accounts: fix TestUpdateKeyfileContents

[stevemilk]

This pull request fixes a flaky test TestUpdatedKeyfileContents in #29830

Problem Statement:
In this test, we start a watcher to monitor file changes and then update memory cache. If the first file change event is sent before the watcher is ready, the watcher may miss this event , resulting in a perpetual blockage and failure to update the cache. This is the root cause behind occasional errors such as "First replacement failed" encountered in TestUpdatedKeyfileContents.

Repro Steps:

1. Add time.Sleep(500 * time.Millisecond) to func (w *watcher) loop() to delay the loop.

func (w *watcher) loop() {
	time.Sleep(500 * time.Millisecond)
	...
}

2. Run TestUpdatedKeyfileContents.

Solution:
Use WaitGroup to ensure that the watcher loop starts before sending file change events.

[fjl]

I like the idea!

[stevemilk]

After further investigation, I discovered a function called waitWatcherStart intended to ensure the watcher begins before file updates. However, this function fails in this test due to the following steps:

1. NewKeyStore fails to initiate the watcher loop since the directory has not been created. Consequently, the watcher is halted, and watcher.runEnded is set to true.
2. MakeDir creates the directory.
3. The watcher loop restarts, and we wait for the watcher to commence. However, watcher.runEnded has been set to true, causing waitWatcherStart to immediately return true. This can occur before the loop actually begins.
4. Consequently, the watcher may miss the file change event and become stuck in a perpetual block.

The initial two commits attempt to rectify the situation after the aforementioned issues arise. However, a simpler solution to fix this test is to create the directory before NewKeyStore. This ensures the watcher successfully starts on the first attempt, and waitWatcherStart functions as intended.

Considering TestUpdatedKeyfileContents aims to verify if the cache updates correctly when the keystore file changes, the straightforward fix is more appropriate. Nonetheless, the initial two commits remain valuable if, in practical usage, the directory is created after NewKeyStore.

@fjl I would like to hear your suggestions.

[holiman]

If this fix fixes the issue, then LGTM. For this specific test, I don't think creating the keystore beforehand makes any difference for the test

[rjl493456442]

I don't think creating the keystore beforehand makes any difference for the test

This will prevent the file watcher failure due to "non-existent folder" and fix the test accordingly.

[namiloh]

I don't think creating the keystore beforehand makes any difference for the test

This will prevent the file watcher failure due to "non-existent folder" and fix the test accordingly.

Right, yes I meant that it will not negatively impact the "things being tested"