When kbn-es downloads a snapshot, it writes the in progress download to
`/tmp` rather than the destination path in case the download fails part
way through, then renames the file from `/tmp` to the final location.
This is a good practice, but in CI the `/tmp` directory isn't stored on
the same disk as the project, which causes a rename error to occur
because we're attempting to rename across devices.
This updates the download logic to write to `${destPath}.tmp` instead,
and do the same renaming, so that we can avoid cross-device errors.
In an attempt to debug random stalls we are seeing in CI, the proc runner will now send SIGKILL if a process does not exit after receiving the standard exit signal it receives (usually SIGTERM). Additionally, if the child process object never emits "close" or "exit" another timeout will trigger an error
* Don't use transforms
* dismiss toasts so they don't interfere with the snapshot tests
* try again
* Fix styling during expanded panel mode
* See if putting the sleep back fixes the build.
* fix typo
* Revert "Revert "Add back support for markdown in custom banners. (#17280)" (#17287)"
This reverts commit 29cbbde599.
* Remove references to Bootstrap banner notifications.
* Prevent non-string banners from being displayed.
Allow plugins to integrate more deeply with webpack where necessary by supporting `__webpackPluginProvider__` uiExports which will be called and expected to return a wepack plugin that will be included in the compiler configuration that Kibana uses.
We found out these errors cause problems on dashboards with panels based
on different index patterns. If a field exists in one index pattern
and not another each search will throw an error. It also changes the
filtering behaviour of the panels with the missing field. Unlike lucene
which will filter out all results that don't have that field, these
errors will prevent the query for the missing field from being added
to the request to ES and the docs with the missing field will still show
up in the results. We could consider changing this behaviour, there are
valid reasons why someone might want to see those results, but to keep
things familiar I think we should preserve lucene's behaviour for now.
* moving plugin template to React and EUI
* fixing method reference error
* adding whitespace and comments per PR feedback
* adjusting test
* fixing test
* reducing Angular dependency and fixing issue with hack not loading properly
* fixing tests
* reacting to PR feedback
* Created general driver to accept an argument select the driver to run.
* Added environment variable for driver, changed package.json to use custim leadfoot, and made changes to browserdriver to use any driver.
* Made changes per PR.
* Fixed all nits.
