Solves #3722 by making the changes in #3539 more compatible with the previous behavior.
This gives a recovery option for "oops I deleted my helm release" by allowing rollback, which is intended to be a working feature of helm. Note that purging releases removes the history required to rollback, so this doesn't work in that case. However, purging takes significantly more time, so it's harder to accidentally purge everything.
Signed-off-by: Brent <bmperrea@gmail.com>
Upgrading a release and override existing values doesn't work as expected for nested values. Maps should be merged recursively, but currently maps are treated just like values and replaced at the top level.
If the existing values are:
```yaml
resources:
requests:
cpu: 400m
something: else
```
and an update is done with ```--set=resources.requests.cpu=500m```, it currently ends up as
```yaml
resources:
requests:
cpu: 500m
```
but it should have been
```yaml
resources:
requests:
cpu: 500m
something: else
```
This PR updates the way override values are merged into the existing set of values to merge rather than replace maps.
Closes: #4792
Signed-off-by: Morten Torkildsen <mortent@google.com>
While investigating a tiller crash on v2.10.0 (see recent comments in #3125), I pulled down the code
and wrote a test replicating the crash I was experiencing. I then
realized that the crash had been fixed, and was able to locate the fix
in #4630 after running a quck bisect.
Since there don't appear to be any tests that cover this crash, and I
had written one myself, I figured I might as well put up a PR for it.
Here's what the test failure on v2.10.0 looks like:
```
-- FAIL: TestUpdateReleasePendingInstall_Force (0.00s)
panic: runtime error: invalid memory address or nil pointer dereference [recovered]
panic: runtime error: invalid memory address or nil pointer dereference
[signal SIGSEGV: segmentation violation code=0x1 addr=0x50 pc=0x1d128d8]
goroutine 235 [running]:
testing.tRunner.func1(0xc420493c20)
/usr/local/Cellar/go/1.10/libexec/src/testing/testing.go:742 +0x29d
panic(0x1eb8d80, 0x2a12db0)
/usr/local/Cellar/go/1.10/libexec/src/runtime/panic.go:505 +0x229
k8s.io/helm/pkg/tiller.(*ReleaseServer).performUpdateForce(0xc4208210b0, 0xc4202c6dc0, 0x0, 0x0, 0x2174220)
/Users/mattrasmus/go/src/k8s.io/helm/pkg/tiller/release_update.go:166 +0x188
k8s.io/helm/pkg/tiller.(*ReleaseServer).UpdateRelease(0xc4208210b0, 0x2191780, 0xc420820f30, 0xc4202c6dc0, 0x29aeb90, 0x38, 0x2d2)
/Users/mattrasmus/go/src/k8s.io/helm/pkg/tiller/release_update.go:43 +0x245
k8s.io/helm/pkg/tiller.TestUpdateReleasePendingInstall_Force(0xc420493c20)
/Users/mattrasmus/go/src/k8s.io/helm/pkg/tiller/release_update_test.go:549 +0x120
testing.tRunner(0xc420493c20, 0x20e5c70)
/usr/local/Cellar/go/1.10/libexec/src/testing/testing.go:777 +0xd0
created by testing.(*T).Run
/usr/local/Cellar/go/1.10/libexec/src/testing/testing.go:824 +0x2e0
FAIL k8s.io/helm/pkg/tiller 0.118s
```
Signed-off-by: Matt Rasmus <mrasmus@betterworks.com>
* Fix race in helm list when partitioning
Problem:
The chunks slice that is passed through the channel is reused for each
partition. This means that encoding the release into a message is racing with
populating the next partition, causing the results to sometimes not fit in the
message, and the release list to be incorrect
Solution:
Allocate a new slice for each partition
Issue #3322
Signed-off-by: Brian Marshall <bmarshall13@users.noreply.github.com>
(cherry picked from commit a0858e29d8)
* fix import sorting
Signed-off-by: Matthew Fisher <matt.fisher@microsoft.com>
* ref(release_server_test): use NewReleaseServer()
Signed-off-by: Matthew Fisher <matt.fisher@microsoft.com>
* add unit test for race condition in `helm list`
Signed-off-by: Matthew Fisher <matt.fisher@microsoft.com>
resolves#4337
Merging maps inside of strings gets a bit tricky. When two
strings consisting of "{}" were being added together, this resulted in
"{}\n{}" instead of "{}" which is what we wanted. This led to YAML
parsing errors and showed up when the `--reuse-values` flag was used
when no overrides via `--set` were provided during install and/or
upgrade.
It's really easy to cause an import cycle on this type; this resolves
the problem by moving it out of the tiller pkg into its own. An alias is
left behind in order to prevent downstream breakage.
If `helm upgrade` fails because a reused name is still in use, the error
message does not specify which name is still in use. Update the error
message.
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.
Brent
Morten Torkildsen
adshmh
Matt Rasmus
Matt Tucker
Adam Reese
Colin Panisset
Steve Wolter
Matthew Fisher
Arash Deshmeh
Michelle Noorali
Michelle Noorali
Mike Lundy
mattjmcnaughton
Matt Butcher
Sergii Manannikov
Nikhil Manchanda
Taylor Thomas
Stuart Leeks
Timofey Kirillov
fibonacci1729
Alexey igrychev
Johnny Bergström
Matt Farina
xuhaigang
Adam Reese
Federico Gimenez