Fix for 54941 pillar_refresh regression

[dwoz]

What does this PR do?

Only refresh pillar data when BaseMinion.gen_modules is called for the first time. Subsequent pillar refreshes are done when the pillar_refresh event if fired.

What issues does this PR fix or reference?

#54941

Tests written?

Yes

Commits signed with GPG?

Yes

[UtahDave]

There's one more scenario I think I'd love to see tested. I'm not sure exactly where you would add this, but it would be great if in one of these spots you add this workflow:

1. Add a key/value pair to the pillar sls file
2. Execute a test.ping
3. Test that the pillar data has NOT been updated with the new key/value pair yet.
4. Run a saltutil.refresh_pillar
5. Then verify that the pillar data has been updated.

The reason for this is it would be great to make sure that a full pillar refresh isn't happening during a regular salt command.

Is this scenario already covered by one of these tests you already have?

[dwoz]

There's one more scenario I think I'd love to see tested. I'm not sure exactly where you would add this, but it would be great if in one of these spots you add this workflow:

1. Add a key/value pair to the pillar sls file
2. Execute a test.ping
3. Test that the pillar data has NOT been updated with the new key/value pair yet.
4. Run a saltutil.refresh_pillar
5. Then verify that the pillar data has been updated.

The reason for this is it would be great to make sure that a full pillar refresh isn't happening during a regular salt command.

Is this scenario already covered by one of these tests you already have?

I thought about adding this exact test because it most closely mirrors the way we found the regression. The only reason I didn't is that we're essentially using pillar.get, pillar.item, and pillar.raw calls in place of test.ping.

[UtahDave]

There's one more scenario I think I'd love to see tested. I'm not sure exactly where you would add this, but it would be great if in one of these spots you add this workflow:

1. Add a key/value pair to the pillar sls file
2. Execute a test.ping
3. Test that the pillar data has NOT been updated with the new key/value pair yet.
4. Run a saltutil.refresh_pillar
5. Then verify that the pillar data has been updated.

The reason for this is it would be great to make sure that a full pillar refresh isn't happening during a regular salt command.
Is this scenario already covered by one of these tests you already have?

I thought about adding this exact test because it most closely mirrors the way we found the regression. The only reason I didn't is that we're essentially using pillar.get, pillar.item, and pillar.raw calls in place of test.ping.

OK, valid point. You are correct. Those calls would trigger the bad behavior, if present.