Address a race condition that arises when running two `helm install`
commands, both of which specify a namespace that does not already exist.
In this specific instance, attempting to create a `namespace` which
already exists shouldn't be a failure which causes `helm install` to
terminate.
This adds support for installing CRDs well before any other resource
kinds are installed.
This PR introduces a new hook, `crd-install`, that fires before
manifests are even validated. It is used to install a CRD before any
other part of a chart is installed.
Currently, this hook is _only implemented for install_. That means we
currently cannot add new CRDs during `helm upgrade`, nor can they
be rolled back. This is the safest configuration, as the update/rollback
cycle gets very challenging when CRDs are added and removed.
Resolves#3655
We were seeing that when running helm upgrade with the reuse-values
flag enabled that you could end up in the position where overrides
a.k.a computed values from previous revisions were not being saved on
the updated revision. This left us in a weird position where some
computed values would disappear mysteriously in the abyss. That
happened because computed values from previous revisions weren't merged
with the new computed values every time the reuse-values flag was used.
This PR merges computed values from the previous revisions so you don't
end up in that kind of conundrum.
Existing helm.sh/hook-delete-policy annotation variables (hook-failed, hook-succeeded) do not allow to leave failed jobs for debugging without blocking the next job launching: every failed job must be deleted manually before the next related release is launching (installing, updating or rolling back).
New policy, before-hook-creation, removes the hook from previous release if there is one before the new hook is launched and can be used with another variable.
When using `helm upgrade --install`, if the first release fails, Helm will respond with an error saying that it cannot upgrade from an unknown state.
With this feature, `helm upgrade --install --force` automates the same process as `helm delete && helm install --replace`. It will mark the previous release as DELETED, delete any existing resources inside Kubernetes, then replace it as if it was a fresh install. It will then mark the FAILED release as SUPERSEDED.
* add test for rolling back from a FAILED deployment
* Update naming of release variables
Use same naming as the rest of the file.
* Update rollback test
- Add logging
- Verify other release names not changed
* fix(tiller): Supersede multiple deployments
There are cases when multiple revisions of a release has been
marked with DEPLOYED status. This makes sure any previous deployment
will be set to SUPERSEDED when doing rollbacks.
Closes#2941#3513#3275
Between grpc 1.2.x and 1.7.x there was an API change. The
previous MaxMsgSize is now a wrapper around MaxRecvMsgSize. This
change now sets the MaxRecvMsgSize and MaxSendMsgSize which need
to be set independently.
Fixes#2437
Two bugs were causing this behavior
- Tiller was marking the previous release superseded when an upgrade
failed.
- Upgrade was diffing against failed releases
When release upgrade fails, updatedRelease is already created
in a storage by *ReleaseServer.UpdateRelease, therefore we should
be updating it's status, not creating it again.
When "helm.sh/hook-delete-policy: hook-succeeded" is provided in a hook's annotation, Tiller will automatically delete the hook after the hook is succeeded. When "helm.sh/hook-delete-policy: hook-failed" is provided in a hook's annotation, Tiller will automatically delete the hook after the hook is failed.
Closes#1769
Tiller currently hangs indefinitely when deadlocks arise from certain
concurrent operations. This commit removes the nested mutex locking
system from pkg/Storage and relies on resource contention controls in k8s.
Closes#2560
* give an uniform check for release process
* fixed as the review of adamreese: update the err message when releasename is empty and update the test units.
* fixed as the review of bacongobbler: add more detail information to return message. the regex rule is added to the return message.
For some reason, GetReleaseContent was configured to return the latest
release only if it is in state DEPLOYED. But a release with a version is
returned regardless of release. This made it really hard to debug failed
releases, even though we have the data to show.
Closes#2525
This builds on previous work and adds more logging to follow
the full process of installing, updating, deleting, and rolling back.
This also standardizes capitalization of logs and small formatting
fixes
Extracted delete specific code from ReleaseServer to external function
which is called from both Local and Remote ReleaseModules.
Made getVersionSet function from tiller package exported.
This change introduces the concept of Rudders - pluggable modules that
Tiller communicates with via grpc, which allow to decouple orchestration
logic from Tiller into separate service.
This commit consists of simple Rudder implementation which does exactly
the same thing as built in Tiller orchestrator - it creates all k8s
objects from provided manifest without orchestrating them.
--experimental-release flag is introduced to enable this behaviour.
This change allows to use the service and tiller outside of the cluster.
Following commits will add Rudder to helm deployment.
Adds an `unreleased` flag to the version if not building from HEAD of a
tag. The compatibility check is bypassed if the client or server are
unreleased.
fixes#2110
Environment is supplied with release lock map which allows to lock a
release by name to make sure that update, rollback or uninstall aren't
running on one release at the same time.
It is now possible to create plugins with chart download capabilities for custom, non-http protocols.
Furthermore it is possible to reuse helm packages to implement alternative clients with these custom downloader functions.