1. only implement post-renderer flags this patch
2. As mumoshu advise, add helmfile flags `--post-render` and add the
postRenderer config in helmDefaults and release. the priority is
helmfile flags > release > helmDefaults.
3. fix the test case in state_test.go and some other tests.
Signed-off-by: guofutan <guofutan@tencent.com>
Signed-off-by: yxxhero <aiopsclub@163.com>
* Fix presync hooks are not called on no diff when run apply subcommand
Signed-off-by: xiaomudk <xiaomudk@gmail.com>
* Update docs/index.md
Co-authored-by: Yusuke Kuoka <ykuoka@gmail.com>
Signed-off-by: xiaomudk <xiaomudk@gmail.com>
Signed-off-by: xiaomudk <xiaomudk@gmail.com>
Co-authored-by: Yusuke Kuoka <ykuoka@gmail.com>
Use the new runWithLogCapture helper instead of the long boilerplate to capture the log for snapshot testing.
Signed-off-by: Yusuke Kuoka <ykuoka@gmail.com>
Allow configuring the lockfile in the state. This makes it possible for
example maintain a lock per environment.
Signed-off-by: Lassi Pölönen <lassi.polonen@iki.fi>
Signed-off-by: Lassi Pölönen <lassi.polonen@iki.fi>
This improves the `helmfile sync` performance.
From the code: `BuildDeps` is used only by `runHelmDepBuilds`, which
only is used by `PrepareCharts` which is finally only used by
`withPreparedCharts`.
`withPreparedCharts` already does `SyncReposOnce` which means we do not
have to refresh the local repository cache on each chart build.
This is only supported in Helm v3.
This seems to be mostly affecting helmfiles which have a lot of releases
and those release charts use sub dependencies.
I saw significant performance improvements for a helmfile with 45
releases, 2 repositories, and most of the charts also had their own
dependencies. Results:
Before the patch:
* real 9m10.565s
* real 9m38.335s
* real 9m14.941s
* real 5m13.106s (with cache)
After the patch:
* real 6m51.965s
* real 6m36.605s
* real 6m31.685s
* real 3m0.271s (with cache)
These were tested with:
```
rm -rf ~/.cache/helmfile ~/.cache/helm ~/.config/helm/repositories.* && helmfile sync ...
```
The result with `(with cache)` was without deleting the caches first.
From these metrics it seems that the sync duration decreased 20-45%
depending on the run, release count, dependencies and if the cache was
used or not.
As far as I understand, this should be backward-compatible change.
Signed-off-by: Indrek Juhkam <indrek@urgas.eu>
Signed-off-by: Indrek Juhkam <indrek@urgas.eu>
I had been unhappy with the fact that our go-test output had a lot of debug log messages which obfuscated test results.
I'm finally removeing all those by directing the test log output to io.Discard.
Signed-off-by: Yusuke Kuoka <ykuoka@gmail.com>
* fix: child process not exit when recive term signal
Signed-off-by: yxxhero <aiopsclub@163.com>
* fix: wait for clean done
Signed-off-by: Yusuke Kuoka <ykuoka@gmail.com>
Signed-off-by: yxxhero <aiopsclub@163.com>
Co-authored-by: Yusuke Kuoka <ykuoka@gmail.com>
guofutan
xiaomudk
yxxhero
Quan TRAN
koralsky
Yusuke Kuoka
yxxhero
Lassi Pölönen
Helge Eichelberg
Indrek Juhkam
Rui Chen