As of mitchellh/panicwrap#25, a call to panicwrap.Wrapped() unsets the
cookie in the env, that makes packer plugin child process inherit an env
without the panicwrap cookie and panicwrap itself.
This trips up CleanupClients() in Packer's plugin client: instead of the
real plugin server it now kills its panicwrap parent -- which doesn't
forward SIGKILL to its child because it's not a signal that can be
caught -- and ends up indefinitely waiting in client.Kill() for an EOF
that will never come.
This workaround is to not even try to panicwrap in a plugin server.
Azure expects the tags format to use a pointer to the string for the map value.
The configuration from the builder is not a pointer so when storing in the state bag for
reference in later execution we convert it when creating the StateBag.
Fixes#10012 and #10013.
* Use the MapToAzureTags helper and error check in resource group creation
* Added test case for tag values not using a pointer
* test/azure/arm: Add test to verify tags stored in state
* test/azure/arm: Add azure_tags to existing acceptance test for Linux
Test Before Fix
```
--- FAIL: TestBuilderAcc_ManagedDisk_Linux (1.81s)
panic: interface conversion: interface {} is map[string]string, not map[string]*string [recovered]
panic: interface conversion: interface {} is map[string]string, not map[string]*string
FAIL github.com/hashicorp/packer/builder/azure/arm 1.822s
```
Test After Fix
```
2020/09/29 17:23:03 ui: ==> test: Resource group has been deleted.
--- PASS: TestBuilderAcc_ManagedDisk_Linux (517.41s)
PASS
ok github.com/hashicorp/packer/builder/azure/arm 517.426s
```
Co-authored-by: Wilken Rivera <dev@wilkenrivera.com>
This change checks for closed iochans related to the log polling
function and will break out of the loop when both channels are closed.
Which is an indicator that the SSM session has been terminated by some
external process. This fixes an issue where Packer hangs, due to the
infinite loop, when an SSM session is killed outside of Packer.
Related to #9442