needed to do the upgrade. Due to dependencies, such as sandboxing and replication, comprehensive testing is currently limited to unit tests. While this logic might evolve during integration, it gives a framework for how the upgrade should function. Additionally, I identified new requirements for state management. Two new annotations have been introduced to the VerticaDB: - vertica.com/replicated-upgrade-sandbox: to record the name of the sandbox created for the upgrade - vertica.com/replicated-upgrade-replicator-name: to track the name of the VerticaReplicator CR utilized for replicating data to the sandbox.
This PR adds a new webhook rule for sandboxes in CRD: cannot have a primary subcluster defined in a sandbox.
This adds a new fetcher in podfacts to fetch node details from the running database inside the pod. This new fetcher will call VClusterOps API which will send an HTTP request to the database. The new fetcher will be enabled when VclusterOps annotation is set, and Vertica server version is not older than v24.3.0. The new fetcher should be faster and more reliable than the old fetcher which will execute vsql within the pod.
In the replicated upgrade process, we must pause and redirect client connections from subclusters in replica group A to those in replica group B. This is the initial stage of the change. Currently, pause/redirect semantics are not supported, as they require new server modifications that have not yet been implemented. Therefore, we will perform an old-style drain to pause the process and maintain service labels correctly to point to subclusters in replica group B for redirection. --------- Co-authored-by: Roy Paulin <rnguetsopken@opentext.com>
This change will modify the upgrade reconcilers, both offline and online, to function in either the main cluster or a sandbox. Our immediate plan is to use the offline upgrade reconciler within the sandbox controller, although this will be done in a follow-on task.
This pull request modifies the Vertica replicator controller to invoke the replication command synchronously. Additionally, it introduces new status conditions within VerticaReplicator to monitor the controller's various states. --------- Co-authored-by: spilchen <mspilchen@opentext.com>
If we had selected an upgrade method but ended up failing back to a different method because something was incompatible, we are going to log an event. For instance, if we requested something other than offline upgrade, but we end up in the offline upgrade code path we will have an event message like this: ``` Normal IncompatibleUpgradeRequested 7m20s verticadb-operator Requested upgrade is incompatible with the Vertica deployment. Falling back to offline upgrade. ```
This PR adds sandbox reconciler to VerticaDB controller. The reconciler will add subclusters to sandboxes in the database. The change of sandbox fields in CRD will trigger sandbox reconciler, then the reconciler will call vclusterOps to add subclusters to sandboxes.
This will add a status message to the `.status.upgradeStatus` field in the VerticaDB as the operator goes through the various stages of replicated upgrade. It reuses the same design for the other upgrade methods in that we only advance the status message forward. This means that we need to reconcile an iteration, we won't report the first status message again.
Automated changes by pre-release.yml GitHub action Signed-off-by: GitHub <noreply@github.com> Signed-off-by: Matt Spilchen <mspilchen@opentext.com> Co-authored-by: spilchen <spilchen@users.noreply.github.com>
36598f8
to
e478aec
Compare
Could we update the PR description please? |
needs one small vcluster change that's under review but then this should be good to merge |
61194a4
to
4789914
Compare
This PR updates the Vertica image to `20240606` to avoid replication errors from the server side. This is a temporary fix, and we will revert this change after fixing the issue on the server side
This PR build a reconciler so that during the upgrade we can just promote the sandbox after all of the data has been replicated to it. --------- Co-authored-by: cchen-vertica <cai.chen@microfocus.com>
f921fa0
to
74459be
Compare
I've synced the vclusterOps to the Github repository. You can update your go.mod to refer to the latest commit in vcluster repo. |
74459be
to
919062d
Compare
This PR used vsql to close all active user sessions before database replication starts in the replicated upgrade. This can speed up the upgrade process. This is a temporary change. We will improve this by pausing all user sessions soon.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Very close. Sorry if what I say is often confusing. Let me know If I misunderstood what you are trying to do.
pkg/vadmin/replication_start_vc.go
Outdated
opts.SourceTLSConfig = s.SourceTLSConfig | ||
opts.IsEon = v.VDB.IsEON() | ||
|
||
opts.IPv6 = net.IsIPv6(s.SourceIP) |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
I guess this should not be a problem. @cchen-vertica ?
…pgrade (#829) This PR added the last two steps of replicated upgrade: 1. remove the old main cluster. After we promote the sandbox, the old main cluster is not needed anymore so we remove it. 2. rename the subclusters in new main cluster. After we promote the sandbox, we rename all subclusters in the sandbox to match the ones in original main cluster.
With the change on the server side, we have already fixed the issue in #832. We can revert to the latest Vertica image to fix the replication issue
The latest vclusterOps improves the promote_sandbox API to clean communal storage after sandbox promotion. This PR simply updates vclusterOps library to clean communal storage in the online upgrade.
After promoting sandbox to main, we need to delete sandbox config map to avoid any kind of conflicts as the sandbox does not exist anymore.
Almost done! Resolve the conflicts and run the tests again. |
565f795
to
8a286fe
Compare
8a286fe
to
768e30d
Compare
spilchen seems not to be a GitHub user. You need a GitHub account to be able to sign the CLA. If you have already a GitHub account, please add the email address used for this commit to your account. You have signed the CLA already but the status is still pending? Let us recheck it. |
allow users to provide a service name to use for source and target databases. if unspecified we use the first service assigned to the database with a cluster ip, if supplied we validate that it's assigned to the appropriate main/sandbox cluster